Power load forecasting method, device, equipment and storage medium

Abstract

The present application discloses a method, apparatus, device, and storage medium for power load forecasting, which includes: obtaining a first power load sequence of a power system; performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data; predicting the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predicting a power load forecast value at a target forecast time; determining a power load residual sequence based on the first power load sequence and the first predicted power load sequence; predicting the power load residual based on the power load residual sequence to obtain a power load residual forecast value; and determining a target power load forecast result based on the power load forecast value and the power load residual forecast value. By using the present application, the accuracy of power load forecasting is improved by performing uncertainty analysis on historical power load sequences and predicting power load based on the uncertainty analysis results.

Description

Technical Field

[0001] The present application relates to the technical field of power load forecasting, and in particular to a power load forecasting method, apparatus, device and storage medium. Background Art

[0002] Currently, there are numerous methods for power load forecasting. Traditional forecasting methods, such as regression analysis, time series analysis, and gray models, are effective and stable when dealing with simple, linear load variations. However, these methods often struggle to accurately capture the complex, nonlinear characteristics and uncertainties of power loads. Intelligent forecasting methods, such as artificial neural networks, support vector machines, and fuzzy logic, while advantageous in handling complex relationships, also have limitations. For example, neural networks can easily fall into local optimal solutions, leading to inaccurate forecasts.

[0003] With the widespread access to distributed energy resources, the diversification of user electricity consumption, and the development of smart grids, the patterns of power load fluctuations have become more complex and variable. Traditional forecasting methods and single intelligent forecasting methods can no longer meet the power system's requirements for high-precision and high-reliability load forecasting. Therefore, how to improve the accuracy of power load forecasting has become an urgent issue. Summary of the Invention

[0004] The embodiments of the present application provide a method, apparatus, device and storage medium for power load forecasting, which effectively reduces the prediction error and improves the accuracy of power load forecasting by performing uncertainty analysis on historical power load sequences and predicting the power load based on the uncertainty analysis results, while predicting and error-correcting the power load residuals.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting power load, comprising:

[0006] Acquire a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence;

[0007] performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data;

[0008] Predicting the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predicting the power load at a target prediction time to obtain a power load prediction value;

[0009] determining an electric load residual sequence according to the first electric load sequence and the first predicted electric load sequence;

[0010] Predicting the power load residual according to the power load residual sequence to obtain a power load residual prediction value;

[0011] A target power load prediction result is determined according to the power load prediction value and the power load residual prediction value.

[0012] In a second aspect, an embodiment of the present application provides a power load forecasting device, comprising: a data acquisition unit, an uncertainty analysis unit, a first load forecasting unit, a data processing unit, and a second load forecasting unit, wherein:

[0013] The data acquisition unit is used to obtain a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence;

[0014] The uncertainty analysis unit is used to perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data;

[0015] The first load prediction unit is used to predict the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at a target prediction time to obtain a power load prediction value;

[0016] The data processing unit is used to determine a power load residual sequence according to the first power load sequence and the first predicted power load sequence;

[0017] The second load prediction unit is used to predict the power load residual according to the power load residual sequence to obtain a power load residual prediction value;

[0018] The data processing unit is further configured to determine a target power load prediction result based on the power load prediction value and the power load residual prediction value.

[0019] In a third aspect, an embodiment of the present application provides an electronic device comprising: a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the steps in the first aspect of the embodiment of the present application.

[0020] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the above-mentioned computer-readable storage medium stores a computer program for electronic data exchange, wherein the above-mentioned computer program enables a computer to execute some or all of the steps described in the first aspect of the embodiment of the present application.

[0021] In a fifth aspect, embodiments of the present application provide a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to perform some or all of the steps described in the first aspect of the embodiments of the present application. The computer program product may be a software installation package.

[0022] It can be seen that the embodiments of the present application have the following beneficial effects:

[0023] By implementing the embodiment of the present application, the historical power load sequence of the power system in a preset time period is obtained, and the historical power load sequence is preprocessed to obtain a first power load sequence; uncertainty analysis is performed on the first power load sequence to obtain uncertainty characteristic data; the power load is predicted based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and the power load at the target prediction moment is predicted to obtain a power load prediction value; the power load residual sequence is determined based on the first power load sequence and the first predicted power load sequence; the power load residual is predicted based on the power load residual sequence to obtain a power load residual prediction value; the target power load prediction result is determined based on the power load prediction value and the power load residual prediction value. It can be seen that by performing uncertainty analysis on the historical power load sequence, predicting the power load based on the uncertainty analysis result, and predicting the power load residual and performing error correction on the power load residual, the prediction error is effectively reduced and the accuracy of the power load prediction is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0025] Figure 1 This is a flow chart of a method for predicting power load provided in an embodiment of the present application;

[0026] Figure 2 This is a schematic diagram of a power load forecast result provided by an embodiment of the present application;

[0027] Figure 3 This is a schematic diagram of a flow chart of power load residual prediction provided by an embodiment of the present application;

[0028] Figure 4 This is a flow chart of another power load forecasting method provided in an embodiment of the present application;

[0029] Figure 5 This is a schematic diagram of the structure of a power load forecasting device provided in an embodiment of the present application;

[0030] Figure 6 This is a structural diagram of a sequence decomposition unit of a second load forecasting unit provided in an embodiment of the present application;

[0031] Figure 7 This is a structural diagram of a residual prediction unit of a second load prediction unit provided in an embodiment of the present application;

[0032] Figure 8 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0034] The terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish between different objects, not to describe a particular order. Furthermore, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0035] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0036] The following describes the relevant contents, concepts, meanings, technical issues, technical solutions, beneficial effects, etc. involved in the embodiments of this application.

[0037] First, some of the terms involved in this application are explained:

[0038] User-side resources: User-side resources refer to equipment and devices such as power generation, power consumption, and energy storage that are installed on the power consumption side, belong to the user, and have the ability to interact with the power grid.

[0039] See Figure 1 , Figure 1 This is a flow chart of a method for power load forecasting provided by an embodiment of the present application, which includes but is not limited to the following steps:

[0040] S101 : Acquire a historical power load sequence of a power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence.

[0041] In the embodiments of this application, the power system is an electric energy production and consumption system consisting of power generation, transmission, transformation, distribution, and consumption. The generated electric energy can be supplied to various user-side resources through transmission, transformation, and distribution. This embodiment of the application primarily analyzes and predicts the power consumption data of user-side resources. Therefore, it is possible to obtain the power load data of the power system for specific analysis.

[0042] The preset time period refers to a pre-set historical time period for power load analysis. This time period is set according to specific application requirements and analysis purposes. For example, if you need to study short-term load fluctuation characteristics, the preset time period can be set to the past few hours or one day to analyze the load variation pattern within the day. If you need to analyze long-term power load trends, the preset time period can be set to the past few years to predict power demand in the next year.

