Short-term load prediction method and system based on VMD and BiLSTM

Abstract

The invention discloses a short-term load prediction method and system based on VMD and BiLSTM, and relates to the technical field of electric power, and the main points of the technical scheme are that the method comprises the steps: carrying out the preprocessing of original power load data, and obtaining initial data; performing optimization processing on parameters in the VMD by using an AWPSO algorithm, and decomposing the initial data through the optimized VMD to obtain a plurality of modal components; and adopting a BiLSTM model to predict each modal component, and adding the predicted values of each modal component to obtain a final predicted value of the original power load data. According to the method, modal decomposition is carried out on an original power load sequence, periodic features and trend features contained in the original power load sequence are mined, finally, short-term power load prediction is realized by utilizing BiLSTM, and prediction results of all modals are fused to reduce load fluctuation and reduce noise interference, so that the load prediction precision and stability are improved.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and more specifically, to a short-term load forecasting method and system based on VMD and BiLSTM. Background Art

[0002] As my country's economy and society gradually enter high-quality development, electricity, as the foundation of the national economy, has put forward higher requirements for power load and power quality in all walks of life. Accurate short-term power load forecasting can provide realistic load forecast curves for power generation planning and power dispatching, as well as provide decision-making for power purchase transactions in the power market. Short-term power load is affected by factors such as temperature, humidity, rainfall, economy, holidays, and intermittent access to new energy power plants, showing the characteristics of large volatility and strong noise, resulting in low accuracy and poor stability in short-term power load forecasting.

[0003] At present, short-term power load forecasting is divided into statistical methods, machine learning methods and neural network methods. Statistical method prediction includes autoregression analysis, exponential smoothing model, Kalman filter, multivariate linear regression, etc.; machine learning methods include random forest (RF), least squares support vector machine (LSSVM), etc.; neural network methods include recurrent neural network (RNN), long short-term memory network (LSTM), bidirectional long short-term memory network (BiLSTM) and gated recurrent unit (GRU), GA-VMD-LSTM combination method, etc.

[0004] However, statistical methods make predictions based on the time dependence and trend of historical data, but the prediction results will have large errors when applied to nonlinear and volatile load sequences; machine learning methods are prone to overfitting, poor stability, unsuitable for complex nonlinear data, and poor prediction accuracy; a single neural network method has high requirements for data set integrity when dealing with long sequences and complex dependencies, and has certain limitations in prediction accuracy. The combined model makes up for the shortcomings of a single model to a certain extent, but does not fully exploit the periodic and trend characteristics in the original power data.

[0005] Therefore, how to study and design a short-term load forecasting method and system based on VMD and BiLSTM that can overcome the above-mentioned defects is a problem that we urgently need to solve. Summary of the invention

[0006] In order to solve the deficiencies in the prior art, the purpose of the present invention is to provide a short-term load forecasting method and system based on VMD and BiLSTM, and to use AWPSO to optimize the modal decomposition parameter K and penalty factor in VMD. And the convergence criterion tolerance tol, and the original power load sequence is modally decomposed to explore the periodic characteristics and trend characteristics contained therein, and finally BiLSTM is used to realize short-term power load forecasting, and the prediction results of each mode are integrated to alleviate load fluctuations and reduce noise interference, thereby improving the load forecasting accuracy and stability.

[0007] The above technical objectives of the present invention are achieved through the following technical solutions:

[0008] First, a short-term load forecasting method based on VMD and BiLSTM is provided, comprising the following steps:

[0009] Preprocess the original power load data to obtain initial data;

[0010] The AWPSO algorithm is used to optimize the parameters in VMD, and the initial data is decomposed through the optimized VMD to obtain multiple modal components;

[0011] The BiLSTM model is used to predict each modal component, and the predicted values ​​of each modal component are added together to obtain the final predicted value of the original power load data.

[0012] Furthermore, the process of preprocessing the original power load data includes filling in missing values, data standardization and normalization processing.

