Power load prediction method based on multi-time scale fusion

Abstract

The invention provides a power load prediction method based on multi-time scale fusion, and the method comprises the steps: collecting the historical time series data of a power load, carrying out the decomposition processing of the historical time series data of the power load, obtaining a multi-time scale intrinsic mode component set, and carrying out the prediction of the multi-time scale intrinsic mode component set according to the multi-time scale load prediction demand. Calculating the correlation between the historical time sequence data of the power load and the coupling factors, and determining strong correlation coupling factors; performing preset processing on the multi-time scale intrinsic mode component set and the strong correlation coupling factor to obtain a plurality of processed data sets; inputting the plurality of data sets into a time sequence analysis model to obtain a plurality of time scale prediction results; and evaluating the precision of the plurality of time scale prediction results, and determining a power future day-level load prediction result. According to the method, static trend and dynamic change are both considered, the comprehensiveness, precision and robustness of load prediction are improved, scientific support is provided for scheduling optimization and resource allocation of the smart power grid, and improvement of safety and reliability of operation of the power grid is facilitated.

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 based on multi-time scale fusion. Background Art

[0002] As the power system develops towards intelligence and refinement, load forecasting, as the core technology for power system planning, scheduling and operation, is crucial to the efficient allocation of power resources and system stability.

[0003] However, load data often exhibits significant nonlinearity, nonstationarity, and multi-timescale characteristics, posing significant challenges for load forecasting. Specifically, load data is influenced by a variety of factors, including seasonal variations, weather conditions, and socioeconomic activities, resulting in complex fluctuations. These characteristics cause load data to exhibit distinct dynamic behaviors at different time scales. For example, hourly load fluctuations typically reflect short-term fluctuations in electricity demand, while daily load fluctuations reflect longer-term trends and cyclical characteristics. Single-time-scale load forecasting methods, which can only capture characteristics at a single scale, often fail to fully characterize the global and local characteristics of load data, resulting in significant limitations in practical applications. On the one hand, existing methods have limited responsiveness to short-term fluctuations, making them unable to meet the power grid's demand for refined management of rapidly changing loads. On the other hand, their inaccurate predictions of long-term trends can impact medium- and long-term power resource planning. The failure to fully explore and utilize the characteristics of load data at different time scales directly limits the accuracy and applicability of existing forecasting methods. Summary of the Invention

[0004] In order to overcome the above technical defects, the present application provides a power load forecasting method based on multi-time scale fusion.

[0005] This application is implemented according to the following technical solutions:

[0006] This application provides a method for power load forecasting based on multi-time scale fusion, including:

[0007] Collecting historical time series data of power load, wherein the historical time series data of power load includes hourly historical load data and daily historical load data;

[0008] Decomposing the hourly historical load data and the daily historical load data respectively to obtain a multi-time-scale eigenmode collection; wherein the multi-time-scale eigenmode collection includes an hourly eigenmode collection and a daily eigenmode collection;

[0009] In response to the multi-time scale load forecasting requirements, the correlation between historical time series data of power load and coupling factors is calculated to identify the strongly correlated coupling factors.

[0010] Performing preset processing on the multi-time-scale intrinsic mode collection and the strongly correlated coupling factors to obtain multiple processed data sets;

[0011] Inputting the multiple data sets into a time series analysis model to obtain prediction results at multiple time scales;

[0012] The accuracy of the prediction results of the multiple time scales is evaluated to determine the future daily load prediction result of the power.

[0013] Optionally, the hourly historical load data and the daily historical load data are decomposed separately to obtain a multi-time-scale eigenmode set, including:

[0014] The variational mode decomposition method is used to decompose the hourly historical load data and the daily historical load data to obtain a multi-time scale eigenmode set, specifically:

[0015] For hourly / daily historical load data, a constrained variational optimization problem is constructed. The construction process includes:

[0016]

[0017] Among them, K is the number of decomposition modes; {u k} represents the set of modal components, {ω k} represents the set of modal center frequencies, where k = 1, 2, ..., K; is the partial derivative with respect to time; t is time; δ(t) is the Dirac generalized function; j is the imaginary unit, j 2 =-1; * represents the convolution operator; f(t) is the hourly historical load data / daily historical load data; st represents the constraint condition.