[0043] A historical power load sequence refers to an ordered collection of power load data recorded by the power system within a preset time period. It is indexed by time and consists of a series of discrete power load data. The historical power load sequence reflects the changes in power load over time within the preset time period, that is, the historical power load sequence has both volatility and regularity. Volatility refers to the fact that power load can change instantly due to various factors, such as the start and stop of industrial production, changes in residents' electricity consumption habits, and sudden weather changes. Regularity is manifested at different time scales. For example, the daily load curve shows that the load is higher during the peak period of daytime electricity consumption and lower during the off-peak period of nighttime electricity consumption. The weekly load curve shows that weekdays and weekends have different electricity consumption patterns. The annual load curve shows that the load is higher during summer cooling and winter heating.

[0044] In a specific embodiment, the historical power load sequence of the power system during a preset time period can be obtained through a Supervisory Control and Data Acquisition (SCADA) system. This SCADA system is capable of real-time monitoring and control of the power system's operating equipment, enabling data acquisition, equipment control, measurement, parameter adjustment, and signal alarm functions. Furthermore, the metering equipment and communication technologies within the Advanced Metering Infrastructure (AMI) can be used to accurately and comprehensively collect user power usage data, thereby obtaining the historical power load sequence of the power system during the preset time period.

[0045] Next, after obtaining the historical power load sequence, it is necessary to preprocess it to obtain the first power load sequence. Since the obtained historical power load sequence may contain abnormal data, which may be caused by equipment failure, communication errors, etc., it is necessary to preprocess the historical power load sequence to correct the abnormal data and ensure the accuracy of the data used for power load forecasting.

[0046] Optionally, when preprocessing the historical power load series, to eliminate the dimensional effects of different factors and limit the data range for further data processing and model training, the data in the historical power load series can also be normalized. For example, Min-Max normalization can be used. Specifically, Min-Max normalization performs a linear transformation based on the maximum and minimum values ​​in the historical power load series, mapping each power load data in the historical power load series to the interval [0, 1]. By normalizing the historical power load series, the efficiency and stability of model training can be improved, effectively reducing model training issues caused by data dimensionality differences.

[0047] Optionally, the historical power load sequence may include n power load data, where n is an integer greater than 1; the above step of preprocessing the historical power load sequence to obtain the first power load sequence may specifically include the following steps:

[0048] A101. Determine the absolute value of the difference between any two adjacent power load data among the n power load data to obtain n-1 load difference values;

[0049] A102. Determine the average value of the n-1 load differences to obtain an average value of the load differences;

[0050] A103. Determine an abnormal load threshold value based on a preset correction coefficient and the average value of the load difference;

[0051] A104. Receive the power load data greater than the abnormal load threshold value among the n power load data as abnormal data, and obtain j abnormal data; j is an integer less than n;

[0052] A105. Determine the power load data excluding the j abnormal data among the n power load data to obtain nj power load data, and determine the average value of the nj power load data to obtain a load average value;

[0053] A106. Replace each of the j abnormal data with the load average value to obtain j corrected data;

[0054] A107. Determine the first power load sequence based on the j correction data and the nj power load data.

[0055] The preset correction coefficient refers to a parameter pre-set based on actual experience and historical data to adjust the abnormal load threshold. By setting the preset correction coefficient, the judgment criteria for abnormal data can be flexibly adjusted. For example, when the power load data fluctuates greatly, the preset correction coefficient can be selected to a larger value to increase the abnormal load threshold and reduce the cases where normal fluctuating data is mistakenly judged as abnormal. Otherwise, when the power load data fluctuates relatively smoothly, the preset correction coefficient can be selected to a smaller value to lower the abnormal load threshold, so as to more accurately identify abnormal data, thereby improving the quality of power load data and, in turn, the accuracy of power load forecasting.

[0056] The historical power load sequence may include n power load data, where n is an integer greater than 1.

[0057] In a specific embodiment, by determining the absolute value of the difference between any two adjacent power load data among n power load data, n-1 load differences can be obtained. The load difference can represent the change amplitude between adjacent data to determine the fluctuation of the power load data.

[0058] Next, the average value of n-1 load differences is determined to obtain the load difference average value. The abnormal load threshold is obtained by multiplying the preset correction coefficient and the load difference average value. The abnormal load threshold is used to identify abnormal data in the n power load data, and its value can be adjusted by changing the size of the preset correction coefficient.

[0059] The power load data that exceeds the abnormal load threshold among the n power load data are considered abnormal data, resulting in j abnormal data, where j is an integer less than n. The power load data other than the j abnormal data among the n power load data are determined, resulting in nj power load data. The average value of the nj power load data is then determined to obtain an average load value. This load average value can be used to represent the average level of normal power load data.

[0060] By replacing each of the j abnormal data with the load average, we can obtain j corrected data. By recombining these j corrected data with the nj power load data that were not determined to be abnormal, we can obtain the first power load sequence. By identifying and correcting abnormal data in the historical power load sequence, we avoid data bias caused by abnormal data, and thus obtain a first power load sequence with higher data quality, which more accurately reflects the actual power load situation.

[0061] S102: Perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data.

[0062] In an embodiment of the present application, in order to clarify the changing characteristics of the power load data of the user-side resources and identify the uncertain factors that may affect the load changes, uncertainty analysis can be performed on the first power load sequence to obtain uncertainty characteristic data, which can be used for accurate load forecasting.

[0063] The power load of user-side resources is affected by many factors, including but not limited to weather factors, time factors, economic activity factors, and user behavior factors. For example, in summer, the extensive use of refrigeration equipment such as air conditioners will lead to a significant increase in power load. In winter, the operation of heating equipment will also lead to a significant increase in power load.

[0064] By analyzing the relationship between various influencing factors and power load, a corresponding mathematical model, such as a regression model, can be established to quantify the impact of the influencing factors on the power load. Based on the analysis of the relationship between the influencing factors and the power load, the uncertainty characteristic data of the power load is evaluated. The uncertainty characteristic data can be represented by variance, standard deviation, probability distribution, etc. For example, variance and standard deviation can measure the degree of dispersion of power load data. The greater the degree of dispersion, the higher the uncertainty of the load. The probability distribution can comprehensively describe the probability of the power load occurring within different value ranges.

[0065] By performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data, considering the uncertainty characteristic data in power load forecasting can make the forecast results closer to the actual situation and reduce the forecast error. At the same time, the uncertainty characteristic data can be used to formulate more flexible and adaptive strategies to cope with the uncertain changes in power load and ensure the safe and stable operation of the power system.

[0066] Optionally, the above step of performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data may specifically include the following steps:

[0067] A201. Determine target attribute information corresponding to a first power load sequence;

[0068] A202. Determine a target application scenario of the power system corresponding to the target attribute information; the target application scenario includes one of the following: a load reduction response scenario, an equipment charging and discharging scenario, and a temperature-controlled load operation scenario;

[0069] A203. Acquire operation data of the power system corresponding to the target application scenario during the preset time period;

[0070] A204. Determine uncertainty factor data based on the operation data; the uncertainty factor data includes at least one of the following: user behavior parameters, device parameters, and environmental parameters;

[0071] A205. Perform feature extraction on the uncertainty factor data to obtain the uncertainty feature data.

[0072] In a specific embodiment, target attribute information corresponding to the first power load sequence is determined, wherein the target attribute information may include: specific source information of data, time stamp information, load magnitude information, load change trend, etc.