[0013] Furthermore, the process of optimizing the parameters in VMD using the AWPSO algorithm is specifically as follows:

[0014] The particle swarm is randomly initialized, and each bird represents a candidate solution, whose position includes three dimensions: the number of modal decompositions, the penalty factor, and the tolerance of the convergence criterion;

[0015] The initialization speed of each bird is randomly initialized, and the fitness of each bird is calculated based on the signal reconstruction error;

[0016] Set the position in the bird flock as the initial optimal position of the bird and the global optimal position;

[0017] Adjust the speed and position of each bird;

[0018] Use the modal decomposition number, penalty factor and convergence criterion tolerance corresponding to the current particle position to perform VMD decomposition, update the modal components and the center frequencies of the modal components, and determine whether the conditions for stopping iteration of each modal component are met. If so, output the component, otherwise continue to iterate;

[0019] Calculate the fitness value and evaluate the decomposition effect of the current particle;

[0020] Compare the fitness value of the current particle with the historical optimal value: if the current fitness is better, update the individual optimal value; if the fitness of the current particle is better than the global optimal value of all particles, update the global optimal value;

[0021] Determine whether the maximum number of iterations has been reached or the change in the global optimal value is lower than the preset threshold. If not, continue to update the particle velocity and particle position. Otherwise, output the optimal combination of modal decomposition number, penalty factor and convergence criterion tolerance.

[0022] Furthermore, the speed adjustment expression for each bird is as follows:

[0023]

[0024] in, represents the velocity of the ith particle in the t+1th iteration in the dth dimension; w(t) represents the dynamic inertia weight of the tth generation; represents the speed of the ith particle in the tth iteration in the dth dimension; c1 represents the acceleration coefficient corresponding to each bird; c2 represents the acceleration coefficient corresponding to the flock of birds; r1 represents a random number in [0,1] corresponding to each bird; r2 represents a random number in [0,1] corresponding to the flock of birds; represents the optimal position found by the i-th particle in the d-th dimension; It represents the global optimal position found by all particles in the particle swarm in the dth dimension.

[0025] Furthermore, the position adjustment expression of each bird is as follows:

[0026]

[0027] in, represents the position of the ith particle at the t+1th iteration in the dth dimension; represents the position of the ith particle at the tth iteration in the dth dimension; represents the penalty factor; represents the velocity of the ith particle in the t+1th iteration in the dth dimension.

[0028] Furthermore, the method also includes:

[0029] Adjust the dynamic inertia weight in the AWPSO algorithm so that the dynamic inertia weight of the t+1th generation is smaller than that of the tth generation.

[0030] Furthermore, the expression of the dynamic inertia weight is specifically:

[0031]

[0032] Where w(t) represents the dynamic inertia weight of the tth generation; w max represents the initial inertia weight; w min Represents the final inertia weight; max_iter represents the maximum number of iterations.

[0033] In a second aspect, a short-term load forecasting system based on VMD and BiLSTM is provided, and the system is used to implement the short-term load forecasting method based on VMD and BiLSTM as described in any one of the first aspects, including:

[0034] A data processing module is used to pre-process the original power load data to obtain initial data;

[0035] The optimization decomposition module is used to optimize the parameters in VMD using the AWPSO algorithm, and decompose the initial data through the optimized VMD to obtain multiple modal components;

[0036] The data prediction module is used to predict each modal component using the BiLSTM model and add the predicted values ​​of each modal component to obtain the final predicted value of the original power load data.

[0037] In a third aspect, a computer terminal is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, a short-term load forecasting method based on VMD and BiLSTM as described in any one of the first aspects is implemented.