[0018] Lagrange multipliers and quadratic penalty factors are introduced to solve the variational optimization problem. The reconstructed variational optimization model includes:

[0019]

[0020] Where L is the Lagrange function; λ is the Lagrange multiplier; α is the quadratic penalty factor; <,> represents the inner product, and λ(t) represents the Lagrange multiplier;

[0021] The alternating direction multiplier method is used to iteratively solve the modal components and center frequencies to obtain the multi-time scale eigenmode set. The iterative formula includes:

[0022]

[0023]

[0024] Among them, u k (ω), f ω ,λ(ω) are the signal u k (t), f(t) and λ(t); n is the number of iterations; represents u obtained in the n+1th iteration k (ω) value; λ n (ω) represents the value of λ(t) obtained in the nth iteration; ω is the frequency, is the center frequency of the nth iteration; τ is the noise tolerance.

[0025] Optionally, the calculation of the correlation between the historical time series data of the power load and the coupling factors for the multi-time-scale load forecasting demand and the determination of the strongly correlated coupling factors includes:

[0026] The unified information coefficient method is used to calculate the correlation between historical time series data of power load and coupling factors for multi-time scale load forecasting needs, and the strongly correlated coupling factors are determined.

[0027] Optionally, performing preset processing on the multi-time-scale intrinsic mode collection and the strongly correlated coupling factors to obtain multiple processed data sets includes:

[0028] The multi-time-scale intrinsic mode collection and the strongly correlated coupling factors are normalized to obtain a normalized hourly data set and a normalized daily data set.

[0029] Optionally, inputting the multiple data sets into a time series analysis model to obtain multiple time scale prediction results includes:

[0030] Inputting the normalized hourly data set into the long short-term memory network model to obtain hourly prediction results;

[0031] Aggregating the hourly prediction results to obtain aggregated prediction results;

[0032] The normalized daily data set is input into the long short-term memory network model to obtain the daily prediction result.

[0033] Optionally, the accuracy of the prediction results of the multiple time scales is evaluated to determine the future daily load of electricity;

[0034] include:

[0035] Performing an accuracy evaluation on the cluster prediction result to determine the accuracy of the cluster prediction result;

[0036] Determining a weighting coefficient α1 of the clustering prediction result based on the accuracy of the clustering prediction result;

[0037] Performing an accuracy assessment on the daily-level forecast results to determine the accuracy of the daily-level forecast results;

[0038] Determine a weighting coefficient α2 of the aggregated prediction result based on the accuracy of the daily-level prediction result;

[0039] Based on α1 and α2, the aggregated prediction result and the daily-level prediction result are weightedly fused to determine the future daily-level load prediction result of electricity.

[0040] Optionally, the performing aggregation processing on the hourly prediction results to obtain an aggregated prediction result includes:

[0041] Aggregation is performed based on the following formula:

[0042]

[0043] Where: P hour (t) is the hourly prediction result; P day,sum (i) is the aggregate prediction result of day i.

[0044] In a second aspect, the present application provides an electronic device, comprising: a memory and a processor;

[0045] The memory is used to store computer programs;

[0046] The processor is configured to call the computer program to execute the method described above.

[0047] In a third aspect, the present application provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed on an electronic device, the electronic device implements the method described above.

[0048] In a fourth aspect, the present application provides a computer program product, including a computer program, which, when executed on an electronic device, enables the electronic device to implement the method described above.

[0049] The specific implementation methods of the second to fourth aspects of this application can refer to the implementation method of the first aspect above, and will not be repeated here.

