Programmable Load Controller

A. Field of the Invention

This invention relates to the field of programmable load controllers for residential electric utility customers.

B. Background Art

One of the most serious problems confronting electric utility companies today is the great variance in total electrical demand on a network between peak and off peak times of the day. The so-called peak demand periods or load shedding intervals are periods of very high demand on the power generating equipment where load shedding may be necessary to maintain proper service to the network. These occur for example during hot summer days occasioned by the widespread simultaneous usage of electric air conditioning devices. Typically the load shedding interval may last many hours and normally occurs during the hottest part of the day. Peaks may also occur during the coldest winter months in areas where the usage of electrical heating equipment is prevalent. The increasing use of electrical space heaters contributes to this problem. In the past, in order to accomodate the very high peak demands, electric utility companies have been forced to spend tremendous amounts of money either in investing in additional power-generating capacity and equipment, or in buying so-called peak power from other utilities which have made such investments.

Typically the utilities have met this fluctuating electricity demand by using three types of generation. Base load plants (usually nuclear or coal fired) provide nearly constant output. These units are the most efficient, and the most economical to operate. Intermediate plants follow the slow variations in the daily electricity demand, and are consequently more expensive to run. Peaking plants are the least efficient and the most costly of all. Peaking plants are usually gas- or oil-fired turbines that can be started quickly to meet high demands for short periods. Times when a utility must resort to using more of the inefficient and uneconomical generation are termed "on-peak" or "peak times".

The power a customer uses is actually what causes the utility problems, since the utility must have enough "capacity" to meet the customer's electricity demand at all times. In order to encourage customers to reduce demand and avoid building new plants, most utilities include a demand or capacity charge in the rate. If the consumer does not reduce its demand this charge helps pay for new generation. The highest electricity demand in a billing period is compared to a demand norm and the credit or penalty is calculated. Once a penalty has been imposed for a billing period it cannot be removed by controlling demand below the norm during the remainder of the billing period. Thus there is no additional economic advantage to controlling below the demand level which invoke the penalty. This type of demand charge has been common in industrial rates for many years, but has recently been introduced for residential purposes. It is based upon energy use over a predetermined period of time (conceptualized as a sliding window), rather than upon instantaneous use. There is thus no additional economic advantage to controlling spikes in demand, no matter how high they are, if their duration is short enough that they do not raise the demand over the fifteen minute sliding window enough to cause an additional penalty.

In the prior art several basic strategies and devices have been utilized for load shedding in order to limit the peak power demands on the power generating capacity of electric utility companies. One such mode involves sending signals from the utility to disconnect selected electric loads. While this direct control of power consumption by the utility achieves usage cutbacks, the lack of flexibility may inconvenience the user.

An alternative is installing load shedding devices at the point of consumption. This provides greater flexibility and control. Several such devices have been available but none has the ability to adjust the control point upward as the billing period progresses. This presents difficulties for users that have priority loads which must be energized on an as needed basis. An overrideable, fixed control point allows all priority loads to be energized. However, the minimum level necessary to accomodate these loads may unavoidably result in a penalty charge. Subsequently, there is no economic advantage to shedding non-priority loads to stay below this minimum level. However, a situation may arise in which the control point has been overridden in order to allow all priority loads to be energized but, currently, not all priority loads are in use and non-priority loads are deenergized to stay below the control point which has already been exceeded. Thus, the non-priority loads would be shed unnecessarily.

A control method and system for minimizing the on peak power demand of a residential electric utility customer having a plurality of loads is implemented by determining the average energy demand over a predetermined sliding time window and comparing it with a control point. Non-priority loads are shed in an attempt to decrease the demand below the control point. If the energy demand is still above the control point after all non-priority loads are shed, the control point is dynamically adjusted upward to a new value based upon the energy demand and no priority loads are shed. This is compatible with the utility protocol in which a penalty is calculated by the utility based upon the highest electricity demand in the billing period. Once this penalty is imposed for a billing period, it cannot be removed by controlling below that level during the remainder of the billing period. Loads are shed by transmitting bits of information to the loads when the average energy demand over the predetermined time windows increases beyond the control point.

FIG. 1 shows a block diagram of the programmable load contoller of the present invention.

FIGS. 2, 3 and 4 show the control routine for the programmable load controller of the present invention.

FIG. 5 shows a graph of energy demand versus time.

FIG. 6 shows the routine used by the programmable load controller to shed loads.

FIGS. 7 and 8 show the routine used by the programmable load controller to restore loads.

FIG. 9 shows the external interrupt service routine.

FIG. 10 shows the timer interrupt service routine.

FIG. 11 shows the relationship between the AC line, the CURDIG counter, and the bits of control information.

FIG. 12 shows the meter conditioning circuit.

FIG. 13 shows the phase relationship of the zero crossing detector.

FIGS. 14 and 15 show the routine used by the programmable load controller to transmit bits of information to the loads.

FIGS. 16A and 16B show the relationship between the AC line voltage and bits of information.

FIGS. 17A, 17B, 18A and 18B show detailed descriptions of the bits of information.

FIGS. 19A and 19B show the relationship between the bits of information and the various phases of a multiphase system as implemented by conventional load controllers.

FIG. 20 shows the oscillator, driver, and coupling used by the programmable load controller to transmit bits of information to the loads.