[0038] In a fourth aspect, a computer-readable medium is provided, on which a computer program is stored, and the computer program is executed by a processor to implement the short-term load forecasting method based on VMD and BiLSTM as described in any one of the first aspects.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] The short-term load forecasting method based on VMD and BiLSTM provided by the present invention optimizes VMD through the adaptive weight particle swarm optimization algorithm (Adaptive Weight Particle Swarm Optimization, AWPSO), and combines the model with BiLSTM, and uses AWPSO to optimize the modal decomposition parameter K and penalty factor in VMD. And the convergence criterion tolerance tol, and the original power load sequence is modally decomposed to explore the periodic characteristics and trend characteristics contained therein, and finally BiLSTM is used to realize short-term power load forecasting, and the prediction results of each mode are integrated to alleviate load fluctuations and reduce noise interference, thereby improving the load forecasting accuracy and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] 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:

[0042] Figure 1 is a flow chart of Embodiment 1 of the present invention;

[0043] Figure 2 It is a LSTM structure diagram in the prior art;

[0044] Figure 3 It is a BiLSTM structure diagram in the prior art;

[0045] Figure 4 is a structural diagram of VMD-AWPSO-BiLSTM in Example 1 of the present invention;

[0046] Figure 5 is a data processing flow chart in Embodiment 1 of the present invention;

[0047] Figure 6 is the original power load sequence diagram in Embodiment 2 of the present invention;

[0048] Figure 7 is a modal decomposition diagram of VMD-AWPSO in Example 2 of the present invention;

[0049] Figure 8 is a test curve diagram of an ablation experiment in Example 2 of the present invention;

[0050] Fig. 9 This is a graph of the test results of the optimal window of each model in Example 2 of the present invention for one week;

[0051] Fig.10 is the error curve of one week test of each model in Example 2 of the present invention;

[0052] Fig.11 is a box plot of the weekly load forecast error of each model in Example 2 of the present invention;

[0053] Fig.12 This is a graph of the optimal window one-day test results of each model in Example 2 of the present invention;

[0054] Fig.13 is the one-day test error curve of each model in Example 2 of the present invention;

[0055] Fig.14 is a box plot of the load forecast error of each model in one day in Example 2 of the present invention;

[0056] Fig.15 This is a system block diagram in Example 3 of the present invention. DETAILED DESCRIPTION

[0057] 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 implementation modes of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0058] Embodiment 1: A short-term load forecasting method based on VMD and BiLSTM is implemented by the following steps.

[0059] Step 1: Preprocess the original power load data to obtain initial data.

[0060] Preprocessing of the data set includes filling missing values, data standardization and normalization, so as to improve the accuracy of load forecasting.

[0061] Step 2: Use the AWPSO algorithm to optimize the parameters in VMD, and decompose the initial data through the optimized VMD to obtain multiple modal components.

[0062] Compared with Empirical Mode Decomposition (EMD), VMD has better noise immunity and overcomes problems such as modal component aliasing and endpoint effects. This technology is widely used in communication signal processing. It can achieve simultaneous decomposition and extraction of modes, and the number of decompositions can be manually selected. VMD is divided into two processes: construction and solution of variational problems. The variational problem construction formula is as follows:

[0063]

[0064] Where k (k = 1, 2, ..., K) is the modal decomposition number, u k are the K modal components obtained after decomposition; w k is the center frequency of each modal component; δ(t) is the unit impulse function; j is the imaginary unit, and f is the original signal.

[0065] After introducing the Lagrange multiplier and quadratic penalty factor, it is transformed into an unconstrained variational problem. The specific formula is as follows:

[0066]

[0067] Where α is the quadratic penalty factor and λ is the Lagrange multiplier.

[0068] The update iterative formulas for each modal component, modal center frequency and Lagrange multiplier are as follows:

[0069]

[0070] In the formula, γ is the noise.

[0071] The formula for stopping iteration condition for each modal component is as follows:

[0072]

[0073] In the formula, ε is a non-negative number that tends to 0.

