Meteorological sensitive load prediction method, device, equipment and medium

Abstract

The invention discloses a meteorological sensitive load prediction method, device and equipment and a medium, and relates to the technical field of electric power prediction, and the method comprises the steps: obtaining original sequence data; the method comprises the following steps: decomposing original sequence data by adopting a CEEMDAN decomposition method to obtain a plurality of IMF components with different frequencies and scales; calculating sample entropies of the plurality of IMF components with different frequencies and scales, and aggregating the plurality of IMF components with different frequencies and scales into a first high-frequency sub-sequence, an intermediate-frequency sub-sequence and a low-frequency sub-sequence based on the sample entropies; performing secondary decomposition on the first high-frequency sub-sequence by adopting a variational mode decomposition method to obtain a second high-frequency sub-sequence; respectively inputting the second high-frequency sub-sequence, the intermediate-frequency sub-sequence and the low-frequency sub-sequence into the LSTM neural network for prediction to obtain respective prediction results of the second high-frequency sub-sequence, the intermediate-frequency sub-sequence and the low-frequency sub-sequence, and carrying out weighted fusion on the respective prediction results of the second high-frequency sub-sequence, the intermediate-frequency sub-sequence and the low-frequency sub-sequence to obtain a prediction result of the second high-frequency sub-sequence; and obtaining a prediction result of the meteorological sensitive load.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power forecasting, and more specifically, to a method, device, equipment and medium for forecasting meteorological sensitive loads. Background Art

[0002] With the intensification of global climate change and the frequent occurrence of extreme weather events, the frequency and intensity of use of meteorological sensitive loads have increased significantly. This change directly affects the load curve of the power system, especially during the high temperature period in summer, when meteorological sensitive loads often account for a large proportion of the total power load, resulting in a sharp increase in power load in a short period of time, which puts tremendous pressure on the safe and stable operation of the power grid.

[0003] Meteorological sensitive load forecasting methods can be roughly divided into two types: physical model driven and data driven. The physical model driven method deeply explores the basic mechanism of meteorological sensitive load consumption and simulates and predicts meteorological sensitive load by building accurate equipment operation and user behavior models. This method can effectively capture the dynamic changes within the system and the interaction between devices, but it often relies on the collection of complex data and the input of a large number of physical parameters. The physical model driven method requires the use of specialized computer simulation software for analysis, including Energy Plus, TRNSYS, etc. However, the main limitations of the physical model driven method are its high complexity and long simulation process. In addition, meteorological sensitive loads are affected by many factors, and the uncertainty of user behavior and external environment has a great impact on the results of simulation software, which makes it particularly difficult to build a physical prediction model that is both general and accurate.

[0004] In contrast, the data-driven method obtains future forecast values ​​by analyzing historical load data and related environmental variables. It has the advantages of low cost, relatively simple model, high computational efficiency and easy implementation, and does not require in-depth knowledge of the physical expertise of the system. Data-driven methods mainly include statistical methods and machine learning methods. In terms of statistical methods, the prior art uses an improved seasonal exponential smoothing method to construct a meteorological sensitive load forecasting model based on the daily average temperature. The prior art uses a simple autoregressive model and weather forecast to construct a meteorological sensitive load forecasting model. The results show that weather data can improve the forecasting accuracy to a certain extent. The prior art proposes a shopping mall air conditioning load forecasting method based on multivariate nonlinear regression, and introduces the actual energy consumption coefficient to preprocess the data to reduce the impact of external interference on the load. However, statistical methods are difficult to capture complex nonlinear relationships and long-term dependencies, which may lead to limitations in forecasting performance in practical applications. In order to overcome the limitations of statistical methods, machine learning methods have been widely used in data-driven meteorological sensitive load forecasting models. The prior art also uses extreme learning machines to construct meteorological sensitive load forecasting models, and uses wild dog optimization algorithms to optimize model parameters, thereby overcoming the uncertainty of model learning parameters. The existing technology uses a method combining principal component analysis and long short-term memory (LSTM) to obtain the prediction results of meteorological sensitive loads based on the dimensionality reduction of multiple influencing factors. The existing technology uses singular value decomposition to reduce the noise of air conditioning load data, and then combines weather data and LSTM neural network to build an air conditioning load prediction model for college student dormitories. The existing technology designs a framework combining artificial neural network (ANN) and ensemble learning to achieve dynamic prediction of air conditioning cooling load.

[0005] Existing technologies show that in addition to developing better model structures to improve prediction accuracy, data decomposition methods such as empirical mode decomposition (EMD), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), and variational mode decomposition (VMD) can reveal the implicit patterns and trends in the data more deeply by decomposing the data and extracting key features of different frequencies, thereby improving prediction accuracy.