A block diagram of a preferred embodiment of a programmable energy load controller 10 for controlling eight loads 40a-h is shown in FIG. 1. Controller 10 includes a microprocessor 14 having a central processing unit 14a and a 128 by 8 RAM 14b. Data buses 18a and 18b allow microprocessor 14 to communicate with an EPROM 16 comprised of a 2K by 8 memory and two 8-bit IO ports. EPROM 16 contains a program which appears at the end of the specification as an appendix and is written in a structured format understandable to those of ordinary skill in the art. For executing the program, microprocessor 14 may be an Intel 8039. The program allows the user to shed loads 40a-h on a priority basis in order to maintain total energy use for up to eight loads below a given control point. The control point may be progressively adjusted higher if necessary in order to energize all priority loads while achieving optimum control.

Microprocessor 14 is connected through a transmitter defined generally by dotted line 31 to an AC line 33 which controls the energizing of eight loads 40a-h. Transmitter 31 includes oscillator 32 and a driver and coupling 30. Microprocessor 14 gates on and off a 121 KHz carrier generated by oscillator 32. This amplitude modulated carrier is injected onto power line 33 allowing microprocessor 14 to transmit a digital signal to loads 40a-h. A 1 bit is represented by the presence of a carrier and a 0 bit is represented by the absence of a carrier. AC line 33 is coupled to the eight loads 40a-h by means of receivers 38a-h respectively as shown within dotted line 37. Receivers 38a-h, which detect this carrier and decode the digital control word, may be conventional receivers such as BSR X-10 or LEVITON. These receivers gate on and off loads 40a-h depending on the control words sent from processor 14. These control words are transmitted to all eight loads simultaneously whenever the status of one of the loads is to be changed. Additionally, control words are transmitted to all eight loads automatically on a periodic basis. For example, the period may be sixteen minutes.

Central load control receiver 13 receives a central control signal from the utility by FM transmission, telephone, or through AC line 33. This binary signal is conditioned by conditioning circuit 15 and applied to microprocessor 14. The binary signal is read by microprocessor 14 as a normally closed switch input signal and is effective to alter the control point of load controller 10. When the switch input signal is opened by the utility, controller 10 reduces demand by 3.0 kW, or to a 4.0 kW minimum. These values may not be altered in the field. The control point may not be adjusted higher when this switch is open.

Keypad 26 is coupled to microprocessor 14 by bus 24 and is used to select a target control point. The algorithm within EPROM 16 may shed non-priority loads as necessary to attempt to maintain energy use below this target. If energy demand remains above the target when only priority loads are energized, the algorithm adjusts the control point up to the demand level. This new level becomes the control point for the remainder of the billing period or until it must be adjusted upward again, or to assure that all priority loads are energized. By selecting a low starting target, a user can assure selection of a control point which is the minimum compatible with the energizing of all priority loads. This also prevents unnecessary shedding of non-priority loads when demand is higher than the target but lower than the previous peak demand of the billing period.

A user may also use keypad 26 to program the peak periods. This allows the user to program system 10 to be compatible with any changes the utility initiates. Additionally, the user may use keypad 26 to establish priority loads which remain energized at all times. This information is stored in a load priority table byte in RAM 14b.

AC conditioning circuit 22 provides a strobe to the interrupt of microprocessor 14 on each negative going zero crossing of AC line 33. EPROM 16 includes a software phase-lock loop which uses this interrupt to synchronize the control word with a zero crossing of the AC line 33. This allows the beginning of each bit of a control word to coincide with the zero crossing of the AC line. Such synchronization is critical, since generally available receivers use the zero crossing of the AC line to sample the line for the presence of a transmitter carrier signal.

Additionally, keypad 26 may be used to request the display items of information on a time multiplexed vacuum florescent display 19. The items of information which may be selected include a user selected target control point, peak periods and priorities; energy consumption including the daily and month to date consumption; and energy demand including current demand and the highest on-peak demand in the current billing period.

Bus 25 connects keypad 26 to EPROM 16 and allows EPROM 16 to energize as appropriate, certain lines in the matrix of keypad 26 to allow microprocessor 14 to scan keypad 26. Bus 24 allows microprocessor 14 to determine which key has been depressed on keypad 26.

Meter conditioning circuit 12 connects the user's power meter to microprocessor 14. It provides logic level conversion upon the output of a conventional induction kilowatt hour meter having a pulse initiator. Each time the wheel on the meter makes a revolution it generates a pulse. The pulse is conditioned and applied to microprocessor 14 by meter conditioning circuit 12 thus providing microprocessor 14 with information on energy use.

Load status display 36 displays the load status byte. This byte is communicated to load status 36 through bus 20 and is latched by a set of eight latches 28. In this byte there is one bit for each of the eight loads 40a-h. A 0 in a given bit position indicates that the corresponding load is energized and this will result in the corresponding LED being turned on. Conversely, a 1 in a given bit position indicates that the corresponding load is not energized and its corresponding LED will be off.

Status block 34 has LED's which indicate whether the system is currently in a control period, whether the current time is within a peak period, and whether the target control point has been exceeded. This information is communicated from microprocessor 14 to status block 34 by way of bus 33.