[0074] The traditional VMD method has the problem that the relevant parameters are difficult to determine. Therefore, the present invention proposes to use AWPSO to optimize the parameters in VMD. AWPSO effectively enhances the global search and local search capabilities of the algorithm by dynamically adjusting the inertia weight, thereby improving the search efficiency and convergence speed. In a D-dimensional space, there are m birds forming a group, and the position of the i-th bird is represented by The best position for each bird to pass is The position with the minimum fitness of the entire flock is recorded as The speed of the i-th bird is The AWPSO iterative solution formula is as follows:

[0075]

[0076] Where i (i = 1, 2, ..., m) represents the value range of the i-th bird; d (d = 1, 2, ..., D) represents the value range of the d-th dimension; represents the velocity of the ith particle in the t+1th iteration in the dth dimension; w(t) represents the dynamic inertia weight of the tth generation; represents the velocity of the ith particle in the tth iteration in the dth dimension; w max represents the initial inertia weight, w min represents the final inertia weight, max_iter represents the maximum number of iterations, t represents the current number of iterations; c1 represents the acceleration coefficient corresponding to each bird; c2 represents the acceleration coefficient corresponding to the flock; r1 represents the random number in [0,1] corresponding to each bird; r2 represents the random number in [0,1] corresponding to the flock; represents the optimal position found so far by the i-th particle in the d-th dimension; represents the global optimal position found by all particles in the particle swarm in the dth dimension; represents the position of the ith particle at the t+1th iteration in the dth dimension; represents the position of the ith particle at the tth iteration in the dth dimension; Represents the penalty factor.

[0077] The process of adaptive weighted particle swarm optimization of variational modal decomposition model (VMD-AWPSO) is as follows Figure 1 The specific steps are as follows:

[0078] 1) Perform data preprocessing, complete the leak detection and supplement of the original input data, and input the original signal;

[0079] 2) Bird swarm initialization: Randomly initialize the particle swarm, each bird represents a candidate solution, and its position includes K, and tol three dimensions;

[0080] 3) Randomly initialize the speed of each bird and calculate the fitness of each bird based on the signal reconstruction error;

[0081] 4) Global and individual optimal initialization: Set the position of the bird in the flock to the initial optimal position of the bird P i and the global optimal position P g ;

[0082] 5) Use formula (7) and formula (9) to adjust the speed and position of each bird;

[0083] 6) Use K corresponding to the current particle position, Perform VMD decomposition with tol parameter, update modal component U and modal component center frequency W, and judge whether the iteration stop condition of each modal component is met. If so, output the component, otherwise continue iteration;

[0084] 7) Calculate the fitness value and evaluate the decomposition effect of the current particle;

[0085] 8) Compare the fitness value of the current particle with its historical optimal value. If the current fitness is better, update the individual optimal P i , if the fitness of the current particle is better than the global optimal value of all particles, then update the global optimal P g ;

[0086] 9) Determine whether the maximum number of iterations is reached or the change in the global optimal value is lower than the preset threshold. If not, continue to update the particle speed and particle position. Otherwise, output the optimal combination of K, α and tol;

[0087] 10) Use the optimal parameters to perform the final VMD decomposition on the original signal to obtain each modal component.

[0088] Step 3: Use the BiLSTM model to predict each modal component, and add the predicted values ​​of each modal component to obtain the final predicted value of the original power load data.

[0089] LSTM uses the logic control of the gating unit to decide whether to retain data, which solves the problem of large weight influence and easy gradient disappearance or gradient explosion in RNN. [ .like Figure 2 As shown in the figure, a single LSTM neuron consists of the following three key parts: input gate, forget gate and output gate. The input gate is responsible for controlling the received information, the forget gate determines which information should be retained, and the output gate determines the final output information. The state update formula of a single LSTM neuron is as follows:

[0090] f t =σ(W f h t-1 +U f x t +b f ) (10);

[0091] i t =σ(W i h t-1 +U i x t +b i ) (11);

[0092] o t =tanh(W o h t-1 +U o x t +b o ) (12);

[0093] C t =f t ⊙C t-1 +i t ⊙tanh(W c h t-1 +U c x t +b c ) (13);

[0094] h t =o t ⊙tanh(C t ) (14);

[0095] In the formula, f t represents the forget gate, i t represents the input gate, o t represents the output gate, C tis the cell state, σ is the sigmoid activation function, W f ,U f ,W i ,U i ,W o ,U o ,W c ,U c is the weight matrix, b f ,b i ,b o ,b c is the bias term, x t and h t-1 They represent the current input value and the previous output value, respectively. t is the output at the current moment, and ⊙ represents the convolution operation.

