Intelligent data pre-processing techniques for facilitating load shape prediction of utility systems

Abstract

The system receives a set of load signals from an archive containing historical load information collected at different locations across the entire power grid distributing electricity to the utility system. Next, the system applies a first-order differential function to this set of load signals to generate a set of differential signals. Then, the system performs spike detection on this set of differential signals to identify pairs of positive-negative and negative-positive spikes, which identifies gaps in the load signals associated with periods of network outage. Next, the system modifies the set of load signals by filling each identified gap with predicted load values ​​determined by performing a local load shape prediction operation based on the consecutive load values ​​immediately preceding the identified gaps. Finally, the system predicts the utility system's power demand based on this modified set of load signals.

Description

Technical Field

[0001] The disclosed embodiments generally relate to techniques for performing electricity demand forecasting to facilitate ongoing operations of utility systems. More specifically, the disclosed embodiments relate to a smart load data preprocessing technique that facilitates improved electricity load shape forecasting for utility systems. Background Technology

[0002] Electricity utility systems typically provide very limited storage for electricity, meaning supply and demand must be constantly matched. Regional blackouts can occur when demand suddenly exceeds supply. To avoid such outages, utility companies continuously buy and sell electricity through regional and national grids. Accurate load shape forecasts, used to predict demand two hours or more in the future, are crucial because real-time spot market rates for buying and selling electricity on the grid can fluctuate by 10 to 20 times. For this reason, most utility systems have begun using machine learning techniques to analyze information from multi-year historian archives and long-term historical weather patterns to create short-term load shape forecasts based on current and near-term weather conditions. These short-term forecasts enable optimized supply and demand decisions to minimize the cost of spot market purchases and maximize hourly revenue from spot market sales.

[0003] Utility systems also focus on long-term load shape forecasting to anticipate demand over the coming weeks and months. This demand is less affected by hourly or daily weather fluctuations but more significantly by projected population growth (or decline) patterns, seasonal weather patterns, and residential / commercial demand growth patterns within the geographic area served by the utility. Power companies use such long-term demand forecasts to perform critical operations such as demand-side management; storage maintenance and dispatch; integration of renewable energy; coordinating the supply of cheaper electricity through alternative means such as energy swapping; and entering into bilateral power supply agreements with neighboring utilities and cogeneration facilities.

[0004] The accuracy of short-term and long-term forecasts is heavily influenced by the fidelity of the archived signals in the data history recorder archive. However, the fidelity of such archived signals is often adversely affected by network outage events, such as transformer failures or scheduled maintenance. Archived data generated during such outages is inconsistent with the normal operation of the utility system and can lead to highly inaccurate short-term and long-term demand forecasts.

[0005] Therefore, there is a need for a technique to mitigate the adverse effects of anomalous disturbances on the archived data history recorder signals generated during such network outage events. Summary of the Invention

[0006] The disclosed embodiments relate to a system for predicting the electricity demand of a utility system. During operation, the system first receives a set of load signals from a archive containing historical load information collected at different locations across the entire power grid distributing electricity to the utility system. Next, the system preprocesses the set of load signals. During this preprocessing operation, the system applies a first-order differential function to the set of load signals to generate a set of differential signals. The system then performs a spike detection operation on the set of differential signals to identify pairs of positive and negative spikes, which identifies gaps in the set of load signals associated with periods of network outage. Next, the system modifies the set of load signals by filling each identified gap with predicted load values ​​determined based on a local load shape prediction operation performed immediately preceding the identified gaps. After the preprocessing operation is complete, the system predicts the electricity demand of the utility system based on the set of preprocessed load signals.

[0007] In some embodiments, the system uses forecasts of electricity demand to control the electricity supply provided by the utility system.

[0008] In some embodiments, controlling the power supply provided by a utility system includes one or more of the following: controlling the amount of electricity generated by one or more power plants in the utility system; purchasing electricity for the utility system through the grid; selling electricity generated by the utility system through the grid; storing electricity for future use by the utility system; and developing plans for the utility system to build new power plants or add other power generation assets (e.g., wind turbines, gas turbines, solar power plants, or geothermal assets).