[0006] Although the methods provided by the existing technology have improved the accuracy of meteorological sensitive load forecasting to a certain extent through better model structure or decomposition technology, there are still the following shortcomings: 1) Meteorological sensitive load data usually show high-frequency fluctuation characteristics, and the currently widely used single decomposition method fails to effectively reduce the high-frequency complexity in the data. This indicates that it is necessary to perform secondary decomposition of high-frequency subsequences to further refine and utilize the hidden information in the data, thereby improving the accuracy and reliability of the forecast. 2) Although the decomposition method can decompose the original data into multiple intrinsic mode functions (IMFs), this method usually leads to low forecasting efficiency when processing multiple components. Each component needs to be predicted separately, which increases the computational burden and reduces the overall forecasting efficiency. 3) Although many existing forecasting models have improved in accuracy, they usually lack sufficient interpretability. In power grid operation, the interpretability of forecasting models is extremely important, which can ensure the transparency of operations and the credibility of decisions. Summary of the invention

[0007] The purpose of the present invention is to provide a meteorological sensitive load prediction method, device, equipment and medium to solve the deficiencies of the prior art.

[0008] In a first aspect of the present application, a method for forecasting meteorological sensitive loads is provided, the method comprising:

[0009] Obtain the original sequence data of meteorological sensitive loads;

[0010] The CEEMDAN decomposition method is used to decompose the original sequence data to obtain multiple IMF components with different frequencies and scales;

[0011] Calculating sample entropies of multiple IMF components with different frequencies and scales, and aggregating the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence, and a low-frequency subsequence based on the sample entropy;

[0012] The first high-frequency subsequence is decomposed twice using the variational mode decomposition method to obtain a second high-frequency subsequence;

[0013] The second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence are respectively input into the LSTM neural network for prediction to obtain the prediction results of the second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence respectively, and the prediction results of the second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence are weighted and fused to obtain the prediction results of meteorological sensitive loads.

[0014] In one implementation, the CEEMDAN decomposition method is used to decompose the original sequence data to obtain multiple IMF components with different frequencies and scales, including:

[0015] Adding white noise multiple times to the original sequence data to generate multiple sequence data; adding white noise once corresponds to generating one sequence data;

[0016] Use EMD decomposition to decompose multiple sequence data until the residual does not exceed two extreme points, then stop EMD decomposition and obtain multiple IMF components with different frequencies and scales.

[0017] In one implementation, the calculation of sample entropy is defined as: Among them, A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability of matching when the two sequences increase in dimension to m+1, n represents the length, and r represents the acceptance threshold for the similarity of the two sequences.

[0018] In one implementation, a variational mode decomposition method is used to perform secondary decomposition on the first high-frequency subsequence to obtain a second high-frequency subsequence, including:

[0019] Construct a variational problem of the first high-frequency subsequence, and solve the variational problem to decompose the first high-frequency subsequence into K modal components;

[0020] Introduce a constraint that makes the sum of the K modal components equal to the first high-frequency subsequence;

[0021] Substitute the constraints into the objective function of the variational problem to construct the Lagrangian function;

[0022] The alternating direction multiplier method is used to solve the Lagerberg function until the convergence condition is reached and the second high frequency subsequence is obtained.

[0023] In one implementation, the method further includes: using the SHAP interpretable method to interpret the prediction results of the meteorological sensitive load and determine the influencing factors that affect the meteorological sensitive load.

[0024] In a second aspect of the present application, a weather-sensitive load prediction device is provided, the device comprising:

[0025] A data acquisition module is used to obtain the original sequence data of meteorological sensitive loads;

[0026] The first decomposition module is used to decompose the original sequence data using the CEEMDAN decomposition method to obtain multiple IMF components with different frequencies and scales;

[0027] An aggregation module is used to calculate the sample entropy of multiple IMF components with different frequencies and scales, and aggregate the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence and a low-frequency subsequence based on the sample entropy;

[0028] A second decomposition module is used to perform secondary decomposition on the first high-frequency subsequence by using a variational mode decomposition method to obtain a second high-frequency subsequence;

[0029] The prediction module is used to input the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence into the LSTM neural network for prediction, obtain the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence respectively, and weightedly fuse the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence to obtain the prediction results of the meteorological sensitive load.

[0030] In one implementation, the first decomposition module is specifically used to: add multiple white noises to the original sequence data to generate multiple sequence data; wherein adding white noise once corresponds to generating one sequence data; decompose the multiple sequence data using EMD decomposition until the residual does not exceed two extreme points, stop EMD decomposition, and obtain multiple IMF components with different frequencies and scales.

[0031] In one implementation, the calculation of sample entropy is defined as: Among them, A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability of matching when the two sequences increase in dimension to m+1, n represents the length, and r represents the acceptance threshold for the similarity of the two sequences.

[0032] The third aspect of the present application provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of a meteorological sensitive load forecasting method provided in the first aspect of the present application are implemented.

