Short-term electricity price prediction method, device and equipment

Abstract

The invention provides a short-term electricity price prediction method, device and equipment, and relates to the technical field of electric power. The method comprises the following steps: acquiring historical electricity price data and related index data influencing the electricity price; performing complete adaptive noise set empirical mode decomposition on the historical electricity price data to obtain a plurality of mode components; calculating the sample entropy of each modal component, and dividing and reconstructing each modal component into a periodic component and a nonlinear component; inputting the nonlinear component and the related index data into a composite prediction model formed by a convolutional neural network and a bidirectional long-short-term memory network to obtain a prediction result of the nonlinear component; inputting the periodic component into the time sequence prediction model to obtain a prediction result of the periodic component; and reconstructing prediction results of the nonlinear component and the periodic component to obtain a predicted electricity price. According to the invention, the problems of low precision, poor model robustness and difficulty in effectively processing complex electricity price data of an electricity price prediction method in the prior art can be solved.

Description

Technical Field

[0001] The present invention relates to the field of electric power technology, and in particular to a short-term electricity price forecasting method, device and equipment. Background Art

[0002] As the proportion of new energy power generation such as wind power and photovoltaic power generation continues to increase, the strong volatility of wind power and photovoltaic power generation makes the traditional power industry structure unfavorable for the absorption of new energy power generation.

[0003] Improving the flexibility of power system operations through market transactions can create a better environment for absorbing wind and photovoltaic power. This means that the power market is no longer dominated by a single supplier, allowing power consumers to more freely choose their power suppliers and purchasing methods. As the link between supply and demand in the power market, electricity prices have a strong regulatory capacity and are the most important factor in the power market. However, due to the combined influence of multiple factors, electricity prices fluctuate significantly and are highly unstable. Therefore, accurate electricity price forecasts can improve the efficiency and cost-effectiveness of resource allocation, avoid supply and demand imbalances and energy waste, and are crucial for achieving a balanced supply and demand of electricity and a balanced allocation of power market resources.

[0004] However, most current research on electricity price forecasting is based on statistical models, machine learning, and other methods. For example, in the field of statistics, the current mainstream method is to use the differential integrated autoregressive moving average model (ARIMA) to predict electricity prices. However, this method is effective for predicting stable electricity prices, but has poor prediction effects on non-stationary, highly random, and nonlinear electricity price objects, and cannot meet current prediction accuracy requirements. Support vector machine (SVM) is a commonly used method in the field of machine learning for short-term electricity price forecasting. This method can obtain the mapping relationship between output data and input data to a certain extent, but it is difficult to explore the laws of features in nonlinear time series, and the accuracy of short-term electricity price forecasting is still insufficient. Summary of the Invention

[0005] In view of this, the embodiments of the present invention provide a short-term electricity price forecasting method, apparatus and device to solve the problems of low accuracy, poor model robustness and difficulty in effectively processing complex electricity price data in the existing electricity price forecasting methods.

[0006] In a first aspect, an embodiment of the present invention provides a short-term electricity price forecasting method, comprising:

[0007] Obtain historical electricity price data and relevant indicator data affecting electricity prices;

[0008] Performing fully adaptive noise ensemble empirical mode decomposition on the historical electricity price data to obtain multiple modal components;

[0009] Calculate the sample entropy of each modal component, and divide and reconstruct each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component;

[0010] Inputting the nonlinear component and the related index data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component;

[0011] Inputting the periodic component into a time series prediction model to obtain a prediction result of the periodic component;

[0012] The prediction results of the nonlinear component and the periodic component are reconstructed to obtain the predicted electricity price.

[0013] As a possible implementation manner, performing fully adaptive noise set empirical mode decomposition on the historical electricity price data includes:

[0014] Establishing a fitness function by decomposing the historical electricity price data using a fully adaptive noise set empirical mode decomposition algorithm and taking the error between the reconstructed data and the historical electricity price data as a target;

[0015] Optimizing noise parameters of the fully adaptive noise ensemble empirical mode decomposition algorithm by a genetic algorithm according to the fitness function;

[0016] The optimized fully adaptive noise ensemble empirical mode decomposition algorithm is used to perform fully adaptive noise ensemble empirical mode decomposition on the historical electricity price data.