[0050] This application has the following beneficial effects:

[0051] This application is based on a load forecasting method that integrates multiple time scales. By integrating hourly and daily load forecasts, it achieves the coordinated forecasting of short-term fluctuations and long-term trends, avoiding the shortcomings of a single time scale forecasting model in terms of accuracy and comprehensiveness. The variational mode decomposition method is used to process load data to ensure the stability and efficiency of the decomposition results. The unified information coefficient method is used to screen strongly correlated coupling factors, avoiding the interference of irrelevant or weakly correlated factors in the forecasting process, thereby improving the validity of the model input data. This method takes into account both static trends and dynamic changes, improves the comprehensiveness, accuracy and robustness of load forecasting, provides scientific support for the scheduling optimization and resource allocation of smart grids, and helps improve the safety and reliability of grid operation.

[0052] In addition to the above-described purposes, features and advantages, the present application has other purposes, features and advantages. The present application will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0054] Figure 1 This is a flow chart of a load forecasting method based on multi-time scale fusion provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] The embodiments of the present application are described in detail below with reference to the accompanying drawings, but the present application can be implemented in many different ways as defined and covered by the claims.

[0056] Example 1:

[0057] In order to solve the problems raised in the above background technology, Figure 1 As shown, the embodiment of the present application proposes a power load forecasting method based on multi-time scale fusion, including:

[0058] Step S101: collecting historical time series data of electric load, wherein the historical time series data of electric load includes hourly historical load data and daily historical load data;

[0059] The historical time series data of power load refers to a series of data sets related to the power system that are arranged in chronological order, including hourly historical load data and daily historical load data. It can include: (1) Power load data, which is the core part of the historical time series data of power load. It records the load carried by the power system at different times and is usually measured in units of active power (kilowatts, megawatts, etc.). It includes the power load of various users such as residents, industry, and commerce, reflecting the actual consumption of electricity. (2) Power generation data: It contains the power generation of various power generation types at different times, such as thermal power generation, hydropower generation, wind power generation, photovoltaic power generation, etc. (3) Power grid operation data: It involves parameters such as voltage, current, power factor, and frequency of the power grid. These data reflect the operating status of the power grid and are crucial to ensuring the safe and stable operation of the power grid. They help to promptly detect problems such as power grid failures and voltage anomalies. (4) Meteorological data: Although it is not direct power data, meteorological factors have a significant impact on power load, so it is often included as relevant data. For example, meteorological information such as temperature, humidity, wind speed, and sunshine duration can be used to analyze the correlation between meteorological conditions and power load, thereby improving the accuracy of load forecasting.

[0060] Step S102: Decomposing the hourly historical load data and the daily historical load data respectively to obtain a multi-time-scale eigenmode collection; wherein the multi-time-scale eigenmode collection includes an hourly eigenmode collection and a daily eigenmode collection;

[0061] After obtaining the above-mentioned historical time series data of power load, the variational mode decomposition method is used to decompose the hourly historical load data and the daily historical load data respectively to obtain a multi-time scale eigenmode collection. The multi-time scale eigenmode collection mainly includes the hourly eigenmode collection and the daily eigenmode collection. The calculation process is as follows:

[0062] For hourly / daily historical load data, a constrained variational optimization problem is constructed. The construction process includes:

[0063]

[0064] Among them, K is the number of decomposition modes; {u k} represents the set of modal components, {ω k} represents the set of modal center frequencies, where k = 1, 2, …, K; is the partial derivative with respect to time; t is time; δ(t) is the Dirac generalized function; j is the imaginary unit, j 2 =-1; * represents the convolution operator; f(t) is the hourly historical load data / daily historical load data; st represents the constraint condition.

[0065] In this embodiment, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B means: A or B. As mentioned above, when solving for hourly historical load data, f(t) takes hourly historical load data, and the obtained {u k} and {ω k} are the hourly eigenmode set and the corresponding modal center frequency set. When solving for the daily historical load data, f(t) takes the daily historical load data, and the obtained {u k} and {ω k} are the daily-level eigenmode sets and the corresponding modal center frequency sets, respectively. The subsequent character “ / ” is used in a similar way.

[0066] Lagrange multipliers and quadratic penalty factors are introduced to solve the variational optimization problem. The reconstructed variational optimization model includes:

[0067]

[0068] Where L is the Lagrange function; λ is the Lagrange multiplier; α is the quadratic penalty factor; <,> represents the inner product, and λ(t) represents the Lagrange multiplier.