[0009] In some embodiments, when forecasting the electricity demand of a utility system, the system trains an inference model using a set of input signals, including a set of load signals and other input signals, which learns the correlations between the input signals. Next, the system uses the inference model to generate a set of inference signals, whereby the inference model generates an inference signal for each of the input signals in the set. Then, the system uses a Fourier-based decomposition and reconstruction technique to decompose each signal in the set of inference signals into deterministic and random components, and uses the deterministic and random components to generate a set of composite signals that are statistically indistinguishable from the inference signals. Finally, the system projects this composite signal into the future to produce a forecast of the utility system's electricity demand.

[0010] In some embodiments, the inference model is trained using the multivariate state estimation technique (MSET).

[0011] In some embodiments, when using Fourier-based decomposition and reconstruction techniques to generate a set of synthetic signals, the system employs telemetry parameter synthesis (TPSS) techniques, which create high-fidelity synthesis equations for generating the combined synthetic signals.

[0012] In some embodiments, when generating a set of composite signals, the system first generates a set of unnormalized signals. The system then performs an environmental weather normalization operation on the set of unnormalized signals to generate the composite signal, wherein the environmental weather normalization operation uses historical, current, and predicted weather measurements, as well as historical electricity consumption data, to adjust the set of unnormalized signals to account for the impact of weather on electricity demand forecasting.

[0013] In some embodiments, other input signals include electricity consumption data from a group of smart meters, each of which collects electricity consumption data from residential and commercial customers of the utility system.

[0014] In some embodiments, when replacing the load values ​​in the identified gaps with predicted load values, the system uses an optimal value interpolation technique that replaces the missing load values ​​in the set of load signals with interpolated load values ​​determined based on the correlation between the load signals. Attached Figure Description

[0015] FIG. 1A An exemplary power grid circuit according to the disclosed embodiment is illustrated, which includes a left-side circuit and a right-side circuit.

[0016] FIG. 1B The illustration shows the same power grid circuit according to the disclosed embodiment, wherein the transformer is moved from the right circuit to the left circuit.

[0017] FIG. 1C The illustration shows the same power grid circuit according to the disclosed embodiment, which has an isolated fault near the transformer.

[0018] FIG. 2A A graph illustrating the load pattern of load transfer according to the disclosed embodiments is presented.

[0019] FIG. 2B A graph illustrating the load pattern under power outage conditions according to the disclosed embodiments is presented.

[0020] FIG. 3 The illustration depicts an electric utility system according to a disclosed embodiment, the electric utility system comprising a group of power plants connected to homes and businesses via a power grid.

[0021] FIG. 4 A flowchart illustrating how load shape prediction is calculated according to the disclosed embodiments is presented.

[0022] FIG. 5 A flowchart is presented illustrating a technique for preprocessing load data according to the disclosed embodiments and then using the preprocessed load data to predict electricity demand.

[0023] FIG. 6 A diagram illustrating a preprocessor according to a disclosed embodiment is presented.

[0024] FIG. 7 A flowchart illustrating a process for predicting electricity demand according to a disclosed embodiment is presented.

[0025] FIG. 8 A flowchart illustrating a process of replacing load values ​​in gaps associated with network outage periods with predicted load values ​​according to a disclosed embodiment is presented. Detailed Implementation

[0026] The following description is presented to enable those skilled in the art to make and use these embodiments, and is provided in the context of a particular application and its requirements. Various modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of these embodiments. Therefore, these embodiments are not limited to the embodiments shown, but are given the widest scope consistent with the principles and features disclosed herein.

[0027] The data structures and code described in this detailed description are typically stored on a computer-readable storage medium, which can be any device or medium capable of storing code and / or data for use by a computer system. Computer-readable storage media include, but are not limited to, volatile memory, non-volatile memory, magnetic and optical storage devices such as disk drives, magnetic tapes, CDs (optical discs), DVDs (digital versatile optical discs or digital video discs), or other media capable of storing computer-readable media now known or hereafter developed.