[0017] As a possible implementation manner, the noise parameter includes at least one of the following: noise amplitude and number of integrated averaging times.

[0018] As a possible implementation, the division and reconstruction of each modal component into a periodic component and a nonlinear component according to the sample entropy of each modal component includes:

[0019] Determine a modal component whose sample entropy is less than a preset sample entropy threshold as a first modal component, and determine a modal component whose sample entropy is greater than or equal to the sample entropy threshold as a second modal component;

[0020] reconstructing the first modal component to obtain the periodic component;

[0021] The second modal component is reconstructed to obtain the nonlinear component.

[0022] As a possible implementation method, the nonlinear component and the related indicator data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component, including:

[0023] Inputting the nonlinear component and the relevant index data into the convolutional neural network to obtain a prediction result output by the convolutional neural network;

[0024] The prediction result output by the convolutional neural network is input into the bidirectional long short-term memory network to obtain the prediction result of the nonlinear component.

[0025] As a possible implementation, before inputting the nonlinear component and the related indicator data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component, the method further includes:

[0026] Performing an optimization operation on the network hyperparameters of the bidirectional long short-term memory network using the White Whale algorithm to obtain the optimal network hyperparameters of the bidirectional long short-term memory network;

[0027] The composite prediction model is trained based on the optimal network hyperparameters.

[0028] As a possible implementation, the time series prediction model is a Prophet model;

[0029] Before inputting the periodic component into the time series prediction model to obtain the prediction result of the periodic component, the method further includes: training the Prophet model.

[0030] As a possible implementation manner, the relevant indicator data includes at least one of the following: power generation cost and new energy consumption rate.

[0031] In a second aspect, an embodiment of the present invention provides a short-term electricity price prediction device, comprising:

[0032] The acquisition module is used to obtain historical electricity price data and related indicator data that affect electricity prices;

[0033] a decomposition module, configured to perform fully adaptive noise set empirical mode decomposition on the historical electricity price data to obtain multiple modal components;

[0034] A processing module is used to calculate the sample entropy of each modal component and divide and reconstruct each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component;

[0035] The prediction module is used to input the nonlinear component and the related indicator data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component; input the periodic component into a time series prediction model to obtain a prediction result of the periodic component; and reconstruct the prediction results of the nonlinear component and the periodic component to obtain a predicted electricity price.

[0036] In a third aspect, an embodiment of the present invention provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method in the first aspect or any possible implementation of the first aspect is implemented.

[0037] In an embodiment of the present invention, historical electricity price data is subjected to fully adaptive noise set empirical mode decomposition into multiple modal components to improve overall prediction accuracy. The sample entropy of each modal component is calculated, and each modal component is divided and reconstructed into a periodic component and a nonlinear component based on the size of the sample entropy of each modal component. That is, the original electricity price data is decoded and reconstructed into two types of data with significant eigenvalues, and predictions are made separately to improve the reliability of the prediction. Furthermore, the nonlinear component and related indicator data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain the prediction result of the nonlinear component. The periodic component is input into a time series prediction model to obtain the prediction result of the periodic component. The prediction results of the nonlinear component and the periodic component are reconstructed to obtain the predicted electricity price. The multi-level model combination can significantly improve the accuracy and robustness of short-term electricity price prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of the short-term electricity price forecasting method provided by the embodiment of the present invention. Figure 1 ;

[0039] Figure 2 This is a flow chart of the short-term electricity price forecasting method provided by the embodiment of the present invention. Figure 2 ;

[0040] Figure 3 1 is a schematic structural diagram of a short-term electricity price prediction device provided by an embodiment of the present invention;

[0041] Figure 4 It is a structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0042] In the following description, specific details such as particular system structures and techniques are provided for purposes of illustration, not limitation, to facilitate a thorough understanding of the embodiments of the present invention. However, it will be apparent to those skilled in the art that the present invention may be practiced in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

[0043] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0044] Compared with a single deep learning method, a combined deep learning method that combines multiple modules has better versatility and accuracy in prediction. Based on the above integrated idea, an embodiment of the present invention proposes a method that combines the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) algorithm, sample entropy, convolutional neural network (CNN) and bidirectional long short-term memory network (BiLSTM), and time series prediction model (Prophet) to perform short-term electricity price forecasting to improve prediction accuracy and robustness.