FIGS. 2-4 show the flow chart of the control routine or procedure which controller 10 uses to determine whether to adjust the control point, to shed loads, or to restore loads. Each time AC conditioning circuit 22 provides microprocessor 14 with a strobe an interrupt service routine is executed at which time registers in microprocessor 14 are advanced. When the seconds register overflows one minute has elapsed and execution of this control routine or procedure is initiated.

On entering this routine a determination of whether the central control flag is set is made in diamond 201. If controller 10 is in the defer mode this flag is set making it possible to control priority loads and diamond 200 is skipped. In diamond 200 a comparison is made between the priority table and the load status table. The priority table contains a 1 or a 0 in each bit location corresponding to each load. Priority for each load is selectable by the user from keypad 26 with a priority 1 indicating no control (always on) or 0 indicating that the load is a non-priority load and is controllable. This allows the user to override controller 10 on specific loads 40a-h with a minimum of effort. If upon making the comparison between the priority table and the current status in diamond 200 of FIG. 1, a priority load is found to be off, its corresponding bit position in the load status table is reset to 0 in block 202. This assures that all priority loads are energized.

Since peak periods are not applicable on Saturday or Sunday the determination made in diamond 206 results in the control flag being reset in block 226 and the on-peak flag being reset in block 228 on those days. On the remaining days the determination defined generally by dotted line 208 results in the on-peak flag being set if the current time is during one of the user selected peak periods. These periods are delineated by T1, T2, T3 and T4. The first peak period is from T1 to T2 and the second is from T3 to T4. These times are easily programmable by the user through keypad 26 to allow for changes in the rate structure. If only one period is desired, the user may program the second period to be of zero length. If the current time is not before T1 but is before T2 as determined by diamonds 210 and 212 or if it is not before T3 but is before T4 as determined by diamonds 214 and 216, then the set of the on-peak flag is executed in block 229. This flag is used by the algorithm to determine whether to adjust the control point and it is one of the flags displayed to the user by status block 34, FIG. 1.

The determination defined generally by dotted line 218 results in a control flag being set twenty-three minutes prior to the beginning of each period. Diamonds 220 and 222 detect current times before the first and second peak periods respectively. Both diamonds send execution to block 224 where the control flag is set. This process permits controller 10 to begin gathering sufficient information to start executing adequate control immediately upon entering a peak period. Twenty-three minutes is determined from the sum of eight minutes which is the time required to exert the maximum control and fifteen minutes. The fifteen minutes is the first set of data for the fifteen minute sliding window since the first average inside the peak period actually uses the 15 minutes of prior data. Both the control and the on-peak flags are reset at the end of either peak period by blocks 226 and 228.

The current average demand is calculated in block 230 of FIG. 3. The energy readings for the previous fifteen minutes are stored on a fifteen level stack in the data RAM 14b. In this step, the energy reading taken fifteen minutes prior to the current time is subtracted from the current reading (KWH). The resulting energy use over the last fifteen minutes is projected as an estimate of a one hour period by multiplying it by four.

In block 231 the central control signal is received from the utility. This signal, which is transmitted by the utility when DEFER mode is desired, is received from central load control receiver 13 of FIG. 1 and applied through conditioning circuit 15 to microprocessor 14. If the central load control signal is low indicating that the utility is not requesting load controller to enter the DEFER mode, the central control flag is reset in block 235. If the central load control signal is high, load controller 10 enters the DEFER mode, which is indicated by the word "defer" flashing on display 19. This message alternates with whatever the user is currently displaying. When controller 10 enters the DEFER mode it calculates the control point by subtracting a predetermined demand KW.sub.DEFER from the 15 minute average calculated in block 230 and uses this result as the new control point. However, it will not reduce the demand below a predetermined minimum level. Controller 10 is currently programmed to reduce the demand by 3.0 kW, or to a 4.0 kW minimum. These values can be altered in the software, but not in the field.

A graphical representation of this procedure is shown in FIG. 5. During normal control mode 291 demand is shown rising as a function of time past the control point level 290 to a peak level 295. At this point a load is shed causing the demand to drop and stay substantially within a dead band defined by control point offset level 292 and a level 294 which is ninety percent of the control point offset level 292. At time 295 the central control mode is activated and the behavior of load controller 10 is shown as central control mode 293. Demand is reduced by shedding other loads until it is below the new control point.

The foregoing operations may be seen in FIG. 3. If the central control signal is not high, as determined in diamond 233, the central control flag is reset in block 235. This takes controller 10 out of the DEFER mode. If the central load controller signal is high a determination of whether the central control flag has been set is made in diamond 237. If this flag is set, the appropriate calculations for the DEFER mode have already been made. If this flag is not set the control point is calculated as shown in block 239. In decision 241, a determination is made whether the control point has fallen below the predetermined minimum. If it has, the control point is set equal to the predetermined minimum block 243. In block 245 the central control flag is set to indicate that the central load control calculations have been performed.

By cycling the central control signal from low to high repeatedly the utility may "ratchet" the control point down to progressively lower levels. This is accomplished as follows. The first time the central control signal is high, block 239 is executed causing a new control point to be defined at a level KW.sub.DEFER below KW.sub.ave. In subsequent passes through the operations shown in FIG. 3, block 230 will always be executed, causing KW.sub.ave to drop to a value below the value which was current when block 239 was executed.