[0028] The methods and processes described in the detailed description section can be implemented as code and / or data, which can be stored in a computer-readable storage medium as described above. When a computer system reads and executes the code and / or data stored on the computer-readable storage medium, the computer system executes the methods and processes implemented as data structures and code and stored in the computer-readable storage medium. Furthermore, the methods and processes described below can be contained within a hardware module. For example, a hardware module can include, but is not limited to, application-specific integrated circuit (ASIC) chips, field-programmable gate arrays (FPGAs), and other programmable logic devices now known or developed hereafter. When a hardware module is activated, the hardware module executes the methods and processes contained within it.

[0029] SUMMARY

[0030] To understand how power loads can be interrupted, we first examine two common scenarios. Consider... FIG. 1A The two circuits originating from the left and right sides of the diagram share a common tie switch 102. These two circuits typically carry a 200kVA load. Now, assuming the closed midpoint switch 104 in the right-side circuit requires maintenance, the transformer 106 is transferred from the right-side circuit to the left-side circuit by closing tie switch 102 and opening tie switch 104, as shown. FIG. 1B As shown in the diagram. At this time, the left circuit carries 300kVA, while the right circuit carries only 100kVA. Furthermore, everything operates normally, and there is no power outage. However, the two circular switches 102 and 104 are in an abnormal state (not "invalid state," but "abnormal configuration"), which renders the load levels of both circuits ineffective for load prediction purposes. The left circuit does not see a permanent 50% load increase, nor should we project such a load increase into the future. Similarly, the left circuit does not see its half-load permanently disappear; it has been temporarily moved to another circuit and will likely be moved back within a day or two. FIG. 2A The diagram illustrates the corresponding load patterns for load transfer in the left and right circuits. Since we only want to use "steady-state" loading to determine the growth rate and other changes, we want to be able to automatically filter out these anomalies.

[0031] exist FIG. 1C A similar situation occurs during the power outage illustrated. In this case, an isolation fault exists near transformer 106, which is no longer receiving power. The left circuit operates normally, but the right circuit only provides a 100kVA load upon restoration of power. FIG. 2B The diagram illustrates the corresponding load patterns for this power outage condition in the left and right circuits. Note that there may be some noise in the load signal during a power outage, depending on what the data recorder picks up. For example, there might be a large current spike indicating a short circuit that is interrupted by a circuit breaker, causing the current to temporarily drop to zero. However, the spike duration may not be sufficient for a utility's analog system control and data acquisition (SCADA) system to pick it up. Therefore, after interrupting the fault to allow the arc to extinguish, the circuit typically remains offline for a few seconds, meaning the sensors will see at least a few zero-current intervals. Once the fault is located and the first sectionalizing switch opens, the first half of the circuit can be quickly restored, and the rest comes back online after the fault is repaired. Similarly, from a predictive perspective, the circuit does not experience a "real" drop in load; it simply lacks load value temporarily due to the power outage.

[0032] For each of the scenarios described above, we want to identify where these anomalies occur in the data history recorder database so that we can "analytune" the anomaly patterns from the database before initiating the training process for short-term and long-term load shape predictions. During this prediction process, we want to compute the underlying long-term trends, but these typically result in changes of only a few percentage points or less over a year, while short-term anomalies can be orders of magnitude larger and easily mask long-term trends.

[0033] While the examples above provide a simple illustration of two data cleaning scenarios using simple line graphs, in reality, actual data history recorder time series plots are highly dynamic, exhibiting a 30% diurnal (night to day) load variation in most major cities. These diurnal load variations are then superimposed on an additional 80% of long-term seasonal variation (coldest winter to hottest summer) in the country's hottest and coldest regions. Therefore, an automated anomaly detection process that merely sets a threshold and then infers network anomalies when the load exceeds that threshold, used to facilitate the preprocessing of training data for load shape prediction, is not effective. An advanced statistical pattern recognition-based technique is needed to efficiently detect what we call "box irregularities"—upward or downward square-wave deviations superimposed on complex dynamic load shape patterns.