[0045] See also Figure 1 , which shows a flow chart for implementing the short-term electricity price prediction method provided by an embodiment of the present invention, as detailed below:

[0046] Step S101: Acquire historical electricity price data and relevant indicator data affecting electricity prices.

[0047] The historical electricity price data is the time series of electricity prices within a preset time period before the forecast period. The historical electricity price data HP can be expressed as HP t =[HP1,HP2,…,HP T ].

[0048] Relevant indicator data that affect electricity prices may include power generation cost GC, new energy consumption rate NR, etc.

[0049] The cost of power generation can be expressed as: GC t =[GC1,GC2,…,GC T ].

[0050] The new energy consumption rate can be expressed as: NR t =[NR1,NR2,…,NR T ].

[0051] After obtaining these data, they need to be pre-processed before use. For example, data standardization:

[0052]

[0053] Among them, μ is the mean of each data, and σ is the standard deviation of each data.

[0054] Step S102 : performing fully adaptive noise ensemble empirical mode decomposition on the historical electricity price data to obtain multiple modal components.

[0055] To effectively process the complex and highly variable raw sample input data, this embodiment uses the ICEEMDAN algorithm to decompose historical electricity price data to predict short-term electricity prices. The ICEEMDAN algorithm is an improvement on the CEEMDAN algorithm. While the CEEMDAN algorithm reduces the impact of noise by adding Gaussian white noise, noise still exists in each component, and each component will have spurious modes. Based on this, the ICEEMDAN method does not directly utilize Gaussian noise. Instead, it extracts the jth-order mode using averaged data to minimize the impact of white noise on modal decomposition.

[0056] Defining the IMF j ″ is the jth modal component generated by decomposition using the ICEEMDAN algorithm; E j (·) is the operator that generates the jth modal component using the empirical mode decomposition algorithm; W i (t) is the Gaussian white noise added for the i-th time, with a mean of 0 and a variance of 1; t is a time variable; ε j-1 In the calculation of the jth modal component IMF″ j The adaptive coefficient is obtained when , and M(·) represents the expected value generated by the local data. The detailed implementation process of the ICEEMDAN algorithm is as follows (Note: The description of the ICEEMDAN algorithm process below is for understanding this solution. Since the ICEEMDAN algorithm is a prior art and is not the focus of this embodiment, any details not detailed in the process can be referred to existing materials):

[0057] (1) Construct and use the empirical mode decomposition algorithm to calculate the first residual data as:

[0058] X i (t) = X(t) + ε0E1(W i (t));

[0059] (2) Calculate the first-order modal component, which can be expressed as:

[0060] r1(t)=MX i (t);

[0061] (3) Estimate the second-order residual data as part r1(t)+ε1E2W i (t), so the second-order modal component can be expressed as:

[0062] IMF″2(t)=r1(t)-r2(t)=r1(t)-M(r1(t)+ε1E2W i (t));

[0063] (4) Recalculate steps (2) and (3) above. At this point, the j-th residual and the j-th order component can finally be expressed as:

[0064] r j (t) = M(r j-1 (t)+ε j-1 E j W i (t));

[0065] IMF j (t) = r j-1 (t)-r j (t).

[0066] Step S103 : calculating the sample entropy of each modal component, and dividing and reconstructing each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component.

[0067] After the electricity price data is decomposed, it will bring about the problem of information diffusion. Since there are many factors that affect the electricity price and the input range of the original data is relatively wide, this embodiment adopts the sample entropy method to classify and integrate components with similar complexity to improve data processing efficiency. Each modal component is a subsequence, and the larger the entropy value, the greater the complexity of the subsequence. For a time series composed of N data, the steps of the sample entropy algorithm are roughly as follows (Note: The description of the following sample entropy algorithm process is used to understand this solution. Since the sample entropy algorithm is a prior art and is not the focus of this embodiment, the incomplete content in the process can refer to existing materials):

[0068] First, a set of m-dimensional vectors X are formed in time order m (i), the vector represents the m consecutive x values ​​starting from the i-th point. m (i) With X m The maximum absolute value of the difference between the corresponding elements in (j) is the distance between the two vectors, which can be expressed as follows:

