Power grid load prediction method and system, storage medium and electronic device

Abstract

The application provides a power grid load prediction method and system, a storage medium and an electronic device, and relates to the technical field of power grid load prediction. On the basis of the currently common time series prediction, the regularity characteristics of the change of the power grid load itself are considered, such as the influence of different months, different weeks, and whether it is a holiday on the power grid load, different time power grid load influence factors are adopted, and then a classic time series prediction model is combined to further improve the power grid load prediction accuracy.

Description

Technical Field

[0001] This invention relates to the field of power grid load forecasting technology, and specifically to a power grid load forecasting method, system, storage medium, and electronic device. Background Technology

[0002] Forecasting power grid load is extremely important for power grid dispatching, provides important scientific basis for future power grid development planning, and is also a guarantee for better serving the production and life of the whole country in the context of power marketization.

[0003] The commonly used forecasting method currently involves decomposing time series data into trend, seasonal, and residual components, then building separate models for each component, and finally integrating the results into a final forecast. For example, patent CN 111161538B uses time series decomposition to forecast traffic flow; similarly, patent CN 109376896B uses the same method to decompose distribution network load into trend, periodic, and residual components, then builds separate forecasting models for each component, and finally integrates the forecast results into a final forecast.

[0004] However, current common methods for time series decomposition cannot fully extract the changing patterns of power grid load sequences, resulting in low accuracy in power grid load forecasting. Summary of the Invention

[0005] (a) Technical problems to be solved

[0006] To address the shortcomings of existing technologies, this invention provides a power grid load forecasting method, system, storage medium, and electronic device, solving the technical problem of low accuracy in power grid load forecasting.

[0007] (II) Technical Solution

[0008] To achieve the above objectives, the present invention provides the following technical solution:

[0009] A power grid load forecasting method involves pre-constructing a power grid load forecasting model, the application process of which includes:

[0010] S1. Obtain the historical load data sequence and perform a stationarity test. If the sequence is not stationary, proceed to S2; otherwise, proceed to S3.

[0011] S2. Perform difference operations on the non-stationary sequence until it becomes stationary;

[0012] S3. Remove time factors from the stationary sequence to obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday.

[0013] S4. Extract the periodic term from the intermediate sequence to obtain the remaining sequence;

[0014] S5. Check whether the remaining sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the remaining sequence and the time factor of the next time step, and go to S8; otherwise, go to S6.

[0015] S6. Construct an autoregressive model based on the remaining sequence to obtain the residual sequence;

[0016] S7. Check whether the residual sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the autoregressive model, the residual sequence, and the time factor of the next time step, and proceed to S8; otherwise, let the residual sequence be the residual sequence for the next round, and proceed to S6.

[0017] S8. If the differential operation was not performed on the historical load data sequence in S1 to S2, the initial load forecast value shall be used as the final load forecast value; otherwise, the inverse differential operation shall be performed on the initial load forecast value, and the result shall be used as the final load forecast value.

[0018] Preferably, S3 includes:

[0019] S31. Obtain time factors for different months, different days of the week, and whether it is a holiday;

[0020]

[0021]

[0022]

[0023] Where month = 1, 2, ..., 12; week = 1, 2, ..., 7; i = 0 when a day is not a holiday, and i = 1 when a day is a holiday;

[0024] S32. Remove time factors from stationary sequences;

[0025]

[0026] Among them, {x t} is a stationary sequence; {y t} represents the intermediate sequence.

[0027] Preferably, in step S4, the X-11 seasonal adjustment model is used to extract the periodic term and obtain the remaining sequence, as follows:

[0028] z t =y t -s t

[0029] Among them, {y t} represents the intermediate sequence; {s t} represents a periodic term; {z} t} represents the remaining sequence.

[0030] Preferably, in step S5, obtaining the initial load forecast value for the next time step based on the remaining sequence and the time factor of the next time step includes:

[0031] For periodic terms {s t Make a prediction at time t+1 and obtain s. t+1 , then y t+1 =s t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1 t+1 .

[0032] Preferably, S6 includes:

[0033] S61. Construct an autoregressive model based on the remaining sequence;

[0034] AR(p) = α0 + α1z t-1 +α2z t-2 +…+α p z t-p +ε t

[0035] Where AR(p) is an autoregressive model, and its order p is determined by the residual sequence {z}. t The autocorrelation coefficient and partial autocorrelation coefficient of} are determined; α0,…,α p All are coefficients; z t-1 z t-2 …z t-p Let {ε} be the residual sequences at times t-1, t-2, ..., tp, respectively; t} is a zero-mean white noise sequence;

[0036] S62. Obtain the residual sequence based on the autoregressive model;

[0037] res t =z t -(α0+α1z t-1 +α2z t-2 +…+α p z t-p )

[0038] Among them, {res t} represents the residual sequence.