[0096] The unidirectional LSTM only considers the impact of the forward sequence on the backward sequence, but cannot handle the impact of the backward sequence on the forward sequence, making it difficult to effectively learn the relationship between the forward and backward sequence data. Figure 3 As shown, the present invention proposes to use BiLSTM to capture bidirectional sequence features and predict future information. The input layer inputs the time series, the LSTM layer processes the time series bidirectionally to obtain feature information, the splicing layer seamlessly merges the acquired feature information, connects the feature information of the previous and next sequences, and the output layer processes the spliced ​​feature information and outputs the final prediction result.

[0097] The specific structure of VMD-AWPSO-BiLSTM is as follows Figure 4 As shown in the figure, it is mainly divided into two parts: VMD-AWPSO and BiLSTM. The improved VMD-AWPSO is used to perform optimal modal decomposition on the preprocessed original power load sequence, and then different BiLSTMs are used to predict each characteristic modal component. Finally, the predicted values ​​of each characteristic modal component are fused to obtain the predicted value of the original power load sequence.

[0098] like Figure 5 As shown in the figure, first, a single load sequence is input, and then the improved VMD-AWPSO algorithm is used to decompose the sequence to obtain n modal components. Secondly, for each modal component, BiLSTM is used to perform load forecasting. Finally, the forecast results are integrated as the final forecast result.

[0099] Example 2: Experimental verification

[0100] (1) Power data preprocessing and VMD-AWPSO modal decomposition

[0101] like Figure 6As shown in the figure, the electricity consumption data set from Belgium is used for training and testing. The data includes 2015-1-1 to 2023-12-31, and the data collection interval is recorded every 15 minutes, in kilowatts (kW). The original power load sequence shows obvious periodicity, so the original power load sequence of 672 time points in a week is selected for the improved VMD-AWPSO modal decomposition test. Figure 7 As shown in the figure, the load value in the IMF1 modal component is large, and the IMF2 modal component shows a gentle fluctuation, which is highly consistent with the trend of the original power load sequence, accurately reflecting the user's power consumption characteristics and rules. The IMF3 to IMF5 modal components have a higher frequency and show strong randomness and volatility. VMD-AWPSO decomposes the original load to effectively reduce the volatility of the load, enhance the load characteristics, and explore its periodic characteristics and trend characteristics.

[0102] (2) Experimental test

[0103] The power data of seven consecutive days in 2015 were extracted from the data set as a weekly test set, and the power data of one day in 2023 were extracted as a one-day test set. The present invention was compared with LSSVM, GRU, BiLSTM, GA-VMD-LSTM, VMD-BiLSTM and other methods for a week and a day test. The number of epochs of training in the model was 100, the number of samples of a single pass batch_size was 128, the same optimizer Adam was used, and the learning rate was 0.001.

[0104] Ablation experiment: The data from 2015 in the dataset are used as the training set and test set for the ablation experiments of BiLSTM, AWPSO-BiLSTM, VMD-BiLSTM and VMD-AWPSO-BiLSTM. 96 data of one day are predicted. The prediction results are shown in Table 1. The improvement of AWPSO-BiLSTM in various indicators compared with BiLSTM is limited, while VMD-BiLSTM has a significant improvement in various indicators compared with BiLSTM, among which RMSE is improved by 62.41%, MAE is improved by 61.45%, MAPE is improved by 61.82%, and R 2 Compared with VMD-BiLSTM, VMD-AWPSO-BiLSTM has further improved various indicators, including RMSE increased by 31.34%, MAE increased by 28.62%, MAPE increased by 28.46%, and R 2 The analysis shows that VMD decomposition can effectively reduce the difficulty of load forecasting and greatly improve the forecasting accuracy. Parameters such as and tol can reduce the loss of VMD decomposition and further enhance the prediction ability of the model. Figure 8 This is the ablation experiment test curve diagram. The prediction curves of BiLSTM and AWPSO-BiLSTM deviate greatly from the true value. The prediction result of VMD-BiLSTM model is better, but VMD-AWPSO-BiLSTM has the highest prediction accuracy.