[0069] d[X m (i),X m (j)]=max g＝0,…,m-1 (|x(i+g)-x(j+g)|);

[0070] Secondly, X m (i) With X m The number of (j) whose spacing is less than or equal to the similarity tolerance r is recorded as B i , definition B (m) (r) is X m (i) With X m (j) The probability of matching m points under r is:

[0071]

[0072] Again, increase the dimension to m+1 and change X m (i) With X m The number of intervals (j) less than or equal to r is denoted as A i Definition A (m) (r) is the probability of matching m+1 points under two sequence vectors r, and the formula is as follows:

[0073]

[0074] Finally, the sample entropy can be defined as:

[0075]

[0076] The calculation of sample entropy measures overall complexity by comparing the similarities between subsequences. When multiple modal components of the data have similar complexity, the corresponding subsequences will exhibit higher similarity. In this way, sample entropy can effectively integrate components of similar complexity and, by globally measuring the complexity of the data, reflect the overall characteristics of the similarly complex parts of the data. That is, the sample entropy of a modal component reflects its complexity; modal components with similar sample entropies have similar complexities. Moreover, after the sample entropy calculation, the sample entropies of different time series often have different characteristics. Sequences with periodic characteristics often exhibit repetitive patterns and regularity, so the sample entropy calculated for time series with periodic characteristics is generally lower. Sequences with nonlinear characteristics are irregular and unpredictable. Their changes do not follow simple patterns and often involve complex dynamic processes. Therefore, the sample entropy calculated for sequences with nonlinear characteristics is higher. Therefore, by setting a sample entropy threshold, it is possible to distinguish between periodic and nonlinear characteristics of modal components:

[0077] A modal component whose sample entropy is less than a preset sample entropy threshold is determined as a first modal component, and a modal component whose sample entropy is greater than or equal to the sample entropy threshold is determined as a second modal component.

[0078] Furthermore, the first modal component is reconstructed by superposition to obtain a periodic component, that is, the first modal component is integrated to form a new sequence with periodic characteristics. Similarly, the second modal component is reconstructed by superposition to obtain a nonlinear component, that is, the second modal component is integrated to form a new sequence with nonlinear characteristics.

[0079] In step S104, the nonlinear component and related index data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component.

[0080] RNNs are neural networks specifically designed for processing sequential data. By introducing recurrent connections, RNNs enable the network to retain and utilize historical information, resulting in excellent performance when processing time series data. This capability has led to widespread application of RNNs in fields such as natural language processing, speech recognition, and time series forecasting. RNNs overcome the limitations of traditional neural networks by introducing recurrent connections within each neuron. In an RNN, each neuron receives input not only at the current time step but also from the hidden state of the previous time step. This mechanism enables the RNN to retain previous information while processing the current input, effectively capturing the temporal dependencies of data.

[0081] In this embodiment, the RNN is primarily used to process time series data for electricity prices. This embodiment can input preprocessed data (including the nonlinear components of historical electricity price data, generation costs, energy consumption rates, etc.) into the RNN network architecture. The RNN network gradually passes this input data to hidden states and uses recurrent connections to capture long-term dependencies in the time series. Ultimately, it outputs hidden states containing time series features, which are used to predict the electricity price at the next time step.

[0082] Furthermore, to better handle the temporal characteristics and uncertainty in electricity price data, the BiLSTM model was introduced. The BiLSTM model uses the output of the RNN as input and simultaneously passes this input to the BiLSTM network in two directions: forward propagation from the past to the future, and backward propagation from the future to the past. After iterative learning, the BiLSTM network outputs the concatenation of the hidden states in both directions, which is the feature vector after bidirectional propagation. This output contains information about both past and future moments, helping the model better capture temporal characteristics and nonlinear patterns.

[0083] Step S105: input the periodic component into the time series prediction model to obtain a prediction result of the periodic component.

[0084] In addition to some nonlinear characteristics, there are also some periodic characteristics in electricity prices. In order to better analyze the periodic characteristics, this embodiment uses the Prophet model to deal with seasonal or periodic trends in time series.

[0085] The Prophet model is a time series forecasting model based on analyzing variables and time, and combining time series decomposition and machine learning fitting. It mainly trains and fits components such as trend, season, holiday, regression, and noise.