[0039] Preferably, in step S7, obtaining the initial load forecast value for the next time step based on the autoregressive model, the residual sequence, and the time factor for the next time step includes:

[0040] For periodic terms {s t Make a prediction at time t+1 and obtain s. t+1 Based on the AR(p) model, the remaining sequence {z} t The predicted value z at time t+1 is obtained by making a prediction. t+1 , then y t+1 =s t+1 +z t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1 t+1 .

[0041] Preferably, during the training phase, the mean squared error loss function is used to measure the error of the power grid load prediction model;

[0042]

[0043] Where MSE is the mean squared error loss function; a t The actual load values ​​of the training set samples at time t; t represents the load prediction value of the model at time t; n represents the sequence length of the sample.

[0044] A power grid load forecasting system, which pre-constructs a power grid load forecasting model, specifically applied to:

[0045] The acquisition module is used to execute S1, acquire historical load data sequences, and perform stationarity tests. If the sequence is not stationary, the process is transferred to the difference module to execute S2; otherwise, the process is transferred to the removal module to execute S3.

[0046] The difference module is used to execute S2, performing difference operations on non-stationary sequences until they become stationary;

[0047] The removal module is used to execute S3 to remove time factors from the stationary sequence and obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday.

[0048] The extraction module is used to execute S4, extract the periodic term from the intermediate sequence, and obtain the remaining sequence;

[0049] The first verification module is used to execute S5 to verify whether the remaining sequence is white noise. If it is, the initial load prediction value for the next time moment is obtained based on the remaining sequence and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the process is transferred to the construction module to execute S6.

[0050] The construction module is used to execute S6, construct an autoregressive model based on the remaining sequence, and obtain the residual sequence;

[0051] The second verification module is used to execute S7 to verify whether the residual sequence is white noise. If it is, the initial load forecast value for the next time moment is obtained based on the autoregressive model, the residual sequence, and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the residual sequence is used as the residual sequence for the next round, and the process is transferred to the construction module to execute S6.

[0052] The prediction module is used to execute S8. If the differential operation is not performed on the historical load data sequence in the acquisition module and the differential module, the initial load prediction value is used as the final load prediction value; otherwise, the inverse operation of the differential operation is performed on the initial load prediction value, and the operation result is used as the final load prediction value.

[0053] A storage medium storing a computer program for power grid load forecasting, wherein the computer program causes a computer to perform the power grid load forecasting method as described above.

[0054] An electronic device, comprising:

[0055] One or more processors;

[0056] Memory; and

[0057] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing grid load forecasting as described above.

[0058] (III) Beneficial Effects

[0059] This invention provides a method, system, storage medium, and electronic device for power grid load forecasting. Compared with existing technologies, it has the following advantages:

[0060] This invention, based on currently common time series forecasting, considers the regularity of power grid load changes, such as the impact of different months, different days of the week, and whether it is a holiday on power grid load. It adopts the influence factors of power grid load at different times, and further improves the accuracy of power grid load forecasting by combining it with the classic time series forecasting model. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 A block diagram illustrating a power grid load forecasting method provided in an embodiment of the present invention;

[0063] Figure 2 This is a flowchart illustrating a power grid load forecasting method provided in an embodiment of the present invention.

[0064] Figure 3 This is a schematic diagram of a periodic term provided in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] This application provides a power grid load forecasting method, system, storage medium, and electronic device, which solves the technical problem of low accuracy in power grid load forecasting.

[0067] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0068] Because the power grid load sequence has its own characteristics and changing patterns, for example, the power grid load changes regularly with different months, and also shows regular changes with different weeks within a week. Whether it is a holiday or not also affects the power grid load.

[0069] The accuracy of power grid load forecasting is crucial and plays a key role in power grid scheduling. To further improve forecast accuracy, this invention considers the regularity of power grid load changes, such as the impact of different months, different days of the week, and whether it is a holiday, on the power grid load based on the currently common time series forecasting. It adopts the influence factors of power grid load at different times and combines them with the classic time series forecasting model to further improve the accuracy of power grid load forecasting.