[0033] The fourth aspect of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a meteorological-sensitive load forecasting method provided in the first aspect of the present application are implemented.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. The present invention firstly decomposes the original sequence data of meteorological sensitive loads through CEEMDAN, then uses sample entropy (SE) to evaluate the sequence complexity of each modal component, and recombines these modal components into high-frequency, medium-frequency and low-frequency subsequences, which represent local features of different time scales. Then VMD is used to perform secondary decomposition on the high-frequency subsequence to reduce the complexity of the high-frequency subsequence and improve predictability. LSTM is used to predict the medium-frequency, low-frequency subsequences and the high-frequency subsequences after secondary decomposition, and finally the prediction results of each subsequence are integrated to obtain the final meteorological sensitive load prediction result.

[0036] 2. The present invention interprets the prediction results through the SHAP method, reveals the key influencing factors affecting the meteorological sensitive load, and enhances the credibility of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the embodiments of the present invention, constitute a part of this application, and do not constitute a limitation of the embodiments of the present invention. In the drawings:

[0038] Figure 1 A schematic diagram of a flow chart of a meteorological sensitive load prediction method provided in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of a meteorological sensitive load curve provided in an embodiment of the present application;

[0040] Figure 3 A schematic diagram of the CEEMDAN decomposition results provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of the sample entropy calculation results provided in an embodiment of the present application;

[0042] Figure 5 A schematic diagram of the sample entropy aggregation results provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of VMD secondary decomposition provided in an embodiment of the present application;

[0044] Figure 7 A comparison diagram of the prediction method provided in the embodiment of the present application and the prediction method provided in the prior art;

[0045] Figure 8 A schematic diagram for a global explanation of the features provided in the embodiments of the present application;

[0046] Fig. 9 A schematic diagram for local explanation of samples provided in the embodiments of the present application;

[0047] Fig.10 A principle block diagram of a meteorological-sensitive load prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0049] It should be noted that the terms "include" or "may include" used in various embodiments of the present application indicate the presence of the function, operation or element applied for, and do not limit the addition of one or more functions, operations or elements. In addition, as used in various embodiments of the present application, the terms "include", "have" and their cognates are only intended to indicate specific features, numbers, steps, operations, elements, components or a combination of the foregoing items, and should not be understood as first excluding the presence of one or more other features, numbers, steps, operations, elements, components or a combination of the foregoing items or the possibility of adding one or more features, numbers, steps, operations, elements, components or a combination of the foregoing items.

[0050] It should be understood that terms such as "first" and "second" are only used for descriptive purposes and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0051] Please refer to Figure 1 , Figure 1 A schematic diagram of a flow chart of a meteorological sensitive load prediction method provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes:

[0052] S101, obtaining original sequence data of meteorological sensitive loads.

[0053] In this embodiment, therefore, the load data of a community in Nanchong City, Sichuan Province in 2023 is selected with a time resolution of 15 minutes and the unit is MWh, in order to analyze the meteorological sensitive load of residents in this typical area under high temperatures in summer. Since the current deployment of smart meters has not yet been fully covered, it is still challenging to obtain users' meteorological sensitive load data separately. Power loads are generally divided into benchmark loads and temperature-sensitive loads, among which meteorological sensitive loads belong to temperature-sensitive loads and are significantly affected by temperature. Based on the load change trends over the years, Sichuan Province generally does not contain meteorological sensitive loads in March and April. Therefore, the load curves in March and April can be regarded as the benchmark loads, and the summer meteorological sensitive load curves can be obtained by subtracting the benchmark loads in March and April from the load curves in July and August. The meteorological sensitive load curve contains a total of 5952 original sequence data, such as Figure 2 shown.

[0054] S102, using CEEMDAN decomposition method to decompose the original sequence data to obtain multiple IMF components with different frequencies and scales.

[0055] In this embodiment, CEEMDAN decomposition is improved from EMD decomposition, and can adaptively decompose the original data into multiple IMF components of different frequencies and scales, which is suitable for nonlinear and non-stationary time series data. CEEMDAN adds an adaptive white noise sequence to the original data, and eliminates the modal aliasing phenomenon that may be introduced by EMD during decomposition by using integrated averaging, thereby enhancing the stability and reliability of the decomposition. In one embodiment, the steps for obtaining multiple IMF components with different frequencies and scales are as follows: adding multiple white noises to the original sequence data to generate multiple sequence data; wherein adding white noise once corresponds to generating one sequence data; using EMD decomposition to decompose multiple sequence data until the residual does not exceed two extreme points, stopping EMD decomposition, and obtaining multiple IMF components with different frequencies and scales.

[0056] The specific steps of CEEMDAN are as follows:

[0057] 1. Add white noise to the original sequence x(t) to generate new sequence data x i (t):

[0058] x i (t) = x(t) + ε 0 ω i (t),i=1,2,...n(1); where ε 0 is the signal-to-noise ratio; ω i (t) is the white noise sequence added for the ith time.