[0086] The general form of the Prophet model algorithm is:

[0087] y(t)=g(t)+s(t)+h(t)+r(t)+ε(t);

[0088] Where y(t) is the predicted value, g(t) is the trend term, s(t) is the cycle term, h(t) is the holiday term, r(t) is the regression term, and ε(t) is the noise term. The trend term is the core of the entire model, primarily used to describe the non-periodic trend of the time series. It includes saturated growth models and linear growth models based on the logistic regression function. The cycle term describes various cyclical trends and primarily relies on the Fourier series to provide the period factor of the cycle term. The holiday term adjusts the model based on the duration of the impact of holidays or other important events, adjusting the time range of potential irregularities in the time series.

[0089] This embodiment obtains the prediction result of the periodic component by inputting the periodic component into the Prophet time series prediction model.

[0090] Step S106 , reconstructing the prediction results of the nonlinear component and the periodic component to obtain the predicted electricity price.

[0091] In an embodiment of the present invention, historical electricity price data is subjected to fully adaptive noise set empirical mode decomposition into multiple modal components to improve overall prediction accuracy. The sample entropy of each modal component is calculated, and each modal component is divided and reconstructed into a periodic component and a nonlinear component based on the size of the sample entropy of each modal component. That is, the original electricity price data is decoded and reconstructed into two types of data with significant eigenvalues, and predictions are made separately to improve the reliability of the prediction. Furthermore, the nonlinear component and related indicator data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain the prediction result of the nonlinear component. The periodic component is input into a time series prediction model to obtain the prediction result of the periodic component. The prediction results of the nonlinear component and the periodic component are reconstructed to obtain the predicted electricity price. The multi-level model combination can significantly improve the accuracy and robustness of short-term electricity price prediction.

[0092] In some embodiments, the accuracy of the ICEEMDAN algorithm is highly dependent on parameter selection when performing fully adaptive noise ensemble empirical mode decomposition on historical electricity price data in step S102. Whether the parameters are correctly selected will affect the algorithm's decomposition results. To ensure the rationality of the data decomposition results, a genetic algorithm can be introduced to optimize the noise parameter selection portion of the ICEEMDAN algorithm.

[0093] Genetic algorithms are computational models that simulate biological evolution, searching for optimal solutions by simulating the natural evolutionary process. The algorithm first initializes the population code; then calculates the fitness of individuals to select the best individuals, forming an optimal population and selecting the best. This optimal population undergoes crossover and mutation for the next generation, selecting individuals based on fitness to ultimately achieve the optimal solution. Genetic algorithms primarily involve selection, crossover, and mutation. For the ICEEMDAN algorithm, the noise parameters optimized primarily include the noise amplitude (i.e., the standard deviation) and the number of ensemble averages.

[0094] The noise amplitude is defined as σ, which represents the amount of white noise added to the original data. If the white noise is too small, the mode may not be excited. If the white noise is too large, it may mask the data characteristics.

[0095] Define the number of integrated averages as N e , which represents the number of times white noise is superimposed and averaged during each modal decomposition process. The greater the number, the more stable the result, but the greater the computational effort and the higher the computational cost.

[0096] Among them, σ is a continuous variable, N e are discrete integers:

[0097] σ∈[σ min ,σ max ];

[0098] N e ∈[N e,min ,N e,max ];

[0099] Next, the reconstruction error is taken as the optimization object and the fitness function is defined to obtain f(σ,N e ), the optimization goal is to minimize the reconstruction error:

[0100]

[0101] Among them, s(t) is the original historical electricity price data, is the data after decomposition and reconstruction by ICEEMDAN, and T is the data length.

[0102] The genetic algorithm steps are:

[0103] (1) Assuming the population size is P, P random individuals are generated when the population is initialized:

[0104] x1,x2,…,x P .

[0105] (2) Calculate the fitness of each individual:

[0106] F i =f(σ i ,N e,i ),i=1,2,…,P.

[0107] (3) According to the fitness F of each individual i The roulette wheel selection method is used for selection, and the probability of each individual being selected is as follows:

[0108]

[0109] (4) Select the parents in pairs from the selected individuals and generate new individuals through linear crossover:

[0110] x new =α·x i +(1-α)·x j ;

[0111] Here, α is a crossover factor, which is usually selected randomly.