[0069] The alternating direction multiplier method is used to iteratively solve the modal components and center frequencies to obtain the multi-time scale eigenmode set. The iterative formula includes:

[0070]

[0071]

[0072]

[0073] Among them, u k (ω), f ω ,λ(ω) are the signal u k (t), f(t) and λ(t); n is the number of iterations; represents u obtained in the n+1th iteration k (ω) value; λ n (ω) represents the value of λ(t) obtained in the nth iteration; ω is the frequency, is the center frequency of the nth iteration; τ is the noise tolerance.

[0074] It should be noted that formula (1) is the objective function, and its input data is f(t), that is, the collected hourly historical load data / daily historical load data, and the output is the modal component set {u k} and {ω k}, that is, the hourly modal component set / daily modal component set, formulas (2)-(5) are all constraints. By solving the objective function and constraints, the objective function value of formula (1) is minimized. k} and {ω k}.

[0075] Step S103: Calculate the correlation between the historical time series data of power load and coupling factors according to the multi-time-scale load forecasting requirements, and determine the strongly correlated coupling factors;

[0076] The need for multi-timescale load forecasting in fields such as power systems requires forecasting power loads for different timeframes, based on diverse application scenarios and decision-making requirements. Power load changes do not exist in isolation but rather interact with multiple factors. Correlation calculations can reveal the inherent connections between historical time series data on power load and coupled factors such as temperature, humidity, economic development level, and industrial structure, helping to understand how these factors influence load changes. The correlations between load and various factors may vary across different timescales. For example, ultra-short-term loads may be more sensitive to sudden changes in temperature, short-term loads may be more closely related to factors such as weekdays and holidays, and long-term loads may be more strongly correlated with factors such as economic growth and demographic changes. Calculating correlations helps analyze the dominant factors and patterns of load changes across different timescales.

[0077] Under different scenarios and time requirements, it is necessary to calculate the key factors affecting the current scenario load in the corresponding scenario, that is, the strongly correlated coupling factors. The specific calculation process is as follows:

[0078] It should be noted that the process of calculating the strong correlation coupling factor is the same for hourly load data and daily load data.

[0079] According to the unified division method, the historical time series data of power load (hourly historical load data / daily historical load data) X and the coupling factor Y are divided into several segments:

[0080]

[0081]

[0082] Among them: x With l y are the partition unit lengths of X and Y respectively; x max with x min are the maximum and minimum values ​​of X respectively; y max with y min are the maximum and minimum values ​​of Y respectively; a and b are the number of segments of X and Y respectively; n 0.6Represents the partition grid size, where n is the number of samples; n 0.6 Used to control the fineness of partitioning to avoid overfitting or underfitting;

[0083] For two sets of eigenvectors X and Y, calculate their marginal probability density functions p(x) and p(y), respectively, as well as the joint probability density function p(x,y) of X and Y;

[0084] Then substitute it into the following formula (8) to calculate its mutual information coefficient. The mutual information coefficient calculation model is:

[0085]

[0086] Among them: I MI (X; Y) is the mutual information coefficient between X and Y;

[0087] Furthermore, according to the unified partitioning method, the results obtained by formulas (7) and (8) are substituted into formula (9) to calculate the unified information coefficient between X and Y:

[0088]

[0089] Among them: I UIC (X; Y) is the unified information coefficient between X and Y; min(a,b) is the minimum value of a and b.

[0090] Here, a threshold can be set according to actual needs. After calculating the value of the unified information coefficient between the power load historical time series data X and the coupling factor Y, it is compared with the threshold. If it is greater than the threshold, the coupling factor Y is determined to be a strongly correlated coupling factor.

[0091] Step S104: performing preset processing on the multi-time-scale eigenmode collection and the strongly correlated coupling factors to obtain a plurality of processed data sets;

[0092] After obtaining the hourly eigenmode collection and its corresponding strong coupling related factors, they are normalized to obtain the normalized hourly data set.