[0073] Among them, the specific source information of the data clarifies the origin of the first power load series data, that is, the specific user-side resources. User-side resources include a variety of equipment and entities, such as various types of production equipment, electric vehicles, energy storage equipment, and residential electrical equipment of industrial and commercial users. The main user-side resources of the power load data are analyzed through the specific source information of the data to analyze the impact of the power consumption characteristics of different types of users on the power load. For example, when the power load data comes from residential electrical equipment, the power load may show a pattern related to the change in electricity price, so as to determine the reason for the change in power load. Time stamp information refers to the specific time corresponding to the recording of power load data, the load magnitude information reflects the size of the power load, and the load change trend shows the change direction of the power load over time.

[0074] According to the target attribute information, the target application scenario of the corresponding power system can be determined, wherein the target application scenario includes one of the following: a load reduction response scenario, an equipment charging and discharging scenario, and a temperature-controlled load operation scenario.

[0075] It should be noted that in a curtailable load response scenario, the first power load sequence can indicate significant load fluctuations during certain time periods, such as when users proactively reduce or increase their electricity consumption in response to incentives (such as rising electricity prices). In a device charging and discharging scenario, the first power load sequence can indicate regular increases and decreases in power load during certain time periods, such as the charging and discharging behavior of electric vehicles and energy storage devices. In a temperature-controlled load operation scenario, the first power load sequence can indicate a correlation between load changes and temperature changes, such as the operation of temperature-controlled devices such as air conditioners and electric heaters.

[0076] Obtain operational data for the power system during a preset time period corresponding to the target application scenario. This data can be obtained from the power system's monitoring system, smart meter records, weather station data, and equipment management systems. In load reduction response scenarios, operational data can include the response time and amount of user participation in load reduction. In equipment charging and discharging scenarios, operational data can include the charging and discharging power, start and end times, and battery capacity of the charging device. In temperature-controlled load operation scenarios, operational data can include indoor and outdoor temperatures, the on / off status of the temperature control device, and the set temperature. This operational data can be used to analyze the uncertainty factors corresponding to the target application scenario and determine the mechanism of power load changes under the target application scenario.

[0077] Uncertainty factor data can be determined based on operational data. This uncertainty factor data includes at least one of the following: user behavior parameters, device parameters, and environmental parameters. User behavior parameters indicate uncertainty in users' willingness and degree of response to load reduction instructions in load reduction response scenarios. This uncertainty may be influenced by factors such as the strength of incentives and their own electricity usage habits. Furthermore, in device charging and discharging scenarios, users vary in how long and how they use their devices, such as when electric vehicle users choose to charge their devices. Device parameters also exhibit uncertainty, such as the cooling and heating efficiency of temperature control devices and performance variations due to device aging. Furthermore, in device charging and discharging scenarios, battery parameters such as charge and discharge efficiency and remaining capacity are not fixed. Furthermore, environmental parameters also exhibit uncertainty. In temperature-controlled load operation scenarios, changes in environmental factors such as outdoor temperature and humidity can affect the operating time and power consumption of indoor temperature control devices. In device charging and discharging scenarios, ambient temperature also affects battery performance.

[0078] Next, feature extraction is performed on the uncertainty factor data to generate uncertainty characteristic data. For example, for user behavior parameters, features such as the probability distribution of user responses, the mean and variance of response times, etc. can be extracted. For device parameters, the range of device performance variations and the amplitude of fluctuations in key parameters can be extracted. For environmental parameters, the trend of changes in parameters such as temperature and humidity, and the frequency of extreme values ​​can be extracted. By extracting features from the uncertainty factor data, uncertainty characteristic data is generated, which intuitively reflects the characteristics and impact of the uncertainty factors, providing data support for subsequent power load forecasting.

[0079] S103 . Predict the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at a target prediction time to obtain a power load prediction value.

[0080] In the embodiments of the present application, a BP neural network model can be used for power load forecasting, or models such as the autoregressive integrated moving average model (ARIMA) and the long short-term memory network (LSTM) can be used for power load forecasting. Different models have different adaptability to data. For example, the BP neural network is good at processing complex nonlinear relationships and can learn the implicit patterns in the power load series, while the ARIMA model is suitable for time series with stationary and seasonal characteristics. When selecting a specific model, the model or model combination can be selected based on the characteristics of the data and the prediction accuracy requirements.

[0081] In a specific embodiment, the first power load sequence can be used as input data, and the uncertainty characteristic data can be input into the BP neural network model as auxiliary information, so that the BP neural network model considers the impact of the uncertainty characteristic data on the power load when predicting the power load. By predicting the power load based on the first power load sequence and the uncertainty characteristic data, a first predicted power load sequence can be obtained, wherein the first power load sequence and the first predicted power load sequence are the same in time, that is, the first power load sequence indicates actual data within a period of time, and the first predicted power load sequence indicates predicted data within the same period of time. Then, the power load at the target prediction time can be predicted based on the BP neural network model to obtain a power load prediction value.

[0082] Optionally, the above step of predicting the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence may specifically include the following steps:

[0083] A301. Obtain a first training data set and a first test data set, and determine an expected output value corresponding to the first test data set to obtain an expected output data set; the first training data set is historical power load data used for model training; and the first test data set is historical power load data used for model testing;

[0084] A302: performing feature fusion on the first training data set and the uncertainty feature data to obtain a second training data set;

[0085] A303: Train a preset neural network model according to the second training data set to obtain a first neural network model;

[0086] A304. Input the first test data set into the first neural network model to obtain a target output data set;

[0087] A305. Determine a training error dataset based on the expected output dataset and the target output dataset;

[0088] A306. Determine model adjustment parameters corresponding to the first neural network model according to the training error data set;

[0089] A307. Adjust the model parameters of the first neural network model according to the model adjustment parameters to obtain a second neural network model;

[0090] A308. Input the first power load sequence into the second neural network model to obtain the first predicted power load sequence.

[0091] Among them, the preset neural network model refers to the neural network architecture pre-selected before performing power load forecasting. In the embodiment of the present application, the preset neural network model is a BP neural network model.

[0092] In a specific embodiment, a first training data set and a first test data set are obtained, and the expected output value corresponding to the first test data set is determined to obtain the expected output data set, wherein the first training data set is the historical power load data used for model training, and the first test data set is the historical power load data used for model testing. The expected output value is the actual power load value corresponding to the test set data, which is used to compare with the output value of the model to determine the accuracy of the model.

[0093] Next, the first training data set and the uncertainty feature data are subjected to feature fusion to obtain a second training data set, and a preset neural network model is trained based on the second training data set to obtain a first neural network model.

[0094] By inputting the first test data set into the first neural network model, the target output data set can be obtained. The training error data set can be determined based on the expected output data set and the target output data set. The expected output data is the true value, and the target output data is the predicted value. By comparing the true value and the predicted value and calculating the error of each data point, the training error data set can be determined.

[0095] The training error indicates the shortcomings of the first neural network model. By analyzing the training error data set and according to the size and distribution of the error, the parameters that need to be adjusted for the first neural network model can be determined, namely the model adjustment parameters. The model adjustment parameters can be the model's weight parameters, bias parameters, learning rate, etc.