The utility then causes the central control signal to go low causing the central control flag to be reset in block 235. Before KW.sub.ave can rise the utility causes the central control signal to go high a second time. When diamond 237 is thereby encountered again, the central control flag is low causing block 239 to he executed a second time. This time KW.sub.DEFER is subtracted from the new (lower) KW.sub.ave producing a control point which is incrementally lower than the control point produced the first time block 239 was executed. This procedure may be repeated as desired producing progressively lower control points.

If the current time is during a peak period as determined by the test of the on-peak flag in diamond 232 then in diamond 234 the average demand calculated in block 230 is compared with the value of the highest average on-peak demand (HKW) that has occurred in the current billing period. If the present demand is higher, then the highest demand is updated in block 236, thus recording that the present demand is the highest on-peak demand of the current billing period. This allows the dynamic readjustment of the control point throughout the billing period. It is this highest on-peak demand upon which the utility levies its demand charge.

Calculating the fifteen minute average every minute is called a sliding window averaging process. This technique eliminates the need to synchronize the fifteen minute window of controller 10 with that of the utility meter. Solid state utility meters of recent design also use a sliding window to prevent customers from loading up at the end of the demand interval.

When controller 10 is in the DEFER mode the control point is not altered even if it is exceeded. A determination is made therefore, in diamond 247, whether the central control flag is set. If this flag is set load controller 10 is in the DEFER mode and execution jumps over the operations defined by dotted line 238 in which the control point is adjusted.

The user programs by way of keypad 26 a desired target for controller 10 to maintain. Actual on-peak demand may exceed this target due to an unrealistic target level, a high level of uncontrolled loads or poor control. If the target is exceeded as determined in diamond 240, the highest on-peak demand is substituted for the target as the control point in block 242, a target exceeded flag is set in block 244 and the appropriate LED in status block 34 is turned on. It is in this step that the dynamic resetting of the control point takes place. This step is an important feature of the control algorithm which reflects the reality of the rate structure. The user is billed on the highest on-peak demand achieved during the billing period. Therefore, to control at a level below this point might cause a user unnecessary discomfort.

An optimum control point is determined if the user selects a very low target value. Controller 10 attempts to control a demand to this level, but due to the fact that the priority loads are not subject to control it may be unable to do so. When demand exceeds the control point during a peak period, the control point is progressively elevated until a minimum controllable level is found. The determination of the adjusted control point is accomplished without user intervention and in finding the minimum control point, the controller maximizes the financial benefit to the user. This allows the control point to be dynamically set to the highest on-peak demand during the billing period.

If the highest on-peak demand does not exceed the user selected target as determined by diamond 240 the control point is set equal to the target in block 246 and the target exceeded flag will be reset in block 248. This causes the controller to limit demand to the user's specifications.

The result of the steps of dotted line block 238 is to use the user selected target as the control point until it is impossible to energize all priority loads while staying below the target average during the fifteen minute window. Once the target is exceeded and a minimum billing for the month has been determined, there is no reason to attempt to maintain the old target and the algorithm dynamically sets a new one. This prevents the unnecessary shedding of loads when the demand exceeds the user selected target but is still below the highest demand of the billing period.

At the end of the billing period, which is signalled by a switch closure when the utility meter is read, the control point is reset to the target value by block 246. The highest on-peak demand (HKW) is moved to the previous month's on-peak demand (PHKW) and the present monthly high is then reset to 0. This approach is consistent with the utility rate structure.

Diamond 250 tests the control flag which is set twenty-three minutes prior to the beginning of a peak period by block 224 of FIG. 2. If this flag is not set then the current time is not in a period where control is to be executed and the RESTORE routine is called in block 252. A call to this routine turns on one load. By repeated calls of RESTORE, the loads are turned on, one each minute. This action prevents a displaced peak or restrike from forming immediately following the end of a peak period.

If a control flag has heen set, as determined in diamond 250, the current time is during a period where control is to be executed and step 254 of FIG. 4 calculates the fourteen minute average. In this calculation, the energy demand of the previous fourteen minutes is added to the demand of the previous one minute. Thus the demand of the previous minute is counted twice, allowing the shed decision to be based on a parameter which is more sensitive to recent changes in demand than the strict fifteen minute average of block 230. This new value is then multiplied by four to create an estimated projection of the energy demand for the next hour based on the information gathered during the previous fourteen minutes. If this predicted demand plus an offset is above the control point as determined in diamond 258, a call to the SHED routine is made in block 260 to turn off one load. Upon returning from the SHED routine, a flag is set in block 262 in order to record the fact that a load has been shed by this section of the control algorithm. Thus if there are any non-priority loads energized, the steps defined generally by dotted line block 256 result in a load being shed in order to attempt to prevent the demand from exceeding the control point.

At this point whether the SHED routine has been called or not, a second estimated projection of the energy demand over the next hour is made. This projection however is based entirely on the previous one minute, as distinguished from the projection made in block 254 which was based on the previous fourteen minutes with the weight of the previous minute doubled.

This second projection is calculated in block 264. The meter reading of one minute ago is subtracted from the current reading to determine the demand of the previous one minute as was done in block 254. However, in this case, the one minute reading only is used to project the demand of the next hour by multiplying it by sixty. The value thus calculated is compared with the same limit as the value calculated in block 254. In diamond 266, a determination is made whether this projection exceeds the control point minus an offset.