[0059] 2. For each x i(t) Use EMD to decompose and get the first modal component IMF 1 (t) and the first residual component r 1 (t): r 1 (t) = x(t) - IMF 1 (t)(3).

[0060] 3. Continue to add white noise to the original sequence x(t) to construct a new sequence to calculate the second modal component IMF 2 (t) and the second residual component r 2 (t): r 2 (t) = r 1 (t)-IMF 2 (t)(5).

[0061] 4. Repeat steps 1-3 above until the residual does not exceed two extreme points and cannot be decomposed any further:

[0062]

[0063] r k (t) = r k-1 (t)-IMF k (t)(7), where K is the total number of modal components after completing CEEMDAN.

[0064] After the decomposition is completed, the relationship between the original sequence, the modal component and the final residual component R(t) is:

[0065]

[0066] S103, calculating sample entropies of multiple IMF components with different frequencies and scales, and aggregating the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence, and a low-frequency subsequence based on the sample entropy.

[0067] In this embodiment, sample entropy is an improvement on approximate entropy, providing a more accurate and stable way to evaluate the complexity of time series data. Sample entropy measures the randomness and unpredictability of data by calculating the probability of occurrence of similar patterns in the data sequence. Compared with approximate entropy, sample entropy avoids the introduction of bias and the deficiency of relying on sample size, so it is more robust and applicable in practical applications. The larger the sample entropy, the higher the complexity of the time series data, the more irregular the changes, the lower the frequency of pattern occurrence, and the more unpredictable the sequence.

[0068] The calculation steps of sample entropy are as follows:

[0069] Construct a sequence of length m to be compared:

[0070] x m (i)=[x(i),x(i+1),...x(i+m-1)],i=1,2,...n-m+1 (9)

[0071] For every vector x m (i) and x m (j), calculate the distance d between vectors:

[0072] d[x m (i),x m (j)] = max[x m (i+k)-x m (j+k)],0≤k≤m-1 (10)

[0073] Calculate similarity probabilities:

[0074] where r is the acceptance threshold for considering two sequences similar; v m is d[x m (i),x m (j)]≤r; ω m+1 is d[x m+1 (i),x m+1 (j)]≤r; A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability that the two sequences match when the dimension is increased to m+1.

[0075] Finally, the calculation of sample entropy is defined as:

[0076] S104, performing secondary decomposition on the first high-frequency subsequence using a variational mode decomposition method to obtain a second high-frequency subsequence.

[0077] In this embodiment, VMD is a time-frequency decomposition method, the basic idea of ​​which is to decompose the signal into several modal components by constructing a variational model, each modal component having a specific center frequency. The goal of VMD is to find these modal functions so that the bandwidth of each modal function is minimized and each mode is adaptively decomposed. In one embodiment, the variational modal decomposition method is used to perform a secondary decomposition on the first high-frequency subsequence to obtain a second high-frequency subsequence, including:

[0078] Construct a variational problem of the first high-frequency subsequence, and solve the variational problem to decompose the first high-frequency subsequence into K modal components;

[0079] Introduce a constraint that makes the sum of the K modal components equal to the first high-frequency subsequence;

[0080] Substitute the constraints into the objective function of the variational problem to construct the Lagrangian function;

[0081] The alternating direction multiplier method is used to solve the Lagerberg function until the convergence condition is reached and the second high frequency subsequence is obtained.

[0082] Specifically, for a given sequence x(t), the specific steps of VMD are as follows:

[0083] Construct a variational problem, VMD decomposes x(t) into K modal components u by solving this variational problem k (t):

[0084]

[0085] in It means to find the partial derivative with respect to t; u k (t) is the kth modal component; ω k is the center frequency of the kth modal component; δ(t) is the Dirac function; * represents the convolution operation; j is the imaginary unit. The goal of this variational problem is to minimize the bandwidth of each modal function, that is, to minimize the spectral bandwidth of each modal function by adjusting the modal function and its corresponding center frequency.

[0086] To ensure that the sum of each mode is equal to the original sequence, the constraint condition is introduced:

[0087] Substitute the constraints into the objective function of the variational problem and construct the Lagrangian function:

[0088] where λ(t) is the Lagrange multiplier and α is the second-order penalty factor.

[0089] Solve the above optimization problem using the alternating direction multiplier method: Where τ is the tolerance to noise; yes The corresponding Fourier transform. Repeat the above steps to gradually optimize u k ,ω k and λ until the convergence condition is reached, the original sequence x(t) can be decomposed into K modal components, i.e., the second high-frequency subsequence.

[0090] S105, input the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence into the LSTM neural network for prediction respectively, obtain the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence respectively, weightedly fuse the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence respectively, and obtain the prediction result of the meteorological sensitive load.

[0091] In this embodiment, the LSTM neural network is a variant of the recurrent neural network (RNN), which can better capture long-term dependencies when processing time series data. Its structure is as follows: Figure 2 As shown in the figure, LSTM effectively solves the gradient vanishing and gradient exploding problems that are prone to occur in traditional RNN when processing long sequences by introducing memory units and gate mechanisms.