[0096] By adjusting the model parameters of the first neural network model according to the model adjustment parameters, a second neural network model can be obtained. Subsequently, the first power load series is input into the second neural network model to obtain a first predicted power load series. By training the neural network model based on uncertainty characteristic data and historical power load data, the model can consider more factors affecting power load, thereby more accurately capturing the changing patterns of power load and improving prediction accuracy.

[0097] See Figure 2 , Figure 2 This is a schematic diagram of the results of a power load forecast provided by an embodiment of the present application. As shown in the figure, it shows how the predicted and actual power load values ​​change over time. The horizontal axis represents time in hours (h), ranging from 2:00 to 24:00, covering the time range of a day. The vertical axis represents load demand in megawatts (MW), ranging from 400MW to 1300MW.

[0098] In the figure, the dotted curve represents the predicted value of the power load demand at different times obtained based on the BP neural network model, and the solid curve represents the real value of the power load demand at different times obtained through actual measurement or recording.

[0099] Analysis of the power load forecast results reveals that both curves exhibit a trend of initially rising and then falling throughout the day. Specifically, at 2 a.m., load demand is relatively low, then begins to rise gradually, reaching a peak in the evening and into the night before gradually falling back. For example, during the peak period, from around 6 p.m. to 10 p.m., load demand reaches its peak for the day. This is likely due to residents returning home from work, the heavy use of various electrical devices, and the active commercial and industrial activities, leading to a concentrated increase in electricity demand. In the declining period, from around 10 p.m. to midnight, after passing the peak, load demand begins to gradually decline. As night progresses, most electricity-using activities decrease, and load demand gradually falls back to a lower level, approaching the initial load demand state in the early morning.

[0100] As can be seen, the predicted value curve and the true value curve do not completely overlap; there is a certain degree of deviation. In some time periods, the predicted value is higher than the true value, while in other time periods, it is lower than the true value. This shows that while the forecasting model can roughly capture the changing trend of power load demand, due to some uncertainties, there is a certain degree of error in the specific values. If the predicted value is consistently higher than the true value, it may lead to overgeneration in the power system, resulting in energy waste and increased costs. If the predicted value is consistently lower than the true value, it may cause power shortages, affecting normal power consumption for users and even leading to unstable grid operation. Therefore, this forecast error caused by uncertainty can be further predicted to improve the accuracy of power load forecasting.

[0101] S104: Determine a power load residual sequence according to the first power load sequence and the first predicted power load sequence.

[0102] In an embodiment of the present application, the first power load sequence is an actually recorded power load data sequence, which reflects the actual situation of the power load in the past preset time period. The first predicted power load sequence is a data sequence obtained by predicting the power load in the preset time period using a prediction model.

[0103] In a specific embodiment, the power load residual sequence can be determined based on the first power load sequence and the first predicted power load sequence, that is, the power load residual sequence can be obtained by subtracting the value of the first power load sequence from the value of the first predicted power load sequence at the corresponding time point.

[0104] The power load residual sequence can intuitively reflect the difference between the predicted value and the actual value. The smaller the value of the power load residual sequence and the more uniform the distribution, the higher the prediction accuracy of the prediction model. Otherwise, if the power load residual sequence fluctuates greatly and the value is dispersed, it means that the prediction model has a large prediction error.

[0105] S105 . Predicting the power load residual according to the power load residual sequence to obtain a power load residual prediction value.

[0106] In the embodiment of the present application, since the prediction results of the prediction model still have a certain error relative to the true value, after obtaining the power load residual sequence, the power load residual sequence can be further predicted to reduce the prediction error. The power load residual sequence can be predicted using time series analysis methods, machine learning algorithms, etc.

[0107] Optionally, the above step of predicting the power load residual according to the power load residual sequence to obtain the power load residual prediction value may specifically include the following steps:

[0108] A501. Performing modal decomposition on the power load residual sequence using a preset modal decomposition model to obtain a plurality of inherent modal components and a residual component;

[0109] A502. Predicting the multiple natural modal components and the residual components using a preset power load prediction model to obtain multiple first power load residuals and one second power load residual;

[0110] A503. Determine the power load residual prediction value based on the multiple first power load residuals and the one second power load residual.

[0111] The preset modal decomposition model refers to the pre-set Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) model. The ICEEMDAN model decomposes the signal into multiple intrinsic modal components and residual components. It improves on ensemble empirical mode decomposition (EEMD) and effectively addresses modal aliasing by adding adaptive white noise.

[0112] The preset power load forecasting model is the Autoregressive Integrated Moving Average (ARIMA) model. The ARIMA model combines the characteristics of the autoregressive (AR) and moving average (MA) models, and also uses differencing (I) to convert non-stationary time series into stationary series for analysis.

[0113] In a specific embodiment, the modal decomposition of the power load residual sequence is performed through a preset modal decomposition model, and multiple intrinsic modal components and a residual component can be obtained. Specifically, by determining the local maximum and minimum values ​​in the residual sequence, fitting the upper and lower envelopes, and calculating the average values ​​of the upper and lower envelopes, the intrinsic modal components that meet the preset component conditions are continuously iterated and screened out until the remaining components can no longer be further decomposed, thereby obtaining the residual component, wherein the preset component condition refers to the condition satisfied by the intrinsic mode function.

[0114] The power load residual sequence includes components of various frequencies and characteristics, making it difficult to predict directly. Modal decomposition can decompose complex sequences into relatively simple components. Each inherent modal component represents the fluctuation characteristics of the power load residual sequence in a certain frequency band, while the residual component represents the long-term trend of the power load residual sequence or other difficult-to-decompose components. Therefore, by decomposing the power load residual sequence, it is easier to select appropriate prediction methods based on the characteristics of different components, thereby improving the accuracy of the prediction.

[0115] Multiple inherent modal components and residual components are predicted through a preset power load forecasting model, and each inherent modal component and residual component is input into the preset power load forecasting model in turn. The model predicts each component based on its own algorithm and learning of historical data, thereby obtaining multiple first power load residual prediction values ​​corresponding to the inherent modal components and second power load residual prediction values ​​corresponding to the residual components. Specifically, the ARIMA model will first perform stationarity processing and white noise test on the data, determine the model parameters (such as p, d, q values) according to the characteristics of the data, and then make predictions. That is, the model parameters of the ARIMA model corresponding to each component are different.

[0116] The power load residual prediction value is determined according to the plurality of first power load residuals and one second power load residual, that is, the power load residual prediction value can be obtained by linearly superimposing the plurality of first power load residuals and one second power load residual.

[0117] By decomposing the complex power load residual sequence into multiple simple components through modal decomposition and then making separate predictions for each component, the changing characteristics of the residual sequence can be captured more accurately. Compared with directly predicting the power load residual sequence, the accuracy of the prediction is improved.

[0118] See Figure 3 , Figure 3 This is a flow chart of a power load residual prediction provided by an embodiment of the present application. As shown in the figure, by inputting a power load residual sequence, the power load residual sequence is decomposed based on the ICEEMDAN (Adaptive Noise Complete Ensemble Empirical Mode Decomposition) model to decompose the power load residual sequence into multiple intrinsic mode components (IMF1, IMF2, ..., IMFn) with different frequency characteristics and a residual component (Rn), where IMF1 represents the first intrinsic mode component, IMFn represents the nth intrinsic mode component, each intrinsic mode component represents the fluctuation characteristics of the residual sequence in a certain frequency band, and the residual component reflects the long-term trend of the sequence or the part that is difficult to decompose.