[0034] Our novel technique for intelligent autonomous preprocessing filters long-term data history recorder signals and identifies box singularities in other steady-state time series signals. Our technique first computes a first-order difference function for each signal in the data history recorder, where the first-order difference function is a numerical approximation of the first derivative of the time series signal. Note that the first-order difference function highlights all plateau regions, regardless of their magnitude. Then, by setting simple criteria for the spikes in the first-order difference function, we can identify and characterize “square wave” deviations. For example, an upward-sloping rectangle in the load signal will include a positive spike followed by a negative spike in the first difference function, and a downward-sloping rectangle in the load signal will include a negative spike followed by a positive spike in the first difference function.

[0035] Note that we use a “spike detection” technique to detect such positive and negative spikes. (See, for example, the spike detection technique described in U.S. Patent Application No. 16 / 215,345, filed December 10, 2018, entitled “Synthesizing High-Fidelity Signals with Spikes for Prognostic-Surveillance Applications”, by inventors Guang C. Wang and Kenny C. Gross, which is incorporated herein by reference.) This spike detection technique is applied to a first difference function to detect positive-negative and negative-positive spike pairs, which allows for the filtering of box singularities from the data history recorder signal to facilitate short-term and long-term load shape prediction.

[0036] After identifying all box singularities in the data history recorder signal, instead of simply cutting off and discarding those singularities, we fill the gaps created by the box singularities with load values ​​inferred from previous load values. For each box singularity, we first (1) extract the longest possible segment of “normal” data that occurred before the box singularity. (2) We then use the extracted segment to train a “mini-prediction model” for load shape prediction techniques. (See, for example, the load shape prediction model described in U.S. Patent Application No. 15 / 715,692, filed September 26, 2017, entitled “Electric Loadshape Forecasting Based on Smartmeter Signals” by inventors Kenny C. Gross, Mengying Li, and Guan C. Wang, which is incorporated herein by reference). (3) Finally, we use this mini-prediction model to predict the load values ​​over the time span of the box singularities.

[0037] For example, if a network has not experienced a singularity in 18 months and then experiences an outage that requires two weeks to repair, then there will be a two-week box singularity in the data stream in the data history archive. In this case, we use load values ​​from the consecutive 18 months prior to the network outage to predict the load values ​​during the two-week outage period. This is superior to simply discarding the data in the box singularity to generate training data for load shape prediction. After all the gaps in the load data associated with the box singularity are filled with the inferred load values, we train the load shape prediction model to project the electrical load into the future.

[0038] Before further describing our preprocessing technique, we first describe the utility system in which it operates.

[0039] Exemplary utility system

[0040] FIG. 3 An exemplary utility system according to the disclosed embodiment is illustrated, comprising a set of power plants 302-304 connected to homes and businesses 310 via a power grid 306. Note that power plants 302-304 can typically include any type of power generation facility, such as a nuclear power plant, a solar power plant, a wind turbine or wind farm, or a coal-fired, natural gas, or oil-fired power plant. Power plants 302-304 are connected to the power grid 306, which transmits electricity to homes and businesses 310 within the area served by utility system 300, and also transmits electricity to and from other utility systems. Note that power grid 306 transmits electricity to homes and businesses 310 via individual smart meters 308, which periodically transmit AMI signals containing electricity consumption data (including kilowatt measurements and kilowatt-hour measurements) to a data center 320.

[0041] The control system within data center 320 receives AMI signals from smart meters 308 and weather data 312, including historical, current, and forecast weather information, and generates load forecasts for sending control signals 325 to power plants 302-304 and the power grid 306. During system operation, load data 327 from the power grid 306 is received by data center 320 and stored in a data history recorder archive 330. This load data is then used to optimize load forecasting, as described in more detail below.