[0070] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0071] Example:

[0072] like Figures 1-2 As shown, this embodiment of the invention provides a power grid load forecasting method, which pre-constructs a power grid load forecasting model. The application process of this model includes:

[0073] S1. Obtain the historical load data sequence and perform a stationarity test. If the sequence is not stationary, proceed to S2; otherwise, proceed to S3.

[0074] S2. Perform difference operations on the non-stationary sequence until it becomes stationary;

[0075] S3. Remove time factors from the stationary sequence to obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday.

[0076] S4. Extract the periodic term from the intermediate sequence to obtain the remaining sequence;

[0077] S5. Check whether the remaining sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the remaining sequence and the time factor of the next time step, and go to S8; otherwise, go to S6.

[0078] S6. Construct an autoregressive model based on the remaining sequence to obtain the residual sequence;

[0079] S7. Check whether the residual sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the autoregressive model, the residual sequence, and the time factor of the next time step, and proceed to S8; otherwise, let the residual sequence be the residual sequence for the next round, and proceed to S6.

[0080] S8. If the differential operation was not performed on the historical load data sequence in S1 to S2, the initial load forecast value shall be used as the final load forecast value; otherwise, the inverse differential operation shall be performed on the initial load forecast value, and the result shall be used as the final load forecast value.

[0081] This invention, based on currently common time series forecasting, considers the regularity of power grid load changes, such as the impact of different months, different days of the week, and whether it is a holiday on power grid load. It adopts the influence factors of power grid load at different times, and further improves the accuracy of power grid load forecasting by combining it with the classic time series forecasting model.

[0082] The following will detail each step of the above technical solution:

[0083] In step S1, the historical load data sequence is obtained and a stationarity test is performed. If the sequence is not stationary, proceed to S2; otherwise, proceed to S3.

[0084] Optionally, the stationarity test in this step may use the ADF method.

[0085] In step S2, the non-stationary sequence is subjected to a difference operation until it becomes stationary.

[0086] For non-stationary sequences, perform difference operations, and then test the stationarity of the resulting sequence. If the sequence is still non-stationary, perform further difference operations until the resulting sequence is stationary. Generally, the sequence becomes stationary after one or two order difference operations.

[0087] In step S3, the stationary sequence is processed to remove time factors to obtain an intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday.

[0088] This invention takes into account that, due to the regularity and periodicity of people's production and daily life, power grid load data has its own unique regularity characteristics compared with other time series data. Specifically, this is reflected in the following aspects:

[0089] First, the power grid load exhibits seasonal variations. During the hot summer months, the use of a large number of refrigeration devices increases the power grid load. In the spring and autumn months, when the temperature is suitable, the power grid load decreases. In winter, the use of related heating facilities causes the power grid load to increase again.

[0090] Secondly, the power grid load not only exhibits seasonal monthly variations, but also shows corresponding variations within a week, with higher loads from Monday to Friday and lower loads on Saturdays and Sundays.

[0091] Third, during holidays, many enterprises and institutions close for the holidays, resulting in a decrease in power grid load. The power grid load gradually recovers after the holidays.

[0092] Furthermore, the power grid load also exhibits regular changes throughout the day. The power grid load is lowest in the early morning, reaches a peak in the morning as factories and enterprises start work, decreases slightly during the midday break, and then peaks again in the afternoon and evening. Starting around 9 p.m., the power grid load gradually decreases.

[0093] Based on the inherent characteristics of power grid load, such as its regularity with different months, weeks, and holidays, this invention aims to improve load forecasting accuracy by combining the aforementioned time-related factors (referred to as time factors) with time series data to predict power grid load.

[0094] Specifically, S3 includes:

[0095] S31. Obtain time factors for different months, different days of the week, and whether it is a holiday;

[0096]

[0097]

[0098]

[0099] Where month = 1, 2, ..., 12; week = 1, 2, ..., 7; i = 0 when a day is not a holiday, and i = 1 when a day is a holiday;

[0100] S32. Remove time factors from stationary sequences;

[0101]

[0102] Among them, {x t} is a stationary sequence; {y t} represents the intermediate sequence.

[0103] By processing the load data through various time factors, the influence of factors such as different months, different weeks, and holidays is eliminated. At this point, the sequence has obvious periodicity, which makes it easier for the subsequent step S4 to extract the periodic term from the processed data.

[0104] In step S4, the periodic term is extracted from the intermediate sequence as follows: Figure 3 As shown, obtain the remaining sequence.