[0119] For the multiple intrinsic modal components and residual components obtained by decomposition, the ARIMA (autoregressive integrated moving average) model corresponding to the component can be used for prediction. That is, the ARIMA model can determine the appropriate model parameters based on the historical data of each component, and then perform power load forecasting based on the corresponding ARIMA model to obtain the IMF1 prediction results, IMF2 prediction results, ..., IMFn prediction results and Rn prediction results respectively.

[0120] By summarizing and calculating the prediction results of each component, the power load residual prediction value can be finally obtained. The power load residual prediction value reflects the deviation from the power load prediction value and can be used to correct the power load prediction value to improve the accuracy of the power load prediction.

[0121] S106. Determine a target power load prediction result according to the power load prediction value and the power load residual prediction value.

[0122] In the embodiment of the present application, the power load prediction value is the predicted value of the power load corresponding to the target prediction moment obtained by processing the first power load sequence and uncertainty characteristic data and predicting through a neural network model, and the power load residual prediction value is the value obtained by synthesizing the power load residual sequence after modal decomposition and prediction of each component. The power load residual prediction value reflects the deviation of the power load prediction value.

[0123] In a specific embodiment, a target power load forecast result is obtained by adding the power load forecast value and the power load residual forecast value. The power load forecast value may have certain errors due to the influence of various uncertainties, and the power load residual forecast value reflects the possible situations of these errors. Therefore, by combining the power load forecast value and the power load residual forecast value, the power load forecast value can be corrected, making the target power load forecast result closer to the actual power load, thereby improving the accuracy of the forecast.

[0124] See Figure 4 , Figure 4 This is a flow chart of another power load forecasting method provided by an embodiment of the present application. As shown in the figure, the flow chart uses the historical power load sequence as the initial input data, and generates a predicted power load sequence and a power load forecast value through BP neural network processing. The predicted power load sequence is compared with the real power load sequence, and the difference between the two is calculated to obtain a power load residual sequence, wherein the real power load sequence is the actual recorded power load data, and the power load residual sequence reflects the deviation between the BP neural network prediction result and the actual value. By performing predictive processing on the power load residual sequence, the power load residual value corresponding to the power load residual sequence can be obtained.

[0125] The power load forecast value and the power load residual value are summarized and calculated to finally obtain the power load forecast result. The power load forecast result comprehensively considers the preliminary forecast and the residual correction, and can more accurately reflect the actual situation of the future power load.

[0126] Optionally, the following steps may also be included:

[0127] A601. Establish a corresponding target state model according to the target application scenario; the target state model includes one of the following: a user responsiveness model, a device operation state model, and a temperature control load operation state model;

[0128] A602: Determine the expected load result corresponding to the target prediction time based on the first predicted power load sequence and the target state model;

[0129] A603: Determine the deviation between the target power load forecast result and the load expected result to obtain a target deviation;

[0130] A604. Determine a target correction factor based on the target deviation;

[0131] A605. Correct the target power load forecast result according to the correction factor to obtain the corrected target power load forecast result.

[0132] In a specific embodiment, a target state model is established based on the target application scenario. The target state model includes one of the following: a user responsiveness model, a device operating state model, and a temperature-controlled load operating state model. For example, a user responsiveness model can be established based on the curtailable load response scenario. This model describes the user's responsiveness to reducing power load under different electricity prices, incentives, and other conditions. A device operating state model can be constructed based on the device charging and discharging scenario. This model reflects the charging and discharging state and patterns of the charging device under different conditions. A temperature-controlled load operating state model can be established based on the temperature-controlled load operating scenario. This model reflects the operating state of the temperature-controlled device in response to factors such as ambient temperature and set temperature.

[0133] The first predicted power load sequence is used as the basis for analysis in conjunction with the established target state model. The target state model can adjust and calculate the first predicted power load sequence based on its own rules and parameters, thereby obtaining the expected load result corresponding to the target prediction time. For example, in the user responsiveness model, the load reduction that users are likely to reduce at the target prediction time is predicted based on the first predicted power load sequence and the user's responsiveness to electricity prices, thereby obtaining the expected load result.

[0134] The target deviation is obtained based on the deviation between the target power load forecast result and the load expectation result. The target power load forecast result is the forecast value obtained by comprehensively considering multiple factors, while the load expectation result is the expected value further revised based on the target application scenario. The target deviation reflects the difference between the forecast result and the expected result based on a specific scenario.

[0135] The target correction factor is determined based on the target deviation. Specifically, a proportional coefficient related to the target deviation can be set based on experience. For example, when the target deviation is positive, the target correction factor can be set to 1-k×target deviation / reference value, where k is a pre-set empirical coefficient with a value range between 0 and 1. The reference value can represent the average value of the first predicted power load sequence. When the target deviation is negative, the target correction factor can be 1+k×|target deviation| / reference value.

[0136] The target power load forecast result is multiplied by the target correction factor to obtain the corrected target power load forecast result. By establishing a target state model and considering the special factors of the target application scenario, the forecast result is corrected to make the final forecast result more consistent with the actual situation and reduce the forecast error.

[0137] Optionally, when the target application scenario is the load curtailment response scenario, the above step A601 of establishing a corresponding target state model according to the target application scenario may specifically include the following steps:

[0138] B601. Obtaining user electricity consumption behavior data in the load reduction response scenario;

[0139] B602. Analyze user response behaviors at each of multiple time scales based on the user electricity usage behavior data to obtain multiple electricity usage behavior change characteristics; the multiple time scales include: a short-term time scale, a medium-term time scale, and a long-term time scale;

[0140] B603. Determine target key factors corresponding to the load curtailment response scenario based on the multiple electricity consumption behavior change characteristics; the target key factors include at least one of the following: curtailment potential index, load transfer rate, and electrical equipment parameters;

[0141] B604. Construct the target state model according to the target key factors.

[0142] When the target application scenario is a load curtailment response scenario, a corresponding target state model may be established according to the load curtailment response scenario.

[0143] In a specific embodiment, the electricity usage behavior data of users in load reduction response scenarios are collected, wherein the electricity usage behavior data may include the user's electricity usage at different times, the user's electricity usage habits, historical payment records, and the user's electricity usage data when facing load reduction measures.

[0144] Time scales are categorized into short-term, medium-term, and long-term. Short-term refers to a day to a week, medium-term refers to a month to a quarter, and long-term refers to a year or longer. At each time scale, user electricity usage behavior data is analyzed. For example, at the short-term time scale, changes in user electricity usage within a few hours of receiving a load reduction notice are analyzed to determine whether users respond quickly and reduce load. At the medium-term time scale, user responses to different electricity prices over a month are analyzed to determine whether users adjust their electricity usage to avoid periods of high electricity prices. At the long-term time scale, overall trends in user electricity usage habits are analyzed to identify changes in user electricity usage patterns. By analyzing user response behavior at each of these multiple time scales, multiple characteristics of electricity usage behavior changes can be derived, such as the timeliness of load reduction and the magnitude of load adjustment.