[0093] After obtaining the daily-level eigenmode collection and its corresponding strong coupling related factors, they are normalized to obtain the normalized daily-level data set;

[0094] The specific calculation process is as follows:

[0095]

[0096]

[0097] Where: m is the original data value in the hourly eigenmode set / daily eigenmode set; m min is the minimum value of the hourly eigenmode set / daily eigenmode set; m max is the maximum value in the hourly eigenmode set / daily eigenmode set; m' is the normalized data of the original data value in the hourly eigenmode set / daily eigenmode set. n is the original data value of the strongly correlated coupling factor; n min is the minimum value in the strongly correlated coupling factor data; n max is the maximum value in the strongly correlated coupling factor data; n' is the normalized data of the strongly correlated coupling factor data.

[0098] The dataset obtained by combining the calculated m' and n' is the normalized hourly dataset / normalized daily dataset.

[0099] Step S105: inputting the multiple data sets into a time series analysis model to obtain multiple time scale prediction results;

[0100] By inputting the normalized hourly dataset and the normalized daily dataset obtained above into the time series analysis model respectively, we can obtain prediction results at multiple time scales.

[0101] The time series analysis model here can be a long short-term memory network (LSTM) model or other time series analysis models, and this application does not impose too many restrictions here.

[0102] By inputting the normalized hourly data set into the long short-term memory network (LSTM) model, hourly prediction results can be obtained.

[0103] Since the ultimate goal is to predict daily power load data and reduce the impact of data noise, it is necessary to aggregate the hourly forecast results to obtain aggregated forecast results. The calculation process is as follows:

[0104]

[0105] Where: P hour (t) is the hourly prediction result; P day,sum (i) is the aggregate prediction result of day i.

[0106] The normalized daily data set is input into the long short-term memory network model to obtain the daily prediction results.

[0107] The processing of the above data in the long short-term memory network model is as follows:

[0108] Calculate the forget gate:

[0109] f t =σ(W f ·[h t-1 ,x t ]+b f ) (13)

[0110] Where: f t Is the output of the forget gate, with a value range of (0,1), which determines how much past information is retained; W f is the weight matrix of the forget gate; b f is the paranoid term of the forget gate; h t-1 Indicates the hidden state of the previous moment; x t is the input at the current moment (i.e., the normalized hourly and daily datasets); σ(·) is the Sigmoid activation function, which ensures that the output is between (0, 1).

[0111] Compute the input gate:

[0112] i t =σ(W i ·[h t-1 ,x t ]+b i ) (14)

[0113]

[0114] Where: i t Is the output of the input gate, with a value range of (0,1), which determines how much new information is written into the cell state; W i is the weight matrix of the input gate; b i is the bias term of the input gate; is the candidate cell state, with a value range of (-1,1), used to update C t ;W c is the weight matrix updated for the cell state; b c is the bias term for cell state update; tanh(·) represents the hyperbolic tangent function, which keeps the value between (-1,1).

[0115] Cell status update:

[0116]

[0117] Where: C t is the cell state at the current moment; C t-1 is the cell state at the previous moment; the symbol ⊙ represents element-by-element multiplication;

[0118] Calculate the output gate:

[0119] o t =σ(Wo ·[h t-1 ,x t ]+b o ) (17)

[0120] h t =o t ⊙tanh(C t ) (18)

[0121] Among them: t is the output of the output gate, with a value range of (0,1), which determines how much information the hidden state contains; W o is the weight matrix of the output gate; b o is the bias term of the output gate; h t is the hidden state at the current moment, which is the output of LSTM; tanh(C t ) represents a nonlinear transformation of the cell state to ensure that the output is between (-1,1).

[0122]

[0123] in: is the final prediction value (i.e., hourly prediction result and daily prediction result); Wy is the weight matrix; by is the bias term.

[0124] Step S106: Evaluate the accuracy of the prediction results of the multiple time scales to determine the future daily power load.