[0042] Generating load shape predictions

[0043] FIG. 4 A flowchart illustrating how the system described above, according to the disclosed embodiment, calculates the optimal load shape prediction 418 is presented. The system begins with AMI meter signals 402 obtained from numerous smart meters in a utility system. FIG. 4 As shown, these AMI meter signals 402 include historical AMI signals 403 and recent AMI signals 404. The system feeds the recent AMI signals 404 to the inference MSET module 405, which trains an inference model to learn the correlations between the recent AMI signals 404, and then uses the trained inference model to generate a set of inference signals 406. Next, the system feeds the inference signals 406 to the TPSS synthesis module 408, which performs a TPSS training operation 410, which decomposes each signal in the set of inference signals 406 into deterministic and random components, and then uses the deterministic and random components to generate a set of corresponding synthetic signals that are statistically indistinguishable from the inference signals. Finally, the system projects this combined signal into the future to generate an unnormalized TPSS prediction 412 for the electricity demand of a set of utility customers.

[0044] Next, the system feeds the unnormalized TPSS forecast 412 to the ambient weather normalization module 416, which normalizes the unnormalized TPSS forecast 412 to account for changes in electricity consumption caused by predicted ambient weather changes. This normalization process involves analyzing historical AMI signals 403 relative to historical weather measurements 414 to determine how AMI meter signals 402 respond to different weather pattern changes. The normalization process then uses current and predicted weather measurements 415 to modify the unnormalized TPSS forecast 412 to account for predicted weather conditions. This produces a final load shape forecast 418, which the utility system can use to perform the various operations described above to control the power supply provided by the utility system.

[0045] Predicting power demand

[0046] FIG. 5 A flowchart illustrating a technique for preprocessing load data and then using the preprocessed load data to predict electricity demand, according to a disclosed embodiment, is presented. The system first receives a set of load signals from a archive containing historical load information collected at various locations throughout the power grid distributing electricity to the utility system (step 502). Next, the system applies a first differential function to the set of load signals to generate a set of differential signals (step 504). Then, the system performs a spike detection operation on the set of differential signals to identify pairs of positive and negative spikes, which identifies gaps in the set of load signals associated with periods of network outage (step 506). Next, the system modifies the set of load signals by filling each identified gap with predicted load values ​​determined based on a local load shape prediction operation performed on consecutive load values ​​immediately preceding the identified gaps (step 508). The system then predicts the electricity demand of the utility system based on the modified set of load signals (step 510). Finally, the system uses the electricity demand prediction to control the electricity supply provided by the utility system (step 512). Note that steps 504, 506, and 508 are derived from... FIG. 6 The preprocessor 604 shown executes to preprocess the data history recorder load signal 602 to generate a preprocessed data history recorder load signal 606.

[0047] FIG. 7 A flowchart illustrating a process for predicting electricity demand according to a disclosed embodiment is presented. (This flowchart illustrates in more detail...) FIG. 5(The operations performed in step 510 of the flow chart). First, the system uses a set of input signals including a set of load signals and other input signals to train an inference model, and the model learns the correlations between the set of input signals (step 702). Next, the system uses the inference model to generate a set of inference signals, where the inference model generates an inference signal for each input signal in the set of input signals (step 704). Then, the system uses Fourier-based decomposition and reconstruction techniques to generate a set of unnormalized synthetic signals that are statistically indistinguishable from the inference signals. This technique decomposes each signal in the set of inference signals into deterministic and random components and uses the deterministic and random components (step 706). Next, the system performs an ambient weather normalization operation on the set of unnormalized synthetic signals to generate the set of synthetic signals, where the ambient weather normalization operation uses historical, current, and predicted weather measurements and historical electricity consumption data to adjust the set of unnormalized signals to account for the impact of weather on the prediction of electricity demand (step 708). Finally, the system projects the set of synthetic signals into the future to generate a prediction of the electricity demand for the utility system (step 710).

[0048] FIG. 8 FIG. presents a flow chart illustrating the process of replacing load values associated with network outage periods in gaps with predicted load values according to the disclosed embodiments. (This flow chart more particularly illustrates the operations performed in step 508 of the flow chart in FIG. 5 (The operations performed in step 508 of the flow chart). First, the system initializes the loop counter i = 1 (step 802). Next, the system determines that the loop counter i < N (step 804). If not (no in step 804), then the process is complete. If so (yes in step 804), then the system collects the continuous load values between gap i-1 and gap i (step 806). Next, the system constructs a micro load shape prediction model based on the collected load values to generate the load values for gap i (step 808). The system then uses the micro load shape prediction model to generate the load values for gap i (step 810). Finally, the system fills gap i with the generated load values (step 812) and increments the loop counter i = i + 1 (step 814) before returning to step 804.

