Net power prediction method, device, equipment, medium and program product

Abstract

The embodiment of the invention provides a net power prediction method and device, equipment, a medium and a program product, and particularly relates to the technical field of electric power. The method comprises the following steps: optimizing parameters of a signal decomposition model by adopting a swarm intelligence optimization algorithm, and determining a target parameter combination; based on the determined target parameter combination, performing variational mode decomposition on the net power historical sequence data to obtain a plurality of mode components; extracting high-frequency mutation features and low-frequency trend features for the plurality of modal components to obtain modal component feature vectors corresponding to the modal components respectively; performing net power prediction based on the modal component feature vectors in combination with the bidirectional time sequence modeling model to obtain prediction results corresponding to the modal components; and generating a final net power prediction result based on the prediction result corresponding to each modal component. The method is used for achieving the effect of improving the net power prediction precision and real-time performance in a complex scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electric power, and in particular to a net power prediction method, device, equipment, medium and program product. BACKGROUND

[0002] Under the background of new power system construction, new energy (such as photovoltaic and wind power) access to distribution network has become the core direction of energy structure transformation. The output of new energy has significant intermittency and volatility, which leads to the complex characteristics of the superposition of high-frequency mutation and low-frequency trend of the net power (the difference between new energy power generation and load power) of the power grid. This characteristic poses a severe challenge to the real-time scheduling, voltage stability control and optimal configuration of the energy storage system of the distribution network. For example, when the photovoltaic power suddenly drops due to cloud cover or the wind power suddenly increases due to wind speed mutation, if the change of the net power cannot be predicted in time, it may lead to power frequency fluctuation, equipment overload and even power failure. Therefore, a high-precision and low-latency real-time net power prediction method is urgently needed to support the second-level scheduling decision. At present, the distribution network regulation system needs to complete the prediction within a time window of 15 minutes or even shorter, and requires the prediction model to have both the sensitive capture ability of high-frequency disturbance and the long-term modeling ability of low-frequency trend. SUMMARY

[0003] The embodiments of the present application provide a net power prediction method, device, equipment, medium and program product to improve the prediction accuracy and real-time performance of the net power in complex scenarios.

[0004] In a first aspect, the embodiments of the present application provide a net power prediction method, comprising:

[0005] Using a swarm intelligence optimization algorithm, the parameters of the signal decomposition model are optimized to determine the target parameter combination;

[0006] Based on the determined target parameter combination, the net power historical sequence data is variational mode decomposed to obtain a plurality of modal components;

[0007] For the plurality of modal components, high-frequency mutation features and low-frequency trend features are extracted to obtain modal component feature vectors corresponding to each modal component, respectively;

[0008] Based on the modal component feature vectors, a bidirectional time series modeling model is used for net power prediction to obtain prediction results corresponding to each modal component, respectively;

[0009] Based on the prediction results corresponding to each modal component, respectively, a final net power prediction result is generated.

[0010] In a second aspect, the embodiments of the present application provide a net power prediction device, comprising:

[0011] The determining module is configured to optimize parameters of the signal decomposition model by using a swarm intelligence optimization algorithm, and determine a target parameter combination.

[0012] The decomposition module is configured to perform variational mode decomposition on the net power historical sequence data based on the determined target parameter combination, and obtain a plurality of modal components.

[0013] The extraction module is configured to extract high-frequency mutation features and low-frequency trend features from the plurality of modal components, and obtain modal component feature vectors corresponding to the modal components, respectively.

[0014] The prediction module is configured to perform net power prediction based on the modal component feature vectors and in combination with a bidirectional time series modeling model, and obtain prediction results corresponding to the modal components, respectively.

[0015] The generation module is configured to generate a final net power prediction result based on the prediction results corresponding to the modal components, respectively.

[0016] In a third aspect, an embodiment of the present application provides a net power prediction device, including a memory and a processor.

[0017] The memory stores computer execution instructions.

[0018] The processor executes the computer execution instructions stored in the memory, so that the processor executes the first aspect and / or various possible implementation manners of the first aspect.

[0019] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed by a processor, the computer execution instructions are used to implement the first aspect and / or various possible implementation manners of the first aspect.

[0020] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the computer program implements the first aspect and / or various possible implementation manners of the first aspect.