[0125] After obtaining the cluster prediction result, the accuracy of the cluster prediction result is evaluated to determine the accuracy of the cluster prediction result, and then the weighting coefficient of the cluster prediction result is determined based on the accuracy of the cluster prediction result.

[0126] After obtaining the daily-level prediction results, the accuracy of the daily-level prediction results is evaluated to determine the accuracy of the daily-level prediction results, and then the weighting coefficient of the aggregated prediction results is determined based on the accuracy of the daily-level prediction results.

[0127] It should be noted that the accuracy indicators selected in this application are mean absolute percentage error (MAPE), root mean square error (RMSE) or mean absolute error (MAE). Here, the MAPE accuracy indicator is taken as an example to illustrate the calculation process of determining the weighting coefficient as follows:

[0128]

[0129] Where MAPE is the mean absolute percentage error; the smaller the value of MAPE, the higher the accuracy of the prediction result.

[0130] After calculating the corresponding accuracy index, the weighting coefficient is calculated using the following formula:

[0131]

[0132]

[0133] Where α1 represents the weighted coefficient of the aggregated prediction result; α2 represents the weighted coefficient of the daily prediction result (fusion weight); MAPE sum The accuracy of the aggregated prediction results; MAPE day The accuracy of the daily forecast results;

[0134] Based on α1 and α2, the aggregated prediction results and the daily-level prediction results are weightedly integrated to determine the future daily-level load prediction results of the power industry. The calculation process is as follows:

[0135] P day,final (i) = α1·P day,sum (i)+α2·P day (i) (23)

[0136] Where, P day (i) is the daily level prediction result for day i; P day,final (i) is the daily load forecast result for the future i-th day, P day,sum (i) is the aggregate prediction result on day i.

[0137] In summary, this application is based on a load forecasting method that integrates multiple time scales. By integrating hourly and daily load forecasts, it achieves the coordinated forecasting of short-term fluctuations and long-term trends, avoiding the shortcomings of a single time scale forecasting model in terms of accuracy and comprehensiveness. The variational mode decomposition method is used to process load data to ensure the stability and efficiency of the decomposition results. The unified information coefficient method is used to screen strongly correlated coupling factors, thereby avoiding the interference of irrelevant or weakly correlated factors in the forecasting process, thereby improving the validity of the model input data. This method takes into account both static trends and dynamic changes, improves the comprehensiveness, accuracy and robustness of load forecasting, provides scientific support for the scheduling optimization and resource allocation of smart grids, and helps improve the safety and reliability of grid operation.

[0138] Example 2:

[0139] This embodiment provides an electronic device, including: a memory and a processor;

[0140] The memory is used to store computer programs;

[0141] The processor is configured to call the computer program to execute the method described in the first embodiment.

[0142] Example 3:

[0143] This embodiment provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed on an electronic device, the electronic device implements the method according to the first embodiment.

[0144] Example 4:

[0145] This embodiment provides a computer program product, including a computer program. When the computer program is run on an electronic device, the electronic device implements the method described in the first embodiment.

[0146] The specific implementation methods of a system, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of the present application can refer to the specific embodiments of the above-mentioned method and will not be repeated here.

[0147] Obviously, those skilled in the art should understand that the above-mentioned units or steps of the present application can be implemented using a general-purpose computing device. They can be concentrated on a single computing device or distributed across a network composed of multiple computing devices. Alternatively, they can be implemented using program codes executable by the computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0148] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Various modifications and variations are possible for those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for power load forecasting based on multi-time scale fusion, characterized in that: include: Collecting historical time series data of power load, wherein the historical time series data of power load includes hourly historical load data and daily historical load data; Decomposing the hourly historical load data and the daily historical load data respectively to obtain a multi-time-scale eigenmode collection; wherein the multi-time-scale eigenmode collection includes an hourly eigenmode collection and a daily eigenmode collection; In response to the multi-time scale load forecasting requirements, the correlation between historical time series data of power load and coupling factors is calculated to identify the strongly correlated coupling factors. Performing preset processing on the multi-time-scale intrinsic mode collection and the strongly correlated coupling factors to obtain multiple processed data sets; Inputting the multiple data sets into a time series analysis model to obtain prediction results at multiple time scales; The accuracy of the prediction results of the multiple time scales is evaluated to determine the future daily load prediction result of the power.