[0105] Table 1 Ablation experiment results

[0106]

[0107] One-week test: When performing time series prediction, the sliding window step size is a key parameter. A larger sliding window step size will lead to information loss and reduce the complexity of the model; while a smaller sliding window step size will increase the overlap between samples, increase the complexity of the model and the computational cost. Therefore, predictions of various models under different sliding windows were performed. The results are shown in Table 2. It can be seen that the optimal window prediction results of each model in the one-week test, and it is found that the various evaluation indicators of the method proposed in the one-week test are better than other models. Compared with the optimal values ​​of other models, RMSE increased by 73.13%, MAE increased by 71.07%, MAPE increased by 71.07%, R 2 Improved by 0.010.

[0108] Fig. 9 The following is a graph showing the test results of the optimal window of each model for one week. The load curve shows obvious periodicity. The lowest point of electricity consumption is 20:30 every day, and the peak electricity consumption occurs at 5:30 and 13:00. For the peaks and troughs that are more difficult to predict, an area is randomly selected and zoomed in. It can be seen that the prediction curve of LSSVM deviates the farthest from the true value. The prediction curves of GRU, BiLSTM and GA-VMD-LSTM have many prediction points that deviate far from the true value. VMD-BiLSTM and VMD-AWPSO-BiLSTM perform well and can effectively fit the true value curve, among which VMD-AWPSO-BiLSTM has a better fitting effect.

[0109] Table 2 Results of testing each model in different windows in one week

[0110]

[0111]

[0112] Fig.10The error curves of each model in a week of testing show that the error curves of LSSVM, GRU and BiLSTM, VMD-BiLSTM and GA-VMD-LSTM fluctuate greatly and have strong randomness, indicating that they are not stable enough. The error curve of VMD-AWPSO-BiLSTM fluctuates steadily and is closer to 0. In summary, VMD-AWPSO-BiLSTM has better prediction accuracy and stability.

[0113] Fig.11 The box plot of the weekly load forecast error of each model reflects the error distribution of each test point and the true value. The height of the box represents the discrete degree of the forecast error of each model. It can be found that the box heights of LSSVM, GRU, BiLSTM, GA-VMD-LSTM and VMD-BiLSTM are similar, indicating that the discreteness of their error value distribution is similar, while the box height of VMD-AWPSO-BiLSTM is much smaller than that of other models, indicating that the discreteness of its error value distribution is the smallest, indicating that the model has good stability. After analyzing the distribution of outliers of each model, it is found that the outlier value of the VMD-AWPSO-BiLSTM model is the smallest and closest to the 0 value line, showing high stability and prediction accuracy. The median line reflects the central trend of the data. The closer the median line is to the 0 value line, the higher the prediction accuracy. It can be observed that the median line of the VMD-AWPSO-BiLSTM model almost coincides with the 0 value line, indicating that its error value is small and close to the 0 value, further showing its high-precision prediction ability.

[0114] One-day test: Table 3 shows the test results of each model in different windows in one day. Analysis of Table 3 shows that the evaluation indicators of the proposed method in one-day test are better than those of other models. Compared with the optimal values ​​of other models, RMSE is improved by 45.21%, MAE is improved by 44.28%, MAPE is improved by 40.21%, and R 2 Improved by 0.001.

[0115] Table 3 Test results of different models in different windows in one day

[0116]

[0117]

[0118] Fig.12 The following are the test results of the optimal window for each model for one day. It can be seen that the prediction curves of LSSVM are mostly far away from the true value, while the prediction curves of other models are relatively close to the true value curve. Among them, the prediction curve of VMD-AWPSO-BiLSTM is most close to the true value curve, indicating that it has better prediction accuracy.