This value (CP minus OFFSET), with which the projections are compared in diamonds 258 and 266, is the shed limit 292 as shown in FIG. 5. It is a fixed offset below the control point 290. The offset is 0.5 kilowatts, which prevents rounding errors and slight control overshoots from pumping up the control point. If the second projection calculated in block 264 is above this shed limit, the SHED routine is called to turn off one load in block 268. This permits controller 10 to respond more quickly to increases in demand.

If the value calculated is not above the shed limit then a determination is made whether it is below the restore limit 294. This is done in diamond 274. The restore limit 294 is 0.9 times the control point minus the same 0.5 kilowatt offset. A deadband of 0.1 times the control point is thus generated between the shed and restore limits. Within this deadband no control action is taken. If the projection of block 264 is below the restore limit, a call to the RESTORE subroutine is made in block 278 to turn on one load. Diamond 276 makes this call of the RESTORE routine conditional however upon the flag of block 262 not being set. This prevents the control loop defined generally by dotted line 272 from negating the effects of the control loop defined generally by dotted line 256.

Another important aspect of this control scheme is that the SHED routine may be called two times during each pass therefore permitting two loads to be shed each minute. This allows the maximum control effort to be realized in four minutes rather than eight minutes. This is useful because the algorithm has no information on which of the controllable loads 40a-h is actually in use. Therefore, it may gate off power to a load which the user has not turned on by means of the local switch on the load itself. Since this would not decrease demand, it is preferable to shed at a higher rate because the instantaneous demand may exceed the control point during the period of time that the controller is searching for a load that is on. Then, when the demand is finally brought back under control, if the uncontrolled demand area is large enough, the average demand may also exceed the control point. Unnecessarily shed loads may be restored later.

As shown in FIG. 6, SHED is the routine which turns one of loads 40a-h off each time the routine is called. The control routine diamonds 258 and 266 of FIG. 4 determine whether the SHED routine is called in diamonds 260 and 268 respectively.

Upon entering the SHED routine, the priority load byte eliminates all the priority 1 loads from possible control action. Priority 1 loads are those loads which the user has defined as not subject to control and which should therefore not be shed unless controller 10 is in the DEFER mode. These loads are masked to appear off in step 302. Since the SHED routine is looking for loads which are on, no control action will be taken with respect to them. This masked byte then contains the status of all controllable loads that are on. A queue pointer, which is a RAM byte containing a 1 in a single bit position, points to the bit position of the load which has been off the longest period of time. This queue pointer is fetched in block 304. It is from this position that the SHED routine searches for the next load to turn off.

In the steps defined generally by dotted line block 306 the pointer is rotated in a circular fashion (block 312) from its present position to the next numerically higher position (i.e. from 2 to 3, 5 to 6, 8 to 1 etc.) until a non-priority load that is on is found or until the pointer has been rotated 8 times.

This is accomplished by determining in diamond 310 whether the load pointed to is on. If it is not, block 312 will rotate the pointer and diamond 314 will send execution back to block 308 if the queue pointer has not been rotated 8 times.

If a load that is on has not been found after all eight rotations as determined by diamond 314, all the non-priority loads are off as seen in block 316. If an on load is found by diamond 310, the load status is modified in block 318 and stored back in RAM 14b in block 320. The transmit state counter is then set to 1 in block 322 to initiate the transmission of the new status to the receivers. The queue pointer is not permanently altered and it is restored to its original position before the return to the calling program.

If controller 10 is not in the defer mode, as detemined in decision 301, priority loads may not be shed. Thus if there are no non-priority loads still energized a return from the SHED routine must be executed without setting any loads to off in block 318. If the central control flag is not set execution from decision 301 passes to the return.

If the central control flag is set, as determined by diamond 301, the same procedure followed when shedding non-priority loads is performed in order to shed priority loads since priority loads may be controlled in the DEFER mode. Thus in block 303 the load status byte is fetched. However this time, as can be seen in block 305 the non-priority loads are masked, as opposed to the case in block 302 where the priority loads were masked. Priority loads are shed in a queue fashion, with a first off being the first on. The priority loads are turned on and off in numerical sequence, that is the priority load with the highest unit number is the first priority load to be turned off. This forms a stack arrangement, which provides for up to eight levels of priority within the priority group. This is implemented in block 307.

In block 307 a load stack pointer is set to point at load number 8. In block 311 this pointer is used to mask the load status byte. In diamond 313 a decision is made whether the load corresponding to the bit at which the load stack pointer points is on. If this load is on execution proceeds to block 318 where that load is turned off. However if the load pointed to is either a non-priority load or a priority load which is already off execution proceeds to block 315 at which the stack pointer is rotated. In diamond 317 a determination is made whether the stack pointer has been rotated eight times. If the answer is no it is still possible to find a priority load which is on and execution is returned to block 311. If the stack pointer has been rotated eight times there are no more loads to be shed as seen in block 319 and a return from the SHED routine is executed.

Thus the operations defined generally by the dotted line 309 perform for priority loads the same functions that the operations defined generally by dotted line 306 performed for non-priority loads. The pointer is rotated from one load to the next until a load which is on is found or until all loads have been tested.

RESTORE is the routine which turns loads back on. It is called from the control routine in blocks 252 and 278 of FIGS. 3 and 4 respectively.