[0021] The method, device, equipment, medium and program product for predicting net power provided by the embodiments of the present application can avoid the local optimal risk of artificial experience parameter adjustment by dynamically optimizing the parameters of the signal decomposition model by using a swarm intelligence optimization algorithm, ensure that the parameter combination is strongly matched with the time-frequency characteristics of the historical sequence data of the net power, extract high-frequency mutation features and low-frequency trend features for each modal component, comprehensively depict the dynamic characteristics of the net power by combining the two, and avoid the prediction deviation caused by single-scale features. The bidirectional time sequence modeling model reduces the prediction error caused by information loss of the unidirectional model by complementing the bidirectional time sequence information, and improves the robustness of the prediction result. Through the above technical means, the accuracy and real-time performance of the net power prediction in a complex scene are improved. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present application and, together with the description, further serve to explain the principles of the application.

[0023] Figure 1 Flowchart of the method for predicting net power provided by the present application Figure One ;

[0024] Figure 2 Flowchart of the method for predicting net power provided by the present application Figure Two ;

[0025] Figure 3 Application process diagram of the sparrow search algorithm provided by the present application;

[0026] Figure 4 Structure diagram of the gated recurrent unit (GRU) provided by the present application;

[0027] Figure 5 Structure diagram of the bidirectional gated recurrent unit provided by the present application;

[0028] Figure 6 Implementation process diagram of the method for predicting net power provided by the present application;

[0029] Figure 7 Structure diagram of the net power prediction device provided by the present application;

[0030] Figure 8 Structure diagram of the net power prediction device provided by the present application.

[0031] Through the above drawings, the specific embodiments of the present application have been shown, and will be described in more detail hereinafter. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application by any means, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0032] The exemplary embodiments will be described in detail herein with reference to the attached drawings. In the following description, like reference numerals refer to like elements, unless the context clearly dictates otherwise. The following description of exemplary embodiments is not representative of all embodiments consistent with the present application. Instead, it is merely an example of apparatus and methods consistent with some aspects of the present application as detailed in the appended claims.

[0033] Existing net power prediction methods mainly rely on the fusion framework of signal decomposition and deep learning, but there are significant limitations:

[0034] Signal decomposition method parameter solidification: such as empirical mode decomposition (EMD) and variational mode decomposition (VMD) need to manually preset the mode number and penalty factor, and this static parameter configuration is difficult to adapt to the complex fluctuations of net power signals.

[0035] Feature modeling fragmentation: existing methods usually independently input the decomposed modal components into the prediction model, ignoring the correlation between different frequency components (such as the modulation effect of high-frequency disturbance on low-frequency trend).

[0036] One-way time series modeling: in existing net power prediction methods, although LSTM models can improve prediction accuracy, they have the defects of low computational efficiency and insufficient real-time performance. The computational complexity increases linearly with the number of modal decomposition, resulting in a significant increase in model inference delay, making it difficult to meet the second-level response demand in power dispatching scenarios.

[0037] Therefore, the present application provides a net power prediction method, the core technical concept of which is: through the collaborative optimization of adaptive signal decomposition algorithm and deep learning model, high-precision real-time prediction of net power is realized. This concept breaks through the limitations of traditional methods in parameter solidification, feature fragmentation, and one-way modeling, with dynamic parameter optimization, multi-scale feature extraction, and bidirectional time series modeling as the core. Specifically, the sparrow search algorithm (SSA) is used to dynamically optimize the parameters of variational mode decomposition (VMD) to realize adaptive decomposition of the signal; the multi-scale convolutional neural network (CNN) is used to extract the local features of each modal component in parallel; the bidirectional gated recurrent unit (BiGRU) is used to fuse historical and future information to construct spatio-temporal correlation features. Finally, through feature superposition, end-to-end prediction is realized, significantly improving the net power prediction accuracy and real-time performance in complex scenarios.

[0038] The technical solutions of the present application and how the technical solutions of the present application solve the above technical problems will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present application will be described below with reference to the drawings.

[0039] Figure 1 Flowchart of the net power prediction method provided in the present application Figure One As shown in Figure 1 The method comprises the following steps:

[0040] S101, adopt a swarm intelligence optimization algorithm to optimize the parameters of the signal decomposition model, and determine a target parameter combination.

[0041] The swarm intelligence optimization algorithm is an optimization algorithm that simulates the group behavior in nature (such as ant colony, bird flock, etc.), and finds the optimal solution through the cooperation and information sharing among individuals in the group.