[0112] (5) In order to increase population diversity, small perturbations are introduced to individual individuals to perform mutation operations.

[0113] (6) The above steps are continuously iterated through the genetic algorithm until the result is continuously improved in the iterations and the optimal solution is obtained.

[0114] The above method can be used to improve the accuracy of data reconstruction and the denoising effect. e The optimization can, on the one hand, reduce the reconstruction error, improve the similarity between the reconstructed data and the original data, and reduce the problem of distortion and noise. On the other hand, it can reduce the modal aliasing phenomenon and ensure that the different components of the data are correctly decomposed. Secondly, it can effectively improve the computational efficiency and the number of integrated averages N e The value of is closely related to the computational overhead, and a larger N e It can improve the stability of the decomposition results, but it will also increase the amount of calculation. Through the optimization of genetic algorithms, the optimal N that can minimize the consumption of computing resources while maintaining the decomposition quality can be found. e value, thereby improving computational efficiency.

[0115] In some embodiments, considering that the performance of BiLSTM is closely related to network hyperparameters such as learning rate, in order to improve the prediction effect of the model, this embodiment uses the Beluga Whale Optimization Algorithm (BWO) to optimize the network hyperparameters of BiLSTM.

[0116] The Beluga algorithm consists of three steps: encircling prey, hunting behavior, and searching for prey. To achieve global search, whales analyze the positions of other whales and search for prey in a random manner, which is a more reliable way to obtain the global optimal solution. In this embodiment, the Beluga algorithm iteratively optimizes the BiLSTM network hyperparameters by inputting network hyperparameters such as the learning rate into the BiLSTM network and combining them with the feature vectors output by the BiLSTM network. It ultimately outputs the optimized network hyperparameter combination, including the learning rate. This ensures that the BiLSTM performs more accurately in electricity price forecasting tasks and avoids inefficient training or overfitting caused by inappropriate hyperparameter settings.

[0117] In addition, for the evaluation indicators of the model, this embodiment selects root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The calculation formulas of the three indicators are as follows:

[0118]

[0119]

[0120] in, represents the predicted value of electricity price, y iRepresents the actual value of electricity price; N represents the number of historical electricity price data.

[0121] Figure 2 This is a schematic diagram of the overall process of the short-term electricity price forecasting method provided by an embodiment of the present invention. The main process includes:

[0122] Step 1: Decomposition of electricity prices. Use the genetic ensemble empirical mode decomposition method to decompose the electricity price time series into n modal components, such as Mode1, Mode2, ..., Moden.

[0123] Step 2: Feature Quantification. By calculating the sample entropy of each modal component, we quantify the complexity of the different components and determine which components have periodic characteristics and which components have nonlinear characteristics. Components of similar complexity are superimposed to form a new sequence of periodic and nonlinear components.

[0124] Step 3: Electricity price prediction. For periodic data with obvious periodic characteristics, the Prophet model is used to calculate the predicted value; for data with obvious nonlinear characteristics, the RBW (RNN+BiLSTM+BWO) model is used for prediction.

[0125] Step 4: Superimpose the predicted values ​​of the above models to obtain the final electricity price forecast value.

[0126] This embodiment uses the genetic ensemble empirical mode decomposition method to process historical electricity price data and combines it with the entropy weight method to decode and reconstruct the raw electricity price data into two types of data with significant eigenvalues, thereby improving the accuracy of subsequent forecasting. By constructing a new RBW network architecture within the neural network architecture, the problem of neural networks reducing their learning rate due to the tendency to reach optimal solutions is alleviated to a certain extent. Using the Prophet model to analyze periodic data can improve the reliability of the overall forecast results.

[0127] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0128] The following are device embodiments of the present invention. For details not fully described therein, reference may be made to the corresponding method embodiments described above.