The RESTORE routine is shown in the flow chart of FIGS. 7 and 8. The load status byte is read in block 400 and in diamond 402 it is tested. If it is determined in diamond 402 that all loads are on there is nothing to restore and a return is executed. If at least one load is off the transmit state counter is set to 1 in block 406 to initiate the transmission of the revised load status to the receivers after returning to the calling routine.

The protocol for restoring loads is as follows. First priority loads are restored in numerical order starting from load one. This permits a hierarchy to be established among the priority loads. After all priority loads are restored non-priority loads may be restored. Non-priority loads are restored in a rotating queue fashion. The restoring of priority loads begins in block 407 where all non-priority loads are masked out of the load status byte. In block 408 a stack pointer pointing to load number one is loaded. In block 410 the status byte which was modified in block 407 is masked with the stack pointer loaded in block 408. On the basis of this masking a determination is made in diamond 411 whether the load pointed to is off. If the answer is yes that load is set to on in block 414 the new status is saved in block 415 and a return to the calling routine is executed.

If a determination is made in decision 411 that the load pointed to is on the stack pointer is rotated in block 412 and a determination is made in decision 413 whether the stack pointer has been rotated through all eight loads in decision 413. If all eight loads have not been tested execution returns to block 410 and the next load is tested. Thus the operations defined generally by dotted line 409 test each of the priority loads sequentially starting from load number one to determine whether any are off. If a priority load is found to be off it is turned on and a return executed. If no such load is found there are no priority loads off to restore as seen in block 419 and the load status byte is reloaded in block 421.

The operations defined generally by dotted line 416 shown in FIG. 8 perform upon the non-priority loads essentially the same operations as those performed for the priority loads by the operations defined generally by dotted line 409. In block 420 the load status byte is masked with the queue pointer. In decision 422 a determination is made whether the non-priority load pointed to is off. If the load is off it is set to on in block 423 and the new status byte is saved in block 425. If the load is not off the queue pointer is rotated in block 418 and a determination is made in decision 424 whether the queue pointer has been rotated through all eight loads. If it has not been rotated eight times execution returns to block 420 and the process is repeated.

If either the queue pointer is rotated through all eight loads and no loads are found which can be restored or, alternatively, if an off load is found and it is restored, the queue pointer is fetched in block 428. This is the same queue pointer that was used as a mask in block 420 and whether an off load has been found or not it must be updated by the operations defined generally by dotted line 430.

If an off load is not found, the next controllable load must be determined for the SHED routine. This is accomplished in the loop defined generally by dotted line 430. The next controllable load is located by getting the original queue pointer in block 428, rotating it in block 432 and masking it with the priority byte in block 434. In diamond 436 a test of the bit which resulted from this masking will determine whether the load pointed to is controllable. If so then the pointer is pointing at the next controllable load and it is saved in block 440. This loop continues until a controllable load is found in diamond 436 or until the pointer is rotated through the seven remaining loads as determined by diamond 438.

Synchronization with AC line 33 is very critical since the receivers use the zero crossings as a strobe to sample line 33 for the presence of the carrier. This synchronization is accomplished by utilizing a software phase lock loop or SPLL. The SPLL external interrupt flow chart is shown in FIG. 9 and the SPLL timer interrupt flow chart is shown in FIG. 10. FIG. 11 shows a graph 540 of one cycle of AC line 33 with which controller 10 is to be synchronized. FIG. 11 also shows the contents of the SPLL loop counter, CURDIG, in RAM 14b both before and after synchronization on graphs 542 and 544 respectively.

When the system is first powered up, the CURDIG counter and AC line 33 are not synchronized. (Compare graphs 540 and 542.) Internal timer interrupts every 1.28 milliseconds cause the CURDIG counter to count down from 12 to 1. This is accomplished by decrementing CURDIG in block 520 after the internal timer has been reloaded in block 518. A total of 15.36 milliseconds has ellapsed when the counter reaches 0 and during this countdown interval the external interrupts by AC line 33 have been disabled by block 502. Counts 0 and -1 are detected by diamond 526 causing the external interrupt to be enabled in block 530. The relationship of the external interrupt to the CURDIG counter can be seen at window 543, FIG. 11.

These external interrupts are caused by the negative going zero crossing of the AC line as detected by AC conditioning circuit 22, FIG. 1. The time period when the external interrupts are enabled is called the external interrupt window or the EXTINT window. If the negative going zero crossing does not occur during this window, the external interrupts are not re-enabled after being disabled in block 502. This technique yields a free running (i.e. no interrupts) frequency at 55.8 Hertz (17.91 milliseconds), which results in a nominal beat frequency of 4.2 Hertz with the 60 Hertz line.

Since the two frequencies are not identical, the phase relationship between the two waveforms changes with time until the negative going zero crossing occurs during the EXTINT window. When the external interrupt is sensed during this window, the timer is reloaded in block 500 and the CURDIG counter is loaded to 12 again by block 504. By reloading the timer in block 500 the unit has guaranteed that the next negative zero crossing will occur during the EXTINT window (shown at 545 of FIG. 11), and the unit is therefore synchronized with the AC line. The window has been chosen large enough to allow for normal variations in both the AC line frequency and the microprocessor clock frequency.