[0119] Fig.13The error curves of each model were tested for one day, and it was found that the error curve of LSSVM was almost completely far away from the value of 0. The error curve of VMD-AWPSO-BiLSTM was closer to the value of 0 at all points compared with the error curves of GRU, BiLSTM, VMD-BiLSTM, and GA-VMD-LSTM, indicating that VMD-AWPSO-BiLSTM has better prediction accuracy and stability.

[0120] Fig.14 The following is a box plot of the load forecast error for each model for one day. By observing the box height in the figure, it is found that the LSSVM box is the highest, followed by GRU, BiLSTM, VMD-BiLSTM, and GA-VMD-LSTM. The shortest box is VMD-AWPSO-BiLSTM, indicating that the VMD-AWAWPSO-BiLSTM has the smallest error value distribution dispersion and the more stable the model is. At the same time, it is found that the median lines of GA-VMD-LSTM, BiLSTM, VMD-BiLSTM and VMD-AWAWPSO-BiLSTM almost coincide with the 0 value line, indicating that the above four models have higher prediction accuracy than the other two models. Combining the box height and median line analysis, VMD-AWPSO-BiLSTM shows the characteristics of high stability and high prediction accuracy.

[0121] In summary, decomposing the load sequence by VMD-AWPSO can effectively reduce the volatility in the load and make the load characteristics more regular. The integrated BiLSTM model is used to predict each modal component in a targeted manner, which can reduce the impact of noise on short-term load forecasting and improve the stability of the model and the prediction accuracy.

[0122] Example 3: Short-term load forecasting system based on VMD and BiLSTM: This system is used to implement the short-term load forecasting method based on VMD and BiLSTM as described in Example 1, such as Fig.15 As shown, it includes a data processing module, an optimization decomposition module and a data prediction module.

[0123] Among them, the data processing module is used to preprocess the original power load data to obtain the initial data; the optimization decomposition module is used to optimize the parameters in the VMD using the AWPSO algorithm, and decompose the initial data through the optimized VMD to obtain multiple modal components; the data prediction module is used to predict each modal component using the BiLSTM model, and add the predicted values ​​of each modal component to obtain the final predicted value of the original power load data.

[0124] The present invention also records a computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the short-term load forecasting method based on VMD and BiLSTM as described in Example 1 is implemented.

[0125] The present invention also records a computer-readable medium on which a computer program is stored. The computer program is executed by a processor to implement the short-term load forecasting method based on VMD and BiLSTM as described in Example 1.

[0126] Working principle: The power load data is affected by electricity prices, temperature and other meteorological factors, and the power load data is highly nonlinear and noisy and difficult to accurately predict. The existing related technical methods lack in-depth mining of the original power load data. This paper optimizes the combined model of VMD and BiLSTM through the Adaptive Weight Particle Swarm Optimization (AWPSO) algorithm, and uses AWPSO to optimize the modal decomposition parameter K and penalty factor in VMD. And the convergence criterion tolerance tol, and the original power load sequence is modally decomposed to explore the periodic characteristics and trend characteristics contained therein, and finally BiLSTM is used to realize short-term power load forecasting, and the prediction results of each mode are integrated to alleviate load fluctuations and reduce noise interference, thereby improving the load forecasting accuracy and stability.

[0127] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0128] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0131] 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. The short-term load forecasting method based on VMD and BiLSTM is characterized by: The following steps are involved: Preprocess the original power load data to obtain initial data; The AWPSO algorithm is used to optimize the parameters in VMD, and the initial data is decomposed through the optimized VMD to obtain multiple modal components; The BiLSTM model is used to predict each modal component, and the predicted values ​​of each modal component are added together to obtain the final predicted value of the original power load data.

2. The short-term load forecasting method based on VMD and BiLSTM according to claim 1 is characterized in that: The process of preprocessing the original power load data includes filling in missing values, data standardization and normalization processing.