[0145] A comprehensive analysis is conducted on the changing characteristics of multiple electricity consumption behaviors to screen out target key factors that have a significant impact on the load reduction response scenario. The target key factors include at least one of the following: a reduction potential index, a load transfer rate, and electrical equipment parameters. The reduction potential index indicates the maximum load that a user can reduce under different circumstances, which can be determined by analyzing the user's load reduction capacity under different time scales. The load transfer rate indicates the proportion of the user's electricity consumption time transferred from high-load periods to low-load periods. The electrical equipment parameters can be determined based on the type and characteristics of the equipment used by the user.

[0146] Based on the key factors of the target, appropriate mathematical methods or model structures, such as regression models and machine learning models, are used to construct a target state model. For example, based on the load transfer rate, a user responsiveness model is constructed, which represents the electricity consumption relationship based on different electricity price signals.

[0147] By analyzing user electricity consumption behavior data, determining target key factors and building corresponding target state models, we can more accurately predict user load changes in load reduction response scenarios and correct the predicted power load results.

[0148] In summary, by implementing the embodiments of the present application, the historical power load sequence of the power system in a preset time period is obtained, and the historical power load sequence is preprocessed to obtain a first power load sequence; uncertainty analysis is performed on the first power load sequence to obtain uncertainty characteristic data; the power load is predicted based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and the power load at the target prediction moment is predicted to obtain a power load prediction value; the power load residual sequence is determined based on the first power load sequence and the first predicted power load sequence; the power load residual is predicted based on the power load residual sequence to obtain a power load residual prediction value; the target power load prediction result is determined based on the power load prediction value and the power load residual prediction value. It can be seen that by performing uncertainty analysis on the historical power load sequence, predicting the power load based on the uncertainty analysis result, and predicting and error-correcting the power load residual, the prediction error is effectively reduced and the accuracy of the power load prediction is improved.

[0149] See Figure 5 , Figure 5 : is a structural diagram of a power load forecasting device provided in an embodiment of the present application. The power load forecasting device 500 includes: a data acquisition unit 501, an uncertainty analysis unit 502, a first load forecasting unit 503, a data processing unit 504, and a second load forecasting unit 505, wherein:

[0150] The data acquisition unit 501 is used to obtain a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence;

[0151] The uncertainty analysis unit 502 is used to perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data;

[0152] The first load prediction unit 503 is used to predict the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at a target prediction time to obtain a power load prediction value;

[0153] The data processing unit 504 is configured to determine a power load residual sequence based on the first power load sequence and the first predicted power load sequence;

[0154] The second load prediction unit 505 is used to predict the power load residual according to the power load residual sequence to obtain a power load residual prediction value;

[0155] The data processing unit 504 is further configured to determine a target power load prediction result according to the power load prediction value and the power load residual prediction value.

[0156] Optionally, the historical power load sequence includes n power load data, where n is an integer greater than 1; in the aspect of preprocessing the historical power load sequence to obtain the first power load sequence, the data acquisition unit 501 is further specifically configured to:

[0157] Determine the absolute value of the difference between any two adjacent power load data among the n power load data to obtain n-1 load difference values;

[0158] Determine an average value of the n-1 load differences to obtain a load difference average value;

[0159] determining an abnormal load threshold value according to a preset correction coefficient and the average value of the load difference;

[0160] The power load data greater than the abnormal load threshold value among the n power load data are taken as abnormal data to obtain j abnormal data; j is an integer less than n;

[0161] Determine the power load data excluding the j abnormal data among the n power load data to obtain nj power load data, and determine the average value of the nj power load data to obtain a load average value;

[0162] Replacing each of the j abnormal data with the load average value to obtain j corrected data;

[0163] The first electric load sequence is determined according to the j correction data and the nj electric load data.

[0164] Optionally, in performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data, the uncertainty analysis unit 502 is further specifically configured to:

[0165] determining target attribute information corresponding to the first power load sequence;

[0166] Determining a target application scenario of the power system corresponding to the target attribute information; the target application scenario includes one of the following: a load reduction response scenario, an equipment charging and discharging scenario, and a temperature-controlled load operation scenario;

[0167] Acquire operation data of the power system corresponding to the target application scenario in the preset time period;

[0168] Determine uncertainty factor data based on the operation data; the uncertainty factor data includes at least one of the following: user behavior parameters, device parameters, and environmental parameters;

[0169] Feature extraction is performed on the uncertainty factor data to obtain the uncertainty feature data.

[0170] Optionally, in predicting the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, the first load prediction unit 503 is further specifically configured to:

[0171] Obtaining a first training data set and a first test data set, and determining an expected output value corresponding to the first test data set to obtain an expected output data set; the first training data set is historical power load data used for model training; the first test data set is historical power load data used for model testing;

[0172] Performing feature fusion on the first training data set and the uncertainty feature data to obtain a second training data set;

[0173] Training a preset neural network model according to the second training data set to obtain a first neural network model;

[0174] Inputting the first test data set into the first neural network model to obtain a target output data set;

[0175] Determine a training error dataset based on the expected output dataset and the target output dataset;

[0176] Determining a model adjustment parameter corresponding to the first neural network model according to the training error data set;

[0177] Adjusting the model parameters of the first neural network model according to the model adjustment parameters to obtain a second neural network model;

[0178] The first power load sequence is input into the second neural network model to obtain the first predicted power load sequence.

[0179] Optionally, in the aspect of predicting the power load residual according to the power load residual sequence to obtain the power load residual prediction value, the second load prediction unit 505 is further specifically configured to:

[0180] Performing modal decomposition on the power load residual sequence using a preset modal decomposition model to obtain a plurality of inherent modal components and a residual component;

[0181] Predicting the plurality of natural modal components and the residual components by using a preset power load prediction model to obtain a plurality of first power load residuals and a second power load residual;

[0182] The power load residual prediction value is determined based on the plurality of first power load residuals and the one second power load residual.

[0183] See Figure 6 , Figure 6 This is a schematic diagram of the structure of a sequence decomposition unit of a second load forecasting unit provided in an embodiment of the present application. As shown in the figure, the sequence decomposition unit 60 includes: an EMD decomposition module 601, a local mean module 602, an intrinsic modal component calculation module 603, and a residual component calculation module 604. Among them, empirical mode decomposition (EMD) can decompose complex signals into multiple intrinsic modal components and one residual component. Therefore, the power load residual sequence can be decomposed into multiple intrinsic modal components and one residual component through the EMD decomposition module 601. The local mean module 602 receives the signal data processed by the EMD decomposition module 601 and calculates the local mean of the signal. By analyzing and calculating the local extreme points of the signal, the local average trend information of the signal at different scales is obtained. The intrinsic modal component calculation module 603 can use information such as the local mean to calculate the intrinsic modal component, which represents the fluctuation characteristics of the original sequence in a specific frequency band. After obtaining each intrinsic modal component, the residual component calculation module 604 calculates the difference between the original sequence and the extracted intrinsic modal component to obtain the residual component, which represents the low-frequency trend or other difficult-to-decompose part in the sequence that is not fully represented by the intrinsic modal component.