[0092] The LSTM neural network consists of multiple LSTM units, each of which mainly includes a forget gate, an input gate, and an output gate.

[0093] Forget gate: The forget gate f determines which information of the memory unit at the previous moment needs to be retained or forgotten.

[0094] f t =σ(W f ·[h t-1 ,x t ]+b f ) (18)

[0095]

[0096] Where σ(x) is the sigmoid activation function; W f is the weight matrix of the forget gate; b f is the bias vector of the forget gate.

[0097] Input gate: The input gate i controls how information is added to the memory cell, which includes two steps: the selection of updated information and the generation of new information.

[0098] Choices to update your information:

[0099] i t =σ(W i ·[h t-1 ,x t ]+b i ) (20)

[0100] Generate candidate information:

[0101]

[0102] Among them, tanh is the hyperbolic tangent function; is the candidate unit state, i.e., new information; Wc is the weight matrix of the unit state; b c is the bias vector of the cell state.

[0103] Unit status update: unit status c t It is updated through the forget gate and the input gate. First, the unit state c at the previous moment t-1 Multiply it by the output of the forget gate and add the candidate information selected by the input gate.

[0104]

[0105] Output gate: The output gate o determines the cell state c t Which part of will be used as output, and the hidden state h t It also originates from here.

[0106] o t =σ(W o ·[h t-1 ,x t ]+b o ) (24); h t =o t ·tanh(o t ) (25).

[0107] Where W o is the weight matrix of the output gate; b o is the bias vector for the output gate.

[0108] The original sequence data of meteorological sensitive loads were preliminarily decomposed using CEEMDAN. The decomposition results are shown in the following figure. Figure 3 As shown. Figure 3 It can be seen that the high-frequency subsequence after decomposition is highly complex and its change trend is not obvious; while the low-frequency subsequence is less complex and its overall change trend is more obvious.

[0109] In order to improve the efficiency of prediction, the sample entropy of the subsequences after CEEMDAN decomposition is calculated to evaluate the complexity of different subsequences. Subsequences with similar complexity are integrated to form high-frequency, medium-frequency, and low-frequency subsequences to represent local features at different time scales. Figure 4 The sample entropy calculation results under different parameters are shown, where IMF8 is the residual sequence. Figure 4It can be seen that no matter how the input parameters change when calculating the sample entropy, the sample entropy values ​​of IMF1 and IMF2 are much higher than other components, indicating that they have higher complexity and no obvious trend of change, and are difficult to predict subsequences. In contrast, the sample entropy values ​​of IMF7 and IMF8 are much lower than other components, indicating that they have lower complexity and obvious trend of change, and are more predictable. Therefore, these 8 IMF components are reintegrated into Co-IMF1 (including IMF1-2), Co-IMF2 (including IMF3-6) and Co-IMF3 (including IMF7-8), which represent the second high-frequency, medium-frequency and low-frequency subsequences after integration, respectively.

[0110] Depend on Figure 5 It can be seen that the high-frequency subsequence Co-IMF1 still has a high complexity after the initial decomposition by CEEMDAN and the integration of sample entropy. Therefore, this paper uses VMD decomposition to perform a secondary decomposition on Co-IMF1 to reduce the complexity of the high-frequency subsequence and improve predictability. The results of VMD secondary decomposition are shown in Figure 6 shown.

[0111] In order to verify the effectiveness of the prediction method proposed in the embodiment of the present application, the following models were selected for comparison:

[0112] (1) LSTM, without any decomposition, directly uses the original features for prediction.

[0113] (2) EMD-LSTM: Use EMD to decompose the original sequence, then use LSTM to predict each component separately, and obtain the final prediction result by summing them up.

[0114] (3) CEEMDAN-LSTM: CEEMDAN is used to decompose the original sequence, and then LSTM is used to predict each component separately, and the final prediction result is obtained by summing them up.

[0115] (4) VMD-LSTM: VMD is used to decompose the original sequence, and then LSTM is used to predict each component separately, and the final prediction result is obtained by summing them up.

[0116] (5) CEEMDAN-SE-LSTM: CEEMDAN is used to decompose the original sequence and then aggregate it according to the sample entropy to obtain high-frequency, medium-frequency, and low-frequency sub-sequences, which are then predicted using LSTM.

[0117] The prediction results are shown in Table 1. Figure 7 The forecast results for 24 randomly selected time steps are shown.

[0118] Table 1 Prediction results

[0119]

[0120] From Table 1 and Figure 7 It can be seen that the proposed CEEMDAN-SE-VMD-LSTM method achieves the highest prediction accuracy, which is 73.4940%, 74.1935%, 73.4180% and 14.4342% higher than the LSTM model in terms of MAE, RMSE, MMAPE and R2 respectively, and 38.0282%, 40.2597%, 37.7507% and 1.6410 higher than the optimal benchmark model VMD-LSTM respectively. It is proved that the proposed method makes full use of the characteristics of different frequency components through multi-level decomposition and aggregation strategies, reduces the complexity of the original meteorological sensitive load sequence, and effectively reduces the prediction error.