[0105] Optionally, this step uses the X-11 seasonally adjusted model, which employs a moving average method to extract the periodic term and obtain the residual sequence, as detailed below:

[0106] z t =y t -s t

[0107] Among them, {y t} represents the intermediate sequence; {s t} represents a periodic term; {z} t} represents the remaining sequence.

[0108] In step S5, it is checked whether the remaining sequence is white noise. If so, the initial load prediction value for the next time moment is obtained based on the remaining sequence and the time factor of the next time moment, and the process proceeds to S8; otherwise, the process proceeds to S6.

[0109] Optionally, the Ljung-Box statistic is used in this step to test the residual sequence {z}. t Is it white noise?

[0110] If {zt If} is white noise, then it means {z t {z} is completely random, meaning that past actions have no impact on future development; therefore, there is no need to consider {z}. t Further analysis reveals that only the periodic term {s} needs to be considered. t Make a prediction at time t+1 and obtain s. t+1 , then y t+1 =s t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1 t+1 .

[0111] If {z t If it is not white noise, then it means that {z} t There are interrelationships between them, which require further processing in step S6, as detailed below.

[0112] In step S6, an autoregressive model is constructed based on the remaining sequence to obtain the residual sequence, including:

[0113] S61. Construct an autoregressive model based on the remaining sequence;

[0114] AR(p) = α0 + α1z t-1 +α2z t-2 +…+α p z t-p +ε t

[0115] Where AR(p) is an autoregressive model, and its order p is determined by the residual sequence {z}. t The autocorrelation coefficient and partial autocorrelation coefficient of} are determined; α0,…,α p All are coefficients; z t-1 z t-2 …z t-p Let {ε} be the residual sequences at times t-1, t-2, ..., tp, respectively; t} is a zero-mean white noise sequence;

[0116] S62. Obtain the residual sequence based on the autoregressive model;

[0117] res t =z t -(α0+α1z t-1 +α2z t-2 +…+α p z t-p )

[0118] Among them, {res t} represents the residual sequence.

[0119] It should be explained that, because the autocorrelation coefficient of the AR model has a tailing property and the partial autocorrelation coefficient has a truncation property, this step is based on the remaining sequence {z}. t}, select the most suitable order p to establish the AR(p) model. The AR(p) model has the following characteristics:

[0120] When the sample size n is sufficiently large, the autocorrelation coefficient ρ k and partial autocorrelation coefficient α kk They all approximately follow a normal distribution:

[0121]

[0122] From the properties of the normal distribution, we know that:

[0123]

[0124]

[0125] Based on the approximate distribution of sample (partial) autocorrelation coefficients, if the initial p-order sample (partial) autocorrelation coefficients are significantly greater than twice the standard deviation, and then almost 95% of the (partial) autocorrelation coefficients fall within twice the standard deviation, and the process of the non-zero (partial) autocorrelation coefficients decaying to small fluctuations is usually very abrupt, then this is generally considered as the (partial) autocorrelation coefficients being truncated, with a truncation order of p.

[0126] In step S7, it is checked whether the residual sequence is white noise. If so, the initial load forecast value for the next time step is obtained based on the autoregressive model, the residual sequence, and the time factor of the next time step, and the process proceeds to S8; otherwise, the residual sequence is used as the residual sequence for the next round, and the process proceeds to S6.

[0127] Optionally, the Ljung-Box statistic test {res} can also be used in this step. t Is it white noise?

[0128] If {res t If} is white noise, then for the periodic term {s} t Make a prediction at time t+1 and obtain s. t+1 Based on the AR(p) model, the remaining sequence {z} t The predicted value z at time t+1 is obtained by making a prediction. t+1 , then y t+1 =s t+1 +z t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1t+1 .

[0129] If {res t If it is not white noise, then it means that {res} t There are interrelationships between them, so the residual sequence is used as the remaining sequence for the next round, and the process returns to step S6 for further processing.

[0130] In step S8, if the differential operation was not performed on the historical load data sequence in S1 to S2, the initial load forecast value is used as the final load forecast value; otherwise, the inverse differential operation is performed on the initial load forecast value, and the result is used as the final load forecast value.

[0131] Specifically, during the training phase, this embodiment of the invention uses the mean squared error loss function to measure the error of the power grid load prediction model;

[0132]

[0133] Where MSE is the mean squared error loss function; a t The actual load values ​​of the training set samples at time t; t represents the load prediction value of the model at time t; n represents the sequence length of the sample.