[0184] See Figure 7 , Figure 7 This is a structural diagram of a residual prediction unit of a second load forecasting unit provided in an embodiment of the present application. As shown in the figure, the residual prediction unit 70 includes: an ARIMA model-based prediction module 701 and a linear superposition module 702, wherein ARIMA (autoregressive integrated moving average model) is a time series prediction model that is suitable for analyzing time series data with certain trends and seasonality. The ARIMA model-based prediction module 701 uses the ARIMA model to model and predict multiple intrinsic modal components and residual components obtained by decomposing the power load residual series, determines the model parameters of different components, and then predicts the residual according to the corresponding model. The linear superposition module 702 can linearly superpose the various residual prediction values ​​obtained by the ARIMA model prediction module to obtain the final residual prediction result.

[0185] Optionally, the power load forecasting device 500 is further specifically configured to:

[0186] Establish a corresponding target state model according to the target application scenario; the target state model includes one of the following: user responsiveness model, equipment operation state model, temperature control load operation state model;

[0187] Determining a load expectation result corresponding to the target prediction time based on the first predicted power load sequence and the target state model;

[0188] Determining a deviation between the target power load prediction result and the load expected result to obtain a target deviation;

[0189] determining a target correction factor based on the target deviation;

[0190] The target power load forecast result is corrected according to the correction factor to obtain the corrected target power load forecast result.

[0191] Optionally, when the target application scenario is the load curtailment response scenario, in terms of establishing a corresponding target state model according to the target application scenario, the power load forecasting device 500 is further specifically configured to:

[0192] Acquiring user electricity consumption behavior data in the load curtailment response scenario;

[0193] Analyzing the user's response behavior at each of multiple time scales based on the user's electricity consumption behavior data to obtain multiple electricity consumption behavior change characteristics; the multiple time scales include: a short-term time scale, a medium-term time scale, and a long-term time scale;

[0194] Determining target key factors corresponding to the load curtailment response scenario based on the multiple electricity consumption behavior change characteristics; the target key factors include at least one of the following: curtailment potential index, load transfer rate, and electrical equipment parameters;

[0195] The target state model is constructed according to the target key factors.

[0196] The power load forecasting device 500 described in the present application can obtain the historical power load sequence of the power system in a preset time period, and pre-process the historical power load sequence to obtain a first power load sequence; perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data; predict the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at the target prediction moment to obtain a power load prediction value; determine the power load residual sequence based on the first power load sequence and the first predicted power load sequence; predict the power load residual based on the power load residual sequence to obtain a power load residual prediction value; determine the target power load prediction result based on the power load prediction value and the power load residual prediction value. It can be seen that by performing uncertainty analysis on the historical power load sequence, predicting the power load based on the uncertainty analysis result, and predicting and correcting the power load residual, the prediction error is effectively reduced and the accuracy of the power load prediction is improved.

[0197] See Figure 8 , Figure 8 : is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include a processor, a memory, a communication interface, and one or more programs. The processor, memory, and communication interface may be interconnected via a bus. The one or more programs are stored in the memory and configured to be executed by the processor. In the embodiment of the present application, the program includes instructions for performing the following steps:

[0198] Acquire a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence;

[0199] performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data;

[0200] Predicting the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predicting the power load at a target prediction time to obtain a power load prediction value;

[0201] determining an electric load residual sequence according to the first electric load sequence and the first predicted electric load sequence;

[0202] Predicting the power load residual according to the power load residual sequence to obtain a power load residual prediction value;

[0203] A target power load prediction result is determined according to the power load prediction value and the power load residual prediction value.

[0204] The electronic device described in this application can obtain the historical power load sequence of the power system in a preset time period, and pre-process the historical power load sequence to obtain a first power load sequence; perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data; predict the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at the target prediction moment to obtain a power load prediction value; determine the power load residual sequence based on the first power load sequence and the first predicted power load sequence; predict the power load residual based on the power load residual sequence to obtain a power load residual prediction value; determine the target power load prediction result based on the power load prediction value and the power load residual prediction value. It can be seen that by performing uncertainty analysis on the historical power load sequence, predicting the power load based on the uncertainty analysis result, and predicting the power load residual and performing error correction on the power load residual, the prediction error is effectively reduced and the accuracy of the power load prediction is improved.

[0205] An embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, and the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.

[0206] The present application also provides a computer program product comprising a non-transitory computer-readable storage medium storing a computer program, wherein the computer program is operable to cause a computer to perform some or all of the steps of any of the methods described in the above method embodiments. The computer program product may be a software installation package, and the computer may comprise an electronic device.

[0207] Those skilled in the art will appreciate that all or part of the process steps in the above-described method embodiments can be implemented by a computer program instructing the relevant hardware. The program can be stored in a computer-readable storage medium, and when executed, the program can include the process steps in the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

[0208] The steps of the method or algorithm described in the embodiments of the present application can be implemented in hardware or by a processor executing software instructions. The software instructions can be composed of corresponding software modules, which can be stored in RAM, flash memory, ROM, EPROM, electrically erasable programmable read-only memory (EEPROM), registers, hard disk, removable hard disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be an integral part of the processor. The processor and storage medium can be located in an ASIC. In addition, the ASIC can be located in a terminal device or a management device. Of course, the processor and storage medium can also exist as discrete components in the terminal device or the management device.

[0209] Those skilled in the art will appreciate that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented in whole or in part via software, hardware, firmware, or any combination thereof. When implemented using software, they can be implemented in whole or in part in the form of a computer program product. This computer program product comprises one or more computer instructions. When these computer program instructions are loaded and executed on a computer, they fully or partially produce the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (eg, a floppy disk, a hard disk, a magnetic tape), an optical medium (eg, a digital video disc (DVD)), or a semiconductor medium (eg, a solid state disk (SSD)).

[0210] The modules / units included in the various devices and products described in the above embodiments may be software modules / units, hardware modules / units, or partly software modules / units and partly hardware modules / units. For example, for various devices and products applied to or integrated into a chip, the modules / units included therein may all be implemented in the form of hardware such as circuits, or at least part of the modules / units may be implemented in the form of software programs, which run on a processor integrated inside the chip, and the remaining (if any) modules / units may be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated into a chip module, the modules / units included therein may all be implemented in the form of hardware such as circuits, and different modules / units may be located in the same component (such as a chip, circuit module, etc.) or different components of the chip module, or at least part of the modules / units may be It is implemented in the form of a software program, which runs on the processor integrated inside the chip module, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits; for various devices and products applied to or integrated in the terminal equipment, the various modules / units contained therein can be implemented in the form of hardware such as circuits, and different modules / units can be located in the same component (for example, chip, circuit module, etc.) or different components in the terminal equipment, or at least some modules / units can be implemented in the form of a software program, which runs on the processor integrated inside the terminal equipment, and the remaining (if any) modules / units can be implemented in the form of hardware such as circuits.

[0211] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the embodiments of the present application. It should be understood that the above description is only a specific implementation method of the embodiments of the present application and is not intended to limit the scope of protection of the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made on the basis of the technical solutions of the embodiments of the present application should be included in the scope of protection of the embodiments of the present application.