[0121] The LSTM model without decomposition method has the lowest prediction accuracy. The main reason is that the original sequence may contain multiple frequency components and noise, and direct modeling is difficult to effectively capture these complex features. In addition, the non-stationarity and high-frequency noise of the original data will interfere with the training of the model, resulting in large prediction deviations.

[0122] Among the prediction methods combined with decomposition, the prediction accuracy of CEEMDAN-SE-LSTM is the lowest. This is because although the input structure of the model is simplified after sample entropy aggregation, the high-frequency subsequences after aggregation increase the complexity, making prediction more difficult. In addition, the prediction accuracy of VMD-LSTM is better than that of CEEMDAN-LSTM and EMD-LSTM, indicating that the VMD method is more effective in signal decomposition and can more effectively separate and extract frequency components that are helpful for prediction.

[0123] This example uses CEEMDAN for preliminary decomposition and then combines it with VMD to perform secondary decomposition of high-frequency subsequences. This combination gives full play to the advantages of the two decomposition methods and avoids the excessive complexity and possible degradation of the effect caused by directly using two VMD decompositions. In addition, considering that using the same decomposition method twice may cause redundancy in signal processing, the combination of CEEMDAN and VMD effectively improves the prediction accuracy of the model.

[0124] In some embodiments, the method further includes: using the SHAP interpretable method to interpret the prediction results of the meteorological sensitive load to determine the influencing factors that affect the meteorological sensitive load.

[0125] Specifically, SHAP is a method for explaining the output of a model based on game theory. After the model prediction is completed, the explanation is provided by calculating the contribution of the features to the model prediction. This method can provide global and local explanations. SHAP is a model-independent explanation method that can be applied to any type of machine learning model.

[0126] SHAP is calculated based on the Shapley value, a solution for fair distribution of benefits in game theory. For machine learning models, features are considered "participants" and the model's prediction results are considered "benefits". The Shapley value gives the importance of each feature by calculating the marginal contribution of each feature to the prediction result. The calculation formula of SHAP value is as follows:

[0127] where φ i represents the SHAP value of the i-th feature; S is the feature subset that does not contain the i-th feature; N is the set of all features; f(S) is the output of the model trained using only the feature subset S. Each term in the formula represents the change in the model output before and after the i-th feature. By taking a weighted average of all possible feature subsets, the SHAP value can fairly distribute the contribution of each feature to the final prediction.

[0128] The relationship between the SHAP value and the model prediction results can be expressed as:

[0129] Where f(x) is the prediction function of the trained machine learning model; x is the input feature vector; φ 0 It is the benchmark output of the model, usually the mean of all sample target variables, which is a fixed value that represents the basic prediction level of the model without considering any features; Indicates the contribution of each feature to the model prediction result, that is, how each feature makes the prediction result from the benchmark output φ 0 An offset occurs.

[0130] The advantage of the SHAP explanation method is that it allows each step of the model prediction to be traced back to the specific feature contribution, providing strong support for the interpretability of the model. Users can understand how each feature affects the prediction results through the SHAP value, thereby gaining a deep understanding of the model decision-making process.

[0131] Global interpretability aims to understand the overall impact of each feature in the model on the model prediction output, rather than just the impact of a single prediction. Given that the subsequences obtained by the decomposition algorithm cannot fully represent the actual level of meteorological sensitive loads, in order to more directly analyze the impact of each feature on meteorological sensitive loads, this paper uses the LSTM model for direct prediction and uses the SHAP method to explain the model.

[0132] Figure 8 It shows the global interpretation of the features. Each row represents a feature, and the horizontal axis is the SHAP value, which indicates the contribution of the feature to the model output. The color indicates the size of the feature value, blue indicates a lower value, and red indicates a higher value. The distribution range of the points reflects the degree of influence of the feature on the prediction result. The wider the range, the greater the influence. The red points on the positive axis indicate that high feature values ​​have a positive impact on the prediction, that is, the higher the feature value, the higher the prediction result; the red points on the negative axis indicate that high feature values ​​have a negative impact on the prediction, that is, the higher the feature value, the lower the prediction result. Figure 8 It can be seen that the feature with the greatest impact on the meteorological sensitive load is the historical load. The SHAP value shows that when the historical load is high, it has a significant positive impact on the predicted value of the model, that is, the higher the historical load, the higher the predicted load value. On the contrary, a lower historical load will produce a lower predicted value. The second is temperature. High temperature has a significant positive impact on the model output, that is, the higher the temperature, the higher the predicted load value. On the contrary, low temperature has a negative impact on the model, and the predicted value will decrease accordingly. Humidity is the third most important feature. High humidity usually leads to an increase in the model predicted value, while low humidity may lead to a decrease in the predicted value. The impact of precipitation is small, and the SHAP value is mainly concentrated near zero. This may be because the precipitation is small in July and August, resulting in limited impact on the model output. Other features such as wind speed, wind direction, hour and time have a limited impact on the prediction of meteorological sensitive loads, and their SHAP values ​​are more concentrated, indicating that their contribution to the model prediction results is not significant.