[0049] 9Various modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the invention. Therefore, the invention is not limited to the embodiments shown, but is accorded the widest scope consistent with the principles and features disclosed herein.

[0050] The descriptions of the embodiments above are for illustrative and descriptive purposes only. They are not intended to be exhaustive or to limit this specification to the forms disclosed. Therefore, many modifications and variations will be apparent to those skilled in the art. Furthermore, the above disclosure is not intended to limit the description. The scope of this specification is defined by the appended claims.

Claims

1. A method for forecasting electricity demand in a utility system, comprising: Receive a set of load signals from a archive containing historical load information collected at various locations throughout the power grid that distributes electricity to the utility system; The set of load signals is preprocessed in the following manner. The first-order difference function is applied to the set of load signals to generate a set of differential signals. A spike detection operation is performed on the set of differential signals to identify paired positive-negative spikes and negative-positive spikes, which identify gaps in the set of load signals associated with network outage periods. The set of load signals is modified by filling each identified gap with a predicted load value determined by performing a local load shape prediction operation based on the consecutive load values ​​immediately preceding the identified gap; as well as Based on the preprocessed set of load signals, the power demand of the utility system is predicted, wherein the prediction includes: The inference model is trained using a set of input signals, including the set of load signals and other input signals, and the inference model learns the correlation between the set of input signals. The inference model is used to generate a set of inference signals, wherein the inference model generates an inference signal for each of the set of input signals; Using a Fourier-based decomposition and reconstruction technique, each signal in the set of inference signals is decomposed into deterministic and random components, and these deterministic and random components are used to generate a set of synthetic signals that are statistically indistinguishable from the inference signals; and The aforementioned set of synthetic signals is projected into the future to generate forecasts of electricity demand for utility systems.

2. The method of claim 1, wherein the method further comprises using electricity demand forecasts to control the electricity supply provided by the utility system.

3. The method of claim 2, wherein controlling the power supply provided by the utility system comprises one or more of the following: Controlling the amount of electricity generated by one or more power plants in a utility system; Purchase electricity from the power grid for utility systems; Electricity generated by utility systems is sold through the power grid; Storing electricity for future use by utility systems; and Develop plans for building new power plants for the utility system.

4. The method of claim 1, wherein the inference model is trained using the multivariate state estimation technique (MSET).

5. The method of claim 1, wherein using Fourier-based decomposition and reconstruction techniques to generate the set of synthesized signals includes using telemetry parameter synthesis (TPSS) techniques to create high-fidelity synthesis equations for generating the set of synthesized signals.

6. The method of claim 1, wherein generating the set of synthesized signals comprises: Generate a set of unnormalized signals; as well as An environmental weather normalization operation is performed on the set of unnormalized signals to generate the set of composite signals, wherein the environmental weather normalization operation uses historical, current and predicted weather measurements and historical electricity consumption data to adjust the set of unnormalized signals to account for the impact of weather on electricity demand forecasting.

7. The method of claim 1, wherein the other input signal includes electricity consumption data from a group of smart meters, wherein each smart meter in the group collects electricity consumption data from a customer of a utility system.

8. The method of claim 1, wherein replacing the load values ​​in the identified gaps with predicted load values ​​involves using an optimal value interpolation technique that replaces the missing load values ​​in the set of load signals with interpolated load values ​​determined based on the correlation between the load signals.