[0134] The method provided in this invention for predicting grid load is applicable to short-term, medium-term, and long-term forecasts, achieving significantly better results compared to traditional methods that decompose time series data and then predict them separately. Furthermore, the method provided in this invention is applicable not only to overall grid load forecasting but also to forecasting the electricity load of individual users; using this method to predict the historical electricity load data of typical electricity-consuming enterprises yields excellent results.

[0135] This invention provides a power grid load forecasting system, which pre-constructs a power grid load forecasting model. This model is specifically applied to:

[0136] The acquisition module is used to execute S1, acquire historical load data sequences, and perform stationarity tests. If the sequence is not stationary, the process is transferred to the difference module to execute S2; otherwise, the process is transferred to the removal module to execute S3.

[0137] The difference module is used to execute S2, performing difference operations on non-stationary sequences until they become stationary;

[0138] The removal module is used to execute S3 to remove time factors from the stationary sequence and obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday.

[0139] The extraction module is used to execute S4, extract the periodic term from the intermediate sequence, and obtain the remaining sequence;

[0140] The first verification module is used to execute S5 to verify whether the remaining sequence is white noise. If it is, the initial load prediction value for the next time moment is obtained based on the remaining sequence and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the process is transferred to the construction module to execute S6.

[0141] The construction module is used to execute S6, construct an autoregressive model based on the remaining sequence, and obtain the residual sequence;

[0142] The second verification module is used to execute S7 to verify whether the residual sequence is white noise. If it is, the initial load forecast value for the next time moment is obtained based on the autoregressive model, the residual sequence, and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the residual sequence is used as the residual sequence for the next round, and the process is transferred to the construction module to execute S6.

[0143] The prediction module is used to execute S8. If the differential operation is not performed on the historical load data sequence in the acquisition module and the differential module, the initial load prediction value is used as the final load prediction value; otherwise, the inverse operation of the differential operation is performed on the initial load prediction value, and the operation result is used as the final load prediction value.

[0144] This invention provides a storage medium storing a computer program for power grid load forecasting, wherein the computer program causes a computer to execute the power grid load forecasting method described above.

[0145] This invention provides an electronic device, comprising:

[0146] One or more processors;

[0147] Memory; and

[0148] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing grid load forecasting as described above.

[0149] It is understood that the power grid load forecasting system, storage medium and electronic device provided in the embodiments of the present invention correspond to the power grid load forecasting method provided in the embodiments of the present invention. The explanation, examples and beneficial effects of the relevant contents can be referred to the corresponding parts of the power grid load forecasting method, and will not be repeated here.

[0150] In summary, compared with existing technologies, it has the following beneficial effects:

[0151] 1. Based on the currently common time series forecasting, the embodiments of the present invention consider the regularity of the changes in the power grid load itself, such as the impact of different months, different days of the week, and whether it is a holiday on the power grid load. The invention adopts the influence factors of the power grid load at different times, and further improves the accuracy of power grid load forecasting by combining the classic time series forecasting model.

[0152] 2. The method provided in this embodiment of the invention for predicting grid load is applicable to short-term, medium-term, and long-term forecasts, achieving excellent results compared to traditional methods that decompose time series data and then predict them separately. Furthermore, the method provided in this embodiment of the invention is applicable not only to overall grid load forecasting but also to forecasting the electricity load of individual users; using this method to predict the historical electricity load data of typical electricity-consuming enterprises yields excellent results.

[0153] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0154] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power grid load forecasting method, characterized in that, The process of pre-constructing a power grid load forecasting model and applying this model includes: S1. Obtain the historical load data sequence and perform a stationarity test. If the sequence is not stationary, proceed to S2; otherwise, proceed to S3. S2. Perform difference operations on the non-stationary sequence until it becomes stationary; S3. Remove time factors from the stationary sequence to obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday. S4. Extract the periodic term from the intermediate sequence to obtain the remaining sequence; S5. Check whether the remaining sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the remaining sequence and the time factor of the next time step, and go to S8; otherwise, go to S6. S6. Construct an autoregressive model based on the remaining sequence to obtain the residual sequence; S7. Check whether the residual sequence is white noise. If so, obtain the initial load forecast value for the next time step based on the autoregressive model, the residual sequence, and the time factor for the next time step, and proceed to S8; otherwise, let the residual sequence be used as the residual sequence for the next round, and proceed to S6. S8. If the differential operation was not performed on the historical load data sequence in S1~S2, the initial load forecast value shall be used as the final load forecast value; otherwise, the inverse operation of the differential operation shall be performed on the initial load forecast value, and the result shall be used as the final load forecast value. In step S5, the initial load forecast value for the next time moment is obtained based on the remaining sequence and the time factor of the next time moment, including: For periodic terms {s t Make a prediction at time t+1 and obtain s. t+1 , then y t+1 =s t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1 t+1 ; In step S7, the initial load forecast value for the next time step is obtained based on the autoregressive model, the residual sequence, and the time factor for the next time step, including: For periodic terms {s t Make a prediction at time t+1 and obtain s. t+1 Based on the AR(p) model, the remaining sequence {z} t The predicted value z at time t+1 is obtained by making a prediction. t+1 , then y t+1 =s t+1 +z t+1 , will y t+1 Multiply by the relevant time factor at that moment to obtain the stationary sequence {x}. t The initial load forecast value x at time t+1 t+1 ; where AR(p) represents an autoregressive model of order p.