[0133] Local interpretability aims to explain the results of a single prediction sample and clarify which features play a key role in the prediction of the sample. Fig. 9 It shows the local interpretability of a randomly selected prediction sample. Each row represents a feature and the specific value of the feature in this sample (normalized result). The bottom is the model's predicted expected value, which represents the average value of the model's output prediction without any feature input. The top is the model's predicted value for this sample, which is obtained by interacting with the contribution of each feature. Fig. 9 It can be seen that the historical load characteristics contribute +0.06 to the predicted value, the temperature characteristics contribute +0.02, and the humidity characteristics contribute -0.01.

[0134] Please refer to Fig.10 , Fig.10 A principle block diagram of a meteorological sensitive load prediction device provided in an embodiment of the present application is shown in FIG. Fig.10 As shown, the device comprises:

[0135] The data acquisition module 1010 is used to acquire the original sequence data of the meteorological sensitive load;

[0136] The first decomposition module 1020 is used to decompose the original sequence data using the CEEMDAN decomposition method to obtain multiple IMF components with different frequencies and scales;

[0137] Aggregation module 1030, used to calculate sample entropies of multiple IMF components with different frequencies and scales, and aggregate the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence, and a low-frequency subsequence based on the sample entropy;

[0138] The second decomposition module 1040 is used to perform secondary decomposition on the first high-frequency subsequence using a variational mode decomposition method to obtain a second high-frequency subsequence;

[0139] The prediction module 1050 is used to input the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence into the LSTM neural network for prediction, obtain the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence respectively, and weightedly fuse the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence to obtain the prediction result of the meteorological sensitive load.

[0140] A meteorological sensitive load prediction device in an embodiment of the present application is Figure 1 The meteorologically sensitive load prediction method shown is a technical solution based on the same inventive concept. Through the above detailed description of a meteorologically sensitive load prediction method, those skilled in the art can clearly understand the implementation process of a meteorologically sensitive load prediction device in this embodiment, so for the sake of brevity of the specification, it will not be repeated here.

[0141] Accordingly, a meteorological sensitive load prediction device provided by an embodiment of the present invention first performs a preliminary decomposition of the original sequence data of the meteorological sensitive load through CEEMDAN, then uses sample entropy (SE) to evaluate the sequence complexity of each modal component, and recombines these modal components into high-frequency, medium-frequency and low-frequency sub-sequences, respectively representing local features of different time scales. Then, VMD is used to perform secondary decomposition on the high-frequency sub-sequence to reduce the complexity of the high-frequency sub-sequence and improve predictability. LSTM is used to predict the medium-frequency, low-frequency sub-sequences and the high-frequency sub-sequences after secondary decomposition, and finally the prediction results of each sub-sequence are integrated to obtain the final meteorological sensitive load prediction result.

[0142] In some embodiments, the first decomposition module 1020 is specifically used to: add multiple white noises to the original sequence data to generate multiple sequence data; wherein adding white noise once corresponds to generating one sequence data; decomposing the multiple sequence data using EMD decomposition until the residual does not exceed two extreme points, stopping EMD decomposition, and obtaining multiple IMF components with different frequencies and scales.

[0143] In some embodiments, the calculation of sample entropy is defined as: Among them, A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability of matching when the two sequences increase in dimension to m+1, n represents the length, and r represents the acceptance threshold for the similarity of the two sequences.

[0144] In another embodiment of the present invention, an electronic device is provided, the electronic device comprising a processor, a memory, a communication interface and at least one communication bus for connecting the processor, the memory and the communication interface. The memory includes but is not limited to a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (PROM) or a portable CD-ROM, and the memory is used for related instructions and data.

[0145] The communication interface is used to receive and send data. The processor may be one or more CPUs. When the processor is a CPU, the CPU may be a single-core CPU or a multi-core CPU. The processor in the electronic device is used to read one or more programs stored in the memory and perform the following operations: obtain the original sequence data of the meteorological sensitive load; decompose the original sequence data using the CEEMDAN decomposition method to obtain multiple IMF components with different frequencies and scales; calculate the sample entropy of multiple IMF components with different frequencies and scales, and aggregate the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence, and a low-frequency subsequence based on the sample entropy; perform secondary decomposition of the first high-frequency subsequence using the variational mode decomposition method to obtain a second high-frequency subsequence; input the second high-frequency subsequence, the medium-frequency subsequence, and the low-frequency subsequence into the LSTM neural network for prediction, respectively, to obtain the prediction results of the second high-frequency subsequence, the medium-frequency subsequence, and the low-frequency subsequence, and weightedly fuse the prediction results of the second high-frequency subsequence, the medium-frequency subsequence, and the low-frequency subsequence to obtain the prediction results of the meteorological sensitive load.