For single phase control, diamond 532 detects when the CURDIG counter reaches 11 or 4. Comparison of graph 546 with graph 544 shows that the beginning of count 11 marks the end of the bit transmission initiated at the negative going zero crossing and that the beginning of count 4 marks the end of the bit transmission initiated at the positive going zero crossing. Thus, whenever counts 11 or 4 are detected, execution is sent to block 534 where 121 KHz oscillator 32 is reset. Oscillator 32, FIG. 1 provides the carrier signal which is injected onto AC line 33 to transmit control words to loads 40a-h. For multiple phase control, diamond 532 and block 534 are eliminated, resulting in transmission for a full half cycle as shown in graph 548 of FIG. 11.

The transmission for multiple phase control shown in graph 548 of FIG. 11 should be compared with the standard transmission procedure shown in FIGS. 19A and 19B. Bit 624 of FIG. 19B begins at approximately the positive going zero crossing of phase 1 of FIG. 19A (.+-.100 microseconds) and continues for 1 millisecond. Bit 626 begins approximately 2.778 msec. after the beginning of bit 624. This is timed to coincide with the zero crossing of phase 2 of FIG. 19A. The timing of bit 628 is such that it coincides with the zero crossing of phase 2. Likewise the timing of bit 628 is such that it coincides with the zero crossing of phase 3. These bits are timed with respect to phase 1 to coincide with the zero crossings of phases 2 and 3 because the receivers on each phase look for transmissions only at their own zero crossings. Thus if phases 2 and 3 had different phase angles with respect to phase 1, bits 626 and 628 would not be received.

One solution would be to place a transceiver between the transmitter and the receiver for phases 2 and 3. The transceiver receives the transmitted bit, re-synchronizes it with the zero crossing, and transmits it to the receiver. This is inefficient because a transceiver would be required for every load not on phase 1.

The problem is solved by load controller 10 by transmitting the bit for a complete half cycle as shown in FIG. 11, graph 548. Regardless of the phase angles of phases 2 and 3, they must have a zero crossing somewhere within the time in which the bit shown is transmitted. Whenever the receivers on phases 1 and 2 encounter zero crossings and look for transmissions the bit is present. Thus the need for transceivers is removed for all possible phase relationships in a multiple phase control system.

Diamond 536 detects when the CURDIG counter reaches 6. Recall that once the SPLL locks onto the AC line frequency, CURDIG counts down from 12 to 1 between one negative going zero crossing and the next. A count of 6 is halfway through CURDIG's range and is used to estimate the time of the positive going zero crossing. When the 6 is detected, execution is sent to block 538 which calls the TRANSMIT subroutine, initiating the transmission of a bit as seen in graph 546 of FIG. 11.

The initiation of the bit transmitted at the negative going zero crossing is done in block 506. This also can be seen in graph 546 of FIG. 11. After transmission has been initiated, time registers are incremented in block 508. When the seconds register overflows, as determined by diamond 510, the control routine of FIGS. 2-4 is called. When diamond 514 detects the passage of sixteen minutes, it is time for the automatic periodic writing to all eight loads. Therefore execution is sent to block 516.

The detailed circuitry of negative zero crossing detection or conditioning circuit 22, FIG. 1 is shown in FIG. 12. Isolation from the line is accomplished via a transformer 550. The voltage of the 24 volt winding 552 is applied through a single stage low pass RC filter formed by resistor 554a and capacitor 554b and then to the interrupt pin of processor 14. High frequency noise which would cause false triggering is eliminated by the filter. However the pole of this filter cannot be too low in frequency or the resultant phase shift would cause errors in the zero crossing detection. Some compensation for this phase shift has been designed into the system by virtue of the finite threshold of the interrupt pin of microprocessor 14.

Since transformer winding 552 is referenced to ground an ideal zero crossing detector would detect the transition of the signal through the ground level. Interrupts are generated by a negative transition and the threshold of the interrupt pin is a minimum of 0.8 volts. Diode 556 clips the voltage at junction 560 preventing a rise above 5 volts due to the 5 volt supply 562. Diode 558 clamps the point 560 at the 0.8 volt level during time the line voltage makes a negative going zero crossing allowing the processor to keep track of the time and to synchronize with AC line 33.

An interrupt is therefore generated early creating a negative phase shift as shown by line 564 in FIG. 13. Examination of the BSR format specification indicates an extremely tight specification (+100 micro seconds) on the delay between the actual zero crossing and the start of a bit. Starting a bit early is acceptable since each receiver 38a-h uses the zero crossing as a strobe to sample the line for the carrier. By starting the bit early, less accurate zero crossing detection is required for proper operation. This early start technique is exploited to a minor degree on the negative going zero crossing due to the positive threshold of the interrupt pin. More importantly it is utilized on timer interrupt 6 (graphs 544 and 546, FIG. 11) which is used to predict a positive going zero crossing and occurs 0.653 milli seconds or 14 electrical degrees prior to the actual positive going zero crossing. By relying on the internal timer and using the early start technique the need to detect the actual positive going zero crossing is eliminated.

Where an external interrupt FIG. 9 or timer interrupt FIG. 10 occurs, the TRANSMIT subroutine is called to determine if a bit should be sent and if so whether it is a 1 or a 0. This routine generates the control word transmitted to loads 40a-h in FIG. 1. The calls of the TRANSMIT subroutine can be seen in blocks 506 and 538 of the interrupt service routines of FIGS. 9 and 10 respectively.