2. The power grid load forecasting method as described in claim 1, characterized in that, S3 includes: S31. Obtain time factors for different months, different days of the week, and whether it is a holiday; Where month=1,2,…,12; week=1,2,…,7; i=0 when a day is not a holiday, and i=1 when a day is a holiday; S32. Remove time factors from stationary sequences; Among them, {x t } is a stationary sequence; {y t } represents the intermediate sequence.

3. The power grid load forecasting method as described in claim 1, characterized in that, In step S4, the X-11 seasonal adjustment model is used to extract the periodic term and obtain the remaining sequence, as follows: Among them, {y t } represents the intermediate sequence; {s t } represents a periodic term; {z} t } represents the remaining sequence.

4. The power grid load forecasting method as described in claim 1, characterized in that, S6 includes: S61. Construct an autoregressive model based on the remaining sequence; Where AR(p) is an autoregressive model, and its order p is determined by the residual sequence {z}. t The autocorrelation coefficient and partial autocorrelation coefficient of} are determined; α0,…,α p All are coefficients; z t-1 ,z t-2 ,…,z t-p The remaining sequences at times t-1, t-2, ..., tp are respectively; {ε t } is a zero-mean white noise sequence; S62. Obtain the residual sequence based on the autoregressive model; Among them, {res t } represents the residual sequence.

5. The power grid load forecasting method according to any one of claims 1 to 4, characterized in that, During the training phase, the mean squared error loss function is used to measure the error of the power grid load forecasting model. Where MSE is the mean squared error loss function; a t The actual load values ​​of the training set samples at time t; t represents the model's predicted load value at time t; n represents the sequence length of the sample.

6. A power grid load forecasting system, characterized in that, For the power grid load forecasting method as described in claim 1, a power grid load forecasting model is pre-constructed, and this model is specifically applied to: The acquisition module is used to execute S1, acquire historical load data sequences, and perform stationarity tests. If the sequence is not stationary, the process is transferred to the difference module to execute S2; otherwise, the process is transferred to the removal module to execute S3. The difference module is used to execute S2, performing difference operations on non-stationary sequences until they become stationary; The removal module is used to execute S3 to remove time factors from the stationary sequence and obtain the intermediate sequence; the time factors include different months, different days of the week, and whether it is a holiday. The extraction module is used to execute S4, extract the periodic term from the intermediate sequence, and obtain the remaining sequence; The first verification module is used to execute S5 to verify whether the remaining sequence is white noise. If it is, the initial load prediction value for the next time moment is obtained based on the remaining sequence and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the process is transferred to the construction module to execute S6. The construction module is used to execute S6, construct an autoregressive model based on the remaining sequence, and obtain the residual sequence; The second verification module is used to execute S7 to verify whether the residual sequence is white noise. If it is, the initial load forecast value for the next time moment is obtained based on the autoregressive model, the residual sequence, and the time factor of the next time moment, and the process is transferred to the prediction module to execute S8; otherwise, the residual sequence is used as the residual sequence for the next round, and the process is transferred to the construction module to execute S6. The prediction module is used to execute S8. If the differential operation is not performed on the historical load data sequence in the acquisition module and the differential module, the initial load prediction value is used as the final load prediction value; otherwise, the inverse operation of the differential operation is performed on the initial load prediction value, and the operation result is used as the final load prediction value.

7. A storage medium, characterized in that, It stores a computer program for power grid load forecasting, wherein the computer program causes the computer to perform the power grid load forecasting method as described in any one of claims 1 to 5.

8. An electronic device, characterized in that, include: One or more processors; Memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including methods for performing the power grid load forecasting method as described in any one of claims 1 to 5.

Images