[0146] It should be noted that the specific implementation of each operation can be Figure 1 The corresponding description of the method embodiment shown, the electronic device can be used to execute a meteorological sensitive load prediction method of the above method embodiment of the present application, which will not be described in detail here.

[0147] In another embodiment of the present invention, a computer-readable storage medium is provided, and the computer-readable storage medium is a memory device in a computer device for storing programs and data. It is understandable that the computer-readable storage medium here can include both built-in storage media in the computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, and the storage space stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of a meteorological sensitive load prediction method in the above embodiment. Those skilled in the art should understand that the embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0148] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for forecasting meteorological sensitive loads, characterized in that: Methods include: Obtain the original sequence data of meteorological sensitive loads; The CEEMDAN decomposition method is used to decompose the original sequence data to obtain multiple IMF components with different frequencies and scales; Calculating sample entropies of multiple IMF components with different frequencies and scales, and aggregating the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence, and a low-frequency subsequence based on the sample entropy; The first high-frequency subsequence is decomposed twice using the variational mode decomposition method to obtain a second high-frequency subsequence; The second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence are respectively input into the LSTM neural network for prediction to obtain the prediction results of the second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence respectively, and the prediction results of the second high-frequency subsequence, medium-frequency subsequence and low-frequency subsequence are weighted and fused to obtain the prediction results of meteorological sensitive loads.

2. The method according to claim 1, characterized in that The CEEMDAN decomposition method is used to decompose the original sequence data to obtain multiple IMF components with different frequencies and scales, including: Adding white noise multiple times to the original sequence data to generate multiple sequence data; adding white noise once corresponds to generating one sequence data; Use EMD decomposition to decompose multiple sequence data until the residual does not exceed two extreme points, then stop EMD decomposition and obtain multiple IMF components with different frequencies and scales.

3. The method according to claim 1, characterized in that The calculation of sample entropy is defined as: Among them, A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability of matching when the two sequences increase in dimension to m+1, n represents the length, and r represents the acceptance threshold for the similarity of the two sequences.

4. The method according to claim 1, characterized in that: The first high-frequency subsequence is decomposed twice using the variational mode decomposition method to obtain a second high-frequency subsequence, including: Construct a variational problem of the first high-frequency subsequence, and solve the variational problem to decompose the first high-frequency subsequence into K modal components; Introduce a constraint that makes the sum of the K modal components equal to the first high-frequency subsequence; Substitute the constraints into the objective function of the variational problem to construct the Lagrangian function; The alternating direction multiplier method is used to solve the Lagerberg function until the convergence condition is reached and the second high frequency subsequence is obtained.

5. The method according to claim 1, characterized in that The method further includes: using the SHAP interpretable method to interpret the prediction results of the meteorological sensitive load and determine the influencing factors that affect the meteorological sensitive load.

6. A meteorological sensitive load prediction device, characterized in that: The device includes: A data acquisition module is used to obtain the original sequence data of meteorological sensitive loads; The first decomposition module is used to decompose the original sequence data using the CEEMDAN decomposition method to obtain multiple IMF components with different frequencies and scales; An aggregation module is used to calculate the sample entropy of multiple IMF components with different frequencies and scales, and aggregate the multiple IMF components with different frequencies and scales into a first high-frequency subsequence, a medium-frequency subsequence and a low-frequency subsequence based on the sample entropy; A second decomposition module is used to perform secondary decomposition on the first high-frequency subsequence by using a variational mode decomposition method to obtain a second high-frequency subsequence; The prediction module is used to input the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence into the LSTM neural network for prediction, obtain the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence respectively, and weightedly fuse the prediction results of the second high-frequency subsequence, the medium-frequency subsequence and the low-frequency subsequence to obtain the prediction results of the meteorological sensitive load.

7. The device according to claim 6, characterized in that The first decomposition module is specifically used to: add multiple white noises to the original sequence data to generate multiple sequence data; wherein adding white noise once corresponds to generating one sequence data; decompose the multiple sequence data using EMD decomposition until the residual does not exceed two extreme points, stop EMD decomposition, and obtain multiple IMF components with different frequencies and scales.

8. The device according to claim 6, characterized in that The calculation of sample entropy is defined as: Among them, A m (r) is the probability that two sequences match in m dimensions; B m (r) is the probability of matching when the two sequences increase in dimension to m+1, n represents the length, and r represents the acceptance threshold for the similarity of the two sequences.

9. An electronic device, characterized in that: The electronic device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of a meteorological sensitive load prediction method as described in any one of claims 1 to 5 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of a meteorological-sensitive load prediction method according to any one of claims 1 to 5 are implemented.

Images