3. The short-term load forecasting method based on VMD and BiLSTM according to claim 1 is characterized in that: The process of optimizing the parameters in VMD using the AWPSO algorithm is specifically as follows: The particle swarm is randomly initialized, and each bird represents a candidate solution, whose position includes three dimensions: the number of modal decompositions, the penalty factor, and the tolerance of the convergence criterion; The initialization speed of each bird is randomly initialized, and the fitness of each bird is calculated based on the signal reconstruction error; Set the position in the bird flock as the initial optimal position of the bird and the global optimal position; Adjust the speed and position of each bird; Use the modal decomposition number, penalty factor and convergence criterion tolerance corresponding to the current particle position to perform VMD decomposition, update the modal components and the center frequencies of the modal components, and determine whether the conditions for stopping iteration of each modal component are met. If so, output the component, otherwise continue to iterate; Calculate the fitness value and evaluate the decomposition effect of the current particle; Compare the fitness value of the current particle with the historical optimal value: if the current fitness is better, update the individual optimal value; if the fitness of the current particle is better than the global optimal value of all particles, update the global optimal value; Determine whether the maximum number of iterations has been reached or the change in the global optimal value is lower than the preset threshold. If not, continue to update the particle velocity and particle position. Otherwise, output the optimal combination of modal decomposition number, penalty factor and convergence criterion tolerance.

4. The short-term load forecasting method based on VMD and BiLSTM according to claim 3 is characterized in that: The speed adjustment expression for each bird is as follows: in, represents the velocity of the ith particle in the t+1th iteration in the dth dimension; w(t) represents the dynamic inertia weight of the tth generation; represents the speed of the ith particle in the tth iteration in the dth dimension; c1 represents the acceleration coefficient corresponding to each bird; c2 represents the acceleration coefficient corresponding to the flock of birds; r1 represents a random number in [0,1] corresponding to each bird; r2 represents a random number in [0,1] corresponding to the flock of birds; represents the optimal position found by the i-th particle in the d-th dimension; It represents the global optimal position found by all particles in the particle swarm in the dth dimension.

5. The short-term load forecasting method based on VMD and BiLSTM according to claim 3 is characterized in that: The specific expression for adjusting the position of each bird is: in, represents the position of the ith particle at the t+1th iteration in the dth dimension; represents the position of the ith particle at the tth iteration in the dth dimension; represents the penalty factor; represents the velocity of the ith particle in the t+1th iteration in the dth dimension.

6. The short-term load forecasting method based on VMD and BiLSTM according to claim 1 is characterized in that: The method further includes: Adjust the dynamic inertia weight in the AWPSO algorithm so that the dynamic inertia weight of the t+1th generation is smaller than that of the tth generation.

7. The short-term load forecasting method based on VMD and BiLSTM according to claim 6 is characterized in that: The expression of the dynamic inertia weight is specifically: Where w(t) represents the dynamic inertia weight of the tth generation; w max represents the initial inertia weight; w min Represents the final inertia weight; max_iter represents the maximum number of iterations.

8. The short-term load forecasting system based on VMD and BiLSTM is characterized by: The system is used to implement the short-term load forecasting method based on VMD and BiLSTM as described in any one of claims 1 to 7, including: A data processing module is used to pre-process the original power load data to obtain initial data; The optimization decomposition module is used to optimize the parameters in VMD using the AWPSO algorithm, and decompose the initial data through the optimized VMD to obtain multiple modal components; The data prediction module is used to predict each modal component using the BiLSTM model and add the predicted values ​​of each modal component to obtain the final predicted value of the original power load data.

9. A computer terminal comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the short-term load forecasting method based on VMD and BiLSTM as described in any one of claims 1 to 7 is implemented.

10. A computer readable medium having a computer program stored thereon, characterized in that: The computer program is executed by a processor to implement the short-term load forecasting method based on VMD and BiLSTM as described in any one of claims 1 to 7.

Images