Claims

1. A method for predicting power load, characterized in that: The method comprises: Acquire a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence; performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data; the first power load sequence is affected by the following factors: weather factors, time factors, economic activity factors, and user behavior factors; Predicting the power load based on the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predicting the power load at a target prediction time to obtain a power load prediction value; determining an electric load residual sequence according to the first electric load sequence and the first predicted electric load sequence; Predicting the power load residual according to the power load residual sequence to obtain a power load residual prediction value; Determining a target power load prediction result according to the power load prediction value and the power load residual prediction value; The performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data includes: determining target attribute information corresponding to the first power load sequence; Determining a target application scenario of the power system corresponding to the target attribute information; the target application scenario includes one of the following: a load reduction response scenario, an equipment charging and discharging scenario, and a temperature-controlled load operation scenario; Acquire operation data of the power system corresponding to the target application scenario in the preset time period; Determine uncertainty factor data based on the operation data; the uncertainty factor data includes: user behavior parameters, device parameters, and environmental parameters; Performing feature extraction on the uncertainty factor data to obtain the uncertainty feature data; The method further comprises: Establish a corresponding target state model according to the target application scenario; the target state model includes one of the following: user responsiveness model, equipment operation state model, temperature control load operation state model; Determining a load expectation result corresponding to the target prediction time based on the first predicted power load sequence and the target state model; Determining a deviation between the target power load prediction result and the load expected result to obtain a target deviation; determining a target correction factor based on the target deviation; Correcting the target power load forecast result according to the target correction factor to obtain a corrected target power load forecast result; Wherein, when the target application scenario is the load curtailable response scenario, establishing a corresponding target state model according to the target application scenario includes: Acquiring user electricity consumption behavior data in the load curtailment response scenario; Analyzing the user's response behavior at each of multiple time scales based on the user's electricity consumption behavior data to obtain multiple electricity consumption behavior change characteristics; the multiple time scales include: a short-term time scale, a medium-term time scale, and a long-term time scale; Determining target key factors corresponding to the load curtailment response scenario based on the multiple electricity consumption behavior change characteristics; the target key factors include at least one of the following: curtailment potential index, load transfer rate, and electrical equipment parameters; The target state model is constructed according to the target key factors.

2. The method according to claim 1, wherein The historical power load sequence includes n power load data, where n is an integer greater than 1; the preprocessing of the historical power load sequence to obtain a first power load sequence includes: Determine the absolute value of the difference between any two adjacent power load data among the n power load data to obtain n-1 load difference values; Determine an average value of the n-1 load differences to obtain a load difference average value; determining an abnormal load threshold value according to a preset correction coefficient and the average value of the load difference; The power load data greater than the abnormal load threshold value among the n power load data are taken as abnormal data to obtain j abnormal data; j is an integer less than n; Determine the power load data excluding the j abnormal data among the n power load data to obtain nj power load data, and determine the average value of the nj power load data to obtain a load average value; Replacing each of the j abnormal data with the load average value to obtain j corrected data; The first electric load sequence is determined according to the j correction data and the nj electric load data.

3. The method according to any one of claims 1 to 2, characterized in that The step of predicting the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence includes: Obtaining a first training data set and a first test data set, and determining an expected output value corresponding to the first test data set to obtain an expected output data set; the first training data set is historical power load data used for model training; the first test data set is historical power load data used for model testing; Performing feature fusion on the first training data set and the uncertainty feature data to obtain a second training data set; Training a preset neural network model according to the second training data set to obtain a first neural network model; Inputting the first test data set into the first neural network model to obtain a target output data set; Determine a training error dataset based on the expected output dataset and the target output dataset; Determining a model adjustment parameter corresponding to the first neural network model according to the training error data set; Adjusting the model parameters of the first neural network model according to the model adjustment parameters to obtain a second neural network model; The first power load sequence is input into the second neural network model to obtain the first predicted power load sequence.

4. The method according to claim 1, wherein The step of predicting the power load residual according to the power load residual sequence to obtain a power load residual prediction value includes: Performing modal decomposition on the power load residual sequence using a preset modal decomposition model to obtain a plurality of inherent modal components and a residual component; Predicting the plurality of natural modal components and the residual components by using a preset power load prediction model to obtain a plurality of first power load residuals and a second power load residual; The power load residual prediction value is determined based on the plurality of first power load residuals and the one second power load residual.

5. A power load forecasting device, characterized in that: The power load forecasting device includes: a data acquisition unit, an uncertainty analysis unit, a first load forecasting unit, a data processing unit, and a second load forecasting unit, wherein: The data acquisition unit is used to obtain a historical power load sequence of the power system in a preset time period, and preprocess the historical power load sequence to obtain a first power load sequence; The uncertainty analysis unit is used to perform uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data; the first power load sequence is affected by the following factors: weather factors, time factors, economic activity factors, and user behavior factors; The first load prediction unit is used to predict the power load according to the first power load sequence and the uncertainty characteristic data to obtain a first predicted power load sequence, and predict the power load at a target prediction time to obtain a power load prediction value; The data processing unit is used to determine a power load residual sequence according to the first power load sequence and the first predicted power load sequence; The second load prediction unit is used to predict the power load residual according to the power load residual sequence to obtain a power load residual prediction value; The data processing unit is further configured to determine a target power load prediction result based on the power load prediction value and the power load residual prediction value; The performing uncertainty analysis on the first power load sequence to obtain uncertainty characteristic data includes: determining target attribute information corresponding to the first power load sequence; Determining a target application scenario of the power system corresponding to the target attribute information; the target application scenario includes one of the following: a load reduction response scenario, an equipment charging and discharging scenario, and a temperature-controlled load operation scenario; Acquire operation data of the power system corresponding to the target application scenario in the preset time period; Determine uncertainty factor data based on the operation data; the uncertainty factor data includes: user behavior parameters, device parameters, and environmental parameters; Performing feature extraction on the uncertainty factor data to obtain the uncertainty feature data; The power load forecasting device is further specifically used for: Establish a corresponding target state model according to the target application scenario; the target state model includes one of the following: user responsiveness model, equipment operation state model, temperature control load operation state model; Determining a load expectation result corresponding to the target prediction time based on the first predicted power load sequence and the target state model; Determining a deviation between the target power load prediction result and the load expected result to obtain a target deviation; determining a target correction factor based on the target deviation; Correcting the target power load forecast result according to the target correction factor to obtain a corrected target power load forecast result; Wherein, when the target application scenario is the load curtailable response scenario, establishing a corresponding target state model according to the target application scenario includes: Acquiring user electricity consumption behavior data in the load curtailment response scenario; Analyzing the user's response behavior at each of multiple time scales based on the user's electricity consumption behavior data to obtain multiple electricity consumption behavior change characteristics; the multiple time scales include: a short-term time scale, a medium-term time scale, and a long-term time scale; Determining target key factors corresponding to the load curtailment response scenario based on the multiple electricity consumption behavior change characteristics; the target key factors include at least one of the following: curtailment potential index, load transfer rate, and electrical equipment parameters; The target state model is constructed according to the target key factors.

6. An electronic device, characterized in that: include: a processor, a memory, a communication interface, and one or more programs; The one or more programs are stored in the memory and configured to be executed by the processor, wherein the programs include instructions for executing the steps of the method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program includes program instructions. When the program instructions are executed by a processor, the processor is caused to perform the method according to any one of claims 1 to 4.

Images