FIG. 14 and FIG. 15 show the flow chart of the TRANSMIT subroutine. The TRANSMIT subroutine checks the state counter in RAM 14b to see if transmission is required. If transmission is requested (state counter equal to 1) or underway (state counter greater than 1) the TRANSMIT subroutine determines the proper bit. The bit is determined in real time and this is another approach that is made possible by the early start strategy.

A high frequency amplitude modulated signal is injected onto line 33 which is connected to receivers 38a-h. The basic format comprises a 22 bit word transmitted serially with each bit being synchronized with a zero crossing of AC line 33. Table 1 shows the serial data format comprising 11 cycles of AC line 33. A 22 bit word can be broken into three sections, the start code, the security code and the key code. Tables 1-5 are attached as an Appendix.

The start or framing code is four bits long and is used to synchronize the receivers with the word being transmitted.

The second section is the security or house code as shown in Table 2. These bits identify 16 possible security codes as shown, absent the intervening compliment bits. All receivers in one residence are typically given the same security code. By assigning different security codes to each transmitter, several transmitters can operate on the same lines without interference. This is particularly useful when two residences each have a transmitter or more importantly when two apartments in the same building wish to utilize transmitters.

The third section, which is shown in Table 3, is the key code or operation code which is ten bits long. This section is either the unit number and half of the address or function command. Sixteen of the 22 defined codes are the unit numbers. A unit number is a second half of a receiver's address. There are four key codes which specify an operation. The remaining two codes are the ALL which is used to address all the units on a particular security code and the CLEAR code which resets receivers 38a-h.

In order to have a receiver 38a-h perform an operation it must first be addressed by sending a word having the proper security code and the proper unit code. This word is transmitted twice to assure correct reception followed by 6 cycles of no transmission. After the 6 cycle delay a second word, again having the proper security code but containing the appropriate operation command, is transmitted. This word is also transmitted twice. In total, 50 cycles or 0.833 seconds are required to transmit a complete message to a single receiver. FIGS. 17A-17B, 18A-18B show a complete message transmission. In the technique employed by present systems, a 1 bit comprises three separate bursts of carrier as shown in FIG. 16B. These three bursts are spaced at 60 electrical degrees so as to allow 3 phase communication as seen in FIGS. 19A-19B. This is required due to the fact that receivers 38a-h use a zero crossing as a strobe. The first burst begins within 100 micro seconds of the actual zero crossing. Each burst is 1 millisecond in duration and the leading edges of the three bursts are spaced at exactly 60 electrical degrees (2.778 milliseconds).

The 22 bit word is synthesized in three segments. In the TRANSMIT subroutine checks are made of the state counter in diamonds 572, 574 and 606, FIGS. 14, 15 to determine what section of the word is being transmitted. There are fifteen states as shown in Table 4. Bit patterns for each section are stored in table form in the EPROM 16 and these tables are scanned by the TRANSMIT routine to determine the proper bit. This scanning is two dimensional in nature with one pointer incrementing each time a bit is transmitted and the other pointer being set at the start of a particular section of the word. In order to be efficient in EPROM usage both vertical and horizontal scanning are utilized. Table 5 shows the EPROM table for the unit section.

After the present bit has been determined, the next byte is fetched from EPROM 16 and compared to the end of the section flag. If this next byte is the end of the section flag the state counter is advanced and the bit counter is zeroed in preparation for the next section. If the state counter is greater than 14 it is reset to 1 and the unit counter is advanced. This action causes a complete message with the appropriate unit code and operation code to be transmitted for each of the eight loads. Thus all loads will receive a control word containing their status each time the routine is executed. This process is performed periodically regardless of whether there is a status update or not.

After all eight loads 40a-h have been signalled, the state counter in RAM 14b is returned to 0 to again idle the TRANSMIT routine. Transmission to the eight loads 40a-h requires seven seconds and is executed every time a load status is changed and additionally every sixteen minutes. The extra transmission every sixteen minutes is to insure that the loads are in the proper state since CPU 14 cannot observe the state directly. This is also the reason for transmitting to the entire eight loads 38a-h rather than just transmitting the load that was just changed.

For single phase control it is necessary to inject the carrier only during a short time (1 millisecond) following the zero crossing. This is implemented by resetting oscillator 32 on timer interrupt counts eleven and four in blocks 532 and 534 of FIG. 10, the result of which is seen in graph 546 in FIG. 11. For multiple phase control the transmitter would only change states at the external interrupt or timer count six as seen in graph 548. This would result in the carrier being transmitted for the entire half cycle. In this manner any phase can be controlled. This technique is more effective than the techniques used by current systems shown in FIG. 19B.

The circuit for transmitter 31, FIG. 1 is shown in FIG. 20 and comprises an oscillator 32 and a driver and coupling 30. Oscillator 32 is gated on and off by a signal to processor 14 through resistor 772 to the enable pin of timer 764. A high level on this pin starts oscillator 32. The output of oscillator 32 is amplified and shaped by transistor 774 and tuned isolation transformer stage 776. This signal is coupled to AC line 33 by transformer secondary 778.

In programmable load controller 10 the following components have been used for the operation and function as described and shown.