9. A non-transitory computer-readable storage medium storing instructions, which, when executed by a computer, cause the computer to perform a method for predicting the electricity demand of a utility system, the method comprising: Receive a set of load signals from a archive containing historical load information collected at various locations throughout the power grid that distributes electricity to the utility system; The set of load signals is preprocessed in the following manner. The first-order difference function is applied to the set of load signals to generate a set of differential signals. A spike detection operation is performed on the set of differential signals to identify paired positive-negative spikes and negative-positive spikes, which identify gaps in the set of load signals associated with network outage periods. The set of load signals is modified by filling each identified gap with a predicted load value determined by performing a local load shape prediction operation based on the consecutive load values ​​immediately preceding the identified gap; as well as Based on the aforementioned set of preprocessed load signals, the electricity demand of the utility system is predicted, wherein the prediction includes: The inference model is trained using a set of input signals, including the set of load signals and other input signals, and the inference model learns the correlation between the set of input signals. The inference model is used to generate a set of inference signals, wherein the inference model generates an inference signal for each of the set of input signals; Using a Fourier-based decomposition and reconstruction technique, each signal in the set of inference signals is decomposed into deterministic and random components, and these deterministic and random components are used to generate a set of synthetic signals that are statistically indistinguishable from the inference signals; and The aforementioned set of synthetic signals is projected into the future to generate forecasts of electricity demand for utility systems.

10. The non-transitory computer-readable storage medium of claim 9, wherein the method further comprises using forecasts of electricity demand to control the power supply provided by the utility system.

11. The non-transitory computer-readable storage medium of claim 10, wherein controlling the power supply provided by the utility system comprises one or more of the following: Controlling the amount of electricity generated by one or more power plants in a utility system; Purchase electricity from the power grid for utility systems; Electricity generated by utility systems is sold through the power grid; Storing electricity for future use by utility systems; and Develop plans for building new power plants for the utility system.

12. The non-transitory computer-readable storage medium of claim 9, wherein the inference model is trained using a multivariate state estimation technique (MSET).

13. The non-transitory computer-readable storage medium of claim 9, wherein using Fourier-based decomposition and reconstruction techniques to generate the set of synthesized signals includes using telemetry parameter synthesis (TPSS) techniques to create high-fidelity synthesis equations for generating the set of synthesized signals.

14. The non-transitory computer-readable storage medium of claim 9, wherein generating the set of synthesized signals comprises: Generate a set of unnormalized signals; as well as An environmental weather normalization operation is performed on the set of unnormalized signals to generate the set of composite signals, wherein the environmental weather normalization operation uses historical, current and predicted weather measurements and historical electricity consumption data to adjust the set of unnormalized signals to account for the impact of weather on electricity demand forecasting.

15. The non-transitory computer-readable storage medium of claim 9, wherein the other input signal includes electricity consumption data from a group of smart meters, wherein each smart meter in the group collects electricity consumption data from a customer of a utility system.

16. The non-transitory computer-readable storage medium of claim 9, wherein replacing the load value in the identified gap with the predicted load value involves using an optimal value interpolation technique that replaces the missing load value in the set of load signals with an interpolated load value determined based on the correlation between the load signals.

17. A system for predicting the electricity demand of a utility system, comprising: At least one processor and at least one associated memory; as well as A prediction mechanism executed on the at least one processor, wherein during operation, the prediction mechanism: Receive a set of load signals from a archive containing historical load information collected at various locations throughout the power grid that distributes electricity to the utility system; The first-order difference function is applied to the set of load signals to generate a set of differential signals. A spike detection operation is performed on the set of differential signals to identify pairs of positive-negative spikes and negative-positive spikes, which identify gaps in the set of load signals associated with network outage periods. The set of load signals is modified by filling each identified gap with predicted load values ​​determined by performing a local load shape prediction operation based on the consecutive load values ​​immediately preceding the identified gap. Based on the modified set of load signals, the electricity demand of the utility system is predicted, wherein the prediction includes: The inference model is trained using a set of input signals, including the set of load signals and other input signals, and the inference model learns the correlation between the set of input signals. The inference model is used to generate a set of inference signals, wherein the inference model generates an inference signal for each of the set of input signals; Using a Fourier-based decomposition and reconstruction technique, each signal in the set of inference signals is decomposed into deterministic and random components, and these deterministic and random components are used to generate a set of synthetic signals that are statistically indistinguishable from the inference signals; and The aforementioned set of synthetic signals is projected into the future to generate forecasts of electricity demand for utility systems.

18. The system of claim 17, wherein the system further uses electricity demand forecasting to control the electricity supply provided by the utility system.

Images