[0129] Figure 3 The following is a schematic diagram showing the structure of a short-term electricity price prediction device provided by an embodiment of the present invention. For ease of explanation, only the parts related to the embodiment of the present invention are shown, which are described in detail as follows:

[0130] like Figure 3 As shown, the short-term electricity price forecasting device 3 includes:

[0131] An acquisition module 31 is used to obtain historical electricity price data and relevant indicator data affecting electricity prices;

[0132] A decomposition module 32 is used to perform fully adaptive noise set empirical mode decomposition on historical electricity price data to obtain multiple modal components;

[0133] The processing module 33 is used to calculate the sample entropy of each modal component and divide and reconstruct each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component;

[0134] The prediction module 34 is used to input the nonlinear component and related indicator data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain the prediction result of the nonlinear component; input the periodic component into the time series prediction model to obtain the prediction result of the periodic component; reconstruct the prediction results of the nonlinear component and the periodic component to obtain the predicted electricity price.

[0135] As a possible implementation, the decomposition module 32 is used to:

[0136] The fitness function is established by using the fully adaptive noise ensemble empirical mode decomposition algorithm to decompose the historical electricity price data and reconstruct the data with the error between the historical electricity price data.

[0137] According to the fitness function, the noise parameters of the fully adaptive noise ensemble empirical mode decomposition algorithm are optimized by genetic algorithm;

[0138] The optimized fully adaptive noise ensemble empirical mode decomposition algorithm is used to perform fully adaptive noise ensemble empirical mode decomposition on historical electricity price data.

[0139] As a possible implementation manner, the noise parameter includes at least one of the following: noise amplitude and integration averaging times.

[0140] As a possible implementation manner, the processing module 33 is configured to:

[0141] Determine the modal component whose sample entropy is less than a preset sample entropy threshold as the first modal component, and determine the modal component whose sample entropy is greater than or equal to the sample entropy threshold as the second modal component;

[0142] Reconstruct the first modal component to obtain the periodic component;

[0143] The second modal component is reconstructed to obtain the nonlinear component.

[0144] As a possible implementation, the prediction module 34 is used to:

[0145] Input the nonlinear component and related index data into the convolutional neural network to obtain the prediction results output by the convolutional neural network;

[0146] The prediction results output by the convolutional neural network are input into the bidirectional long short-term memory network to obtain the prediction results of the nonlinear component.

[0147] As a possible implementation manner, the prediction module 34 is further configured to:

[0148] The network hyperparameters of the bidirectional long short-term memory network are optimized by the White Whale algorithm to obtain the optimal network hyperparameters of the bidirectional long short-term memory network;

[0149] Based on the optimal network hyperparameters, the composite prediction model is trained.

[0150] As a possible implementation, the time series prediction model is a Prophet model; the prediction module 34 is further used to train the Prophet model.

[0151] As a possible implementation method, the relevant indicator data include at least one of the following: power generation cost and new energy consumption rate.

[0152] In an embodiment of the present invention, historical electricity price data is subjected to fully adaptive noise set empirical mode decomposition into multiple modal components to improve overall prediction accuracy. The sample entropy of each modal component is calculated, and each modal component is divided and reconstructed into a periodic component and a nonlinear component based on the size of the sample entropy of each modal component. That is, the original electricity price data is decoded and reconstructed into two types of data with significant eigenvalues, and predictions are made separately to improve the reliability of the prediction. Furthermore, the nonlinear component and related indicator data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain the prediction result of the nonlinear component. The periodic component is input into a time series prediction model to obtain the prediction result of the periodic component. The prediction results of the nonlinear component and the periodic component are reconstructed to obtain the predicted electricity price. The multi-level model combination can significantly improve the accuracy and robustness of short-term electricity price prediction.

[0153] Figure 4 Schematic diagram of an electronic device provided by an embodiment of the present invention. Figure 4 As shown, the electronic device 4 of this embodiment includes a processor 40 and a memory 41. The memory 41 stores a computer program 42. When the processor 40 executes the computer program 42, the steps of the above-described method embodiments are implemented. Alternatively, when the processor 40 executes the computer program 42, the functions of the modules / units in the above-described device embodiments are implemented.

[0154] For example, the computer program 42 may be divided into one or more modules / units, which are stored in the memory 41 and executed by the processor 40 to implement the present invention. The one or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 42 in the electronic device 4.

[0155] The electronic device 4 may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 4 and does not constitute a limitation on the electronic device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device 4 may also include input and output devices, network access devices, buses, etc.