2. The method according to claim 1, characterized in that The hourly historical load data and the daily historical load data are decomposed and processed respectively to obtain a multi-time-scale eigenmode set, including: The variational mode decomposition method is used to decompose the hourly historical load data and the daily historical load data to obtain a multi-time scale eigenmode set, specifically: For hourly / daily historical load data, a constrained variational optimization problem is constructed. The construction process includes: Among them, K is the number of decomposition modes; {u k } represents the set of modal components, {ω k } represents the set of modal center frequencies, where k = 1, 2, …, K; is the partial derivative with respect to time; t is time; δ(t) is the Dirac generalized function; j is the imaginary unit, j 2 =-1; * represents the convolution operator; f(t) is the hourly historical load data / daily historical load data; st represents the constraint condition. Lagrange multipliers and quadratic penalty factors are introduced to solve the variational optimization problem. The reconstructed variational optimization model includes: Where L is the Lagrange function; λ is the Lagrange multiplier; α is the quadratic penalty factor; <,> represents the inner product, and λ(t) represents the Lagrange multiplier; The alternating direction multiplier method is used to iteratively solve the modal components and center frequencies to obtain the multi-time scale eigenmode set. The iterative formula includes: Among them, u k (ω), f ω ,λ(ω) are the signal u k (t), f(t) and λ(t); n is the number of iterations; represents u obtained in the n+1th iteration k (ω) value; λ n (ω) represents the value of λ(t) obtained in the nth iteration; ω is the frequency, is the center frequency of the nth iteration; τ is the noise tolerance.

3. The method according to claim 1, characterized in that The method of calculating the correlation between the historical time series data of power load and coupling factors and determining the strongly correlated coupling factors for the multi-time scale load forecasting needs includes: The unified information coefficient method is used to calculate the correlation between historical time series data of power load and coupling factors for multi-time scale load forecasting needs, and the strongly correlated coupling factors are determined.

4. The method according to claim 1, wherein The step of performing preset processing on the multi-time-scale intrinsic mode collection and the strongly correlated coupling factors to obtain multiple processed data sets includes: The multi-time-scale intrinsic mode collection and the strongly correlated coupling factors are normalized to obtain a normalized hourly data set and a normalized daily data set.

5. The method according to claim 4, characterized in that Inputting the multiple data sets into a time series analysis model to obtain multiple time scale prediction results includes: Inputting the normalized hourly data set into the long short-term memory network model to obtain hourly prediction results; Aggregating the hourly prediction results to obtain aggregated prediction results; The normalized daily data set is input into the long short-term memory network model to obtain the daily prediction result.

6. The method according to claim 5, characterized in that The hourly prediction results are aggregated. Get the aggregate prediction results, including: Aggregation is performed based on the following formula: Where: P hour (t) is the hourly prediction result; P day,sum (i) is the aggregate prediction result of day i.

7. The method according to claim 5, characterized in that The step of evaluating the accuracy of the prediction results at multiple time scales to determine the future daily load of electricity includes: Performing an accuracy evaluation on the cluster prediction result to determine the accuracy of the cluster prediction result; Determining a weighting coefficient α1 of the clustering prediction result based on the accuracy of the clustering prediction result; Performing an accuracy assessment on the daily-level forecast results to determine the accuracy of the daily-level forecast results; Determine a weighting coefficient α2 of the aggregated prediction result based on the accuracy of the daily-level prediction result; Based on α1 and α2, the aggregated prediction result and the daily-level prediction result are weightedly fused to determine the future daily-level load prediction result of electricity.

8. An electronic device, characterized in that: include: memory and processor; The memory is used to store computer programs; The processor is configured to call the computer program to execute the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed on an electronic device, the electronic device implements the method according to any one of claims 1 to 7.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed on an electronic device, the electronic device implements the method according to any one of claims 1 to 7.

Images