[0156] For the sake of convenience and brevity, the division of the above functional modules / units is only used as an example. In actual applications, the above functions can be assigned to different functional modules / units as needed. The above modules / units can be implemented in the form of hardware, software, or a combination of hardware and software.

[0157] In the above embodiments, the descriptions of each embodiment have their own focus. For parts not described or recorded in detail in one embodiment, please refer to the relevant descriptions of other embodiments. Unless otherwise specified or there is a logical conflict, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features of different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0158] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A short-term electricity price forecasting method, characterized in that: include: Obtain historical electricity price data and relevant indicator data affecting electricity prices; Performing fully adaptive noise ensemble empirical mode decomposition on the historical electricity price data to obtain multiple modal components; Calculate the sample entropy of each modal component, and divide and reconstruct each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component; Inputting the nonlinear component and the related index data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component; Inputting the periodic component into a time series prediction model to obtain a prediction result of the periodic component; The prediction results of the nonlinear component and the periodic component are reconstructed to obtain the predicted electricity price.

2. The short-term electricity price forecasting method according to claim 1, characterized in that: The performing of fully adaptive noise set empirical mode decomposition on the historical electricity price data includes: Establishing a fitness function by decomposing the historical electricity price data using a fully adaptive noise set empirical mode decomposition algorithm and taking the error between the reconstructed data and the historical electricity price data as a target; Optimizing noise parameters of the fully adaptive noise ensemble empirical mode decomposition algorithm by a genetic algorithm according to the fitness function; The optimized fully adaptive noise ensemble empirical mode decomposition algorithm is used to perform fully adaptive noise ensemble empirical mode decomposition on the historical electricity price data.

3. The short-term electricity price forecasting method according to claim 2, characterized in that: The noise parameter includes at least one of the following: noise amplitude and integration averaging times.

4. The short-term electricity price forecasting method according to claim 1, characterized in that: The method of dividing and reconstructing each modal component into a periodic component and a nonlinear component according to the sample entropy of each modal component includes: Determine a modal component whose sample entropy is less than a preset sample entropy threshold as a first modal component, and determine a modal component whose sample entropy is greater than or equal to the sample entropy threshold as a second modal component; reconstructing the first modal component to obtain the periodic component; The second modal component is reconstructed to obtain the nonlinear component.

5. The short-term electricity price forecasting method according to claim 1, characterized in that: The nonlinear component and the related index data are input into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component, including: Inputting the nonlinear component and the relevant index data into the convolutional neural network to obtain a prediction result output by the convolutional neural network; The prediction result output by the convolutional neural network is input into the bidirectional long short-term memory network to obtain the prediction result of the nonlinear component.

6. The short-term electricity price forecasting method according to claim 1, characterized in that: Before inputting the nonlinear component and the related index data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component, the method further includes: Performing an optimization operation on the network hyperparameters of the bidirectional long short-term memory network using the White Whale algorithm to obtain the optimal network hyperparameters of the bidirectional long short-term memory network; The composite prediction model is trained based on the optimal network hyperparameters.

7. The short-term electricity price forecasting method according to claim 1, characterized in that: The time series prediction model is a Prophet model; Before inputting the periodic component into the time series prediction model to obtain the prediction result of the periodic component, the method further includes: training the Prophet model.

8. The short-term electricity price forecasting method according to any one of claims 1 to 7, characterized in that: The relevant indicator data include at least one of the following: power generation cost and new energy consumption rate.

9. A short-term electricity price forecasting device, characterized in that: include: The acquisition module is used to obtain historical electricity price data and related indicator data that affect electricity prices; a decomposition module, configured to perform fully adaptive noise set empirical mode decomposition on the historical electricity price data to obtain multiple modal components; A processing module is used to calculate the sample entropy of each modal component and divide and reconstruct each modal component into a periodic component and a nonlinear component according to the size of the sample entropy of each modal component; A prediction module, configured to input the nonlinear component and the related indicator data into a composite prediction model composed of a convolutional neural network and a bidirectional long short-term memory network to obtain a prediction result of the nonlinear component; The periodic component is input into a time series prediction model to obtain a prediction result of the periodic component; the prediction results of the nonlinear component and the periodic component are reconstructed to obtain a predicted electricity price.

10. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

Images