[0042] The signal decomposition model (such as variational mode decomposition VMD) needs to set some parameters, such as the number of modes, the penalty factor, etc. The values of these parameters will affect the effect of signal decomposition. Through the swarm intelligence optimization algorithm, the optimal parameter combination can be automatically searched in the parameter space.

[0043] After the search by the optimization algorithm, a set of parameters is finally determined, and this set of parameters can make the signal decomposition model achieve good decomposition effect on the current net power historical sequence data.

[0044] S102, based on the determined target parameter combination, perform variational mode decomposition on the net power historical sequence data to obtain multiple modal components.

[0045] The net power historical sequence data refers to the record of the difference between the new energy power and the load power changing with time within a certain time range, usually in the form of time series. This data reflects the dynamic changes of the net power of the power grid after the new energy is connected to the grid.

[0046] In the signal decomposition process, the original net power historical sequence data is decomposed into multiple modal components, and each modal component itself is also a time series.

[0047] S103, for the multiple modal components, extract high-frequency mutation features and low-frequency trend features to obtain modal component feature vectors corresponding to each modal component.

[0048] For each modal component, its high-frequency mutation features and low-frequency trend features are extracted respectively. The high-frequency mutation features reflect the short-term fluctuations and mutations in the signal, and the low-frequency trend features reflect the long-term trend of the signal.

[0049] The extracted high-frequency mutation features and low-frequency trend features are combined to form the feature vector corresponding to each modal component. These feature vectors can comprehensively describe the characteristics of each modal component.

[0050] S104, based on the modal component feature vector, combine the bidirectional time series modeling model to perform net power prediction, and obtain the prediction results corresponding to each modal component respectively.

[0051] The bidirectional time series modeling model is a model that can consider the forward and reverse information of the time series at the same time. For example, the BiGRU model includes a forward GRU and a backward GRU.

[0052] The modal component feature vector is input into the bidirectional time series modeling model, and the model will predict the net power according to these feature vectors. Since the model considers bidirectional time series information, it can more accurately predict the change of net power.

[0053] S105, based on the prediction results corresponding to each modal component respectively, generate the final net power prediction result.

[0054] The prediction results of each modal component are comprehensively processed, which can be a simple weighted sum or a more complex fusion strategy. The purpose is to integrate the prediction results of each modal component into a complete net power prediction result.

[0055] The net power prediction method provided by the embodiment of the application can avoid the local optimal risk of manual experience parameter tuning by using a swarm intelligence optimization algorithm to dynamically optimize the parameters of the signal decomposition model, ensure that the parameter combination is strongly matched with the time-frequency characteristics of the net power historical sequence data, extract high-frequency mutation features and low-frequency trend features for each modal component, and combine the two to comprehensively depict the dynamic characteristics of the net power, avoid prediction deviation caused by single-scale features, and reduce prediction errors caused by information loss of a unidirectional model through bidirectional time series information complementation, and improve the robustness of the prediction result. Through the above technical means, the net power prediction accuracy and real-time performance in a complex scene are improved.

[0056] Figure 2 The flowchart of the net power prediction method provided by the application Figure Two As shown in Figure 2 the embodiment Figure 1 The net power prediction method is described in detail based on the embodiment, and the method comprises the following steps:

[0057] S201, determine the current scene type based on the net power historical sequence data.

[0058] The original data is preprocessed, including data cleaning, feature data extraction, time alignment processing, time granularity standardization, and normalization processing.

[0059] Data cleaning refers to handling outliers, missing values, and noise in the data. For missing values, different strategies are taken according to the missing rate: if the missing rate exceeds 30%, the corresponding sample is deleted; if the missing rate is low, linear interpolation is used to complete the data.

[0060] Feature data extraction refers to extracting valuable feature data from raw data. These feature data may include time feature data, environmental feature data (such as weather type, irradiance), new energy output feature data, etc.

[0061] Time alignment processing refers to the need for time alignment of different types of feature data extracted, in order to facilitate subsequent analysis and processing.

[0062] Time granularity standardization refers to the unification of sampling frequencies of different types of feature data, ensuring that all data is collected within the same time interval. For example, the present application provides that data sampling is performed every 15 minutes.

[0063] Normalization processing refers to normalization processing of different types of feature data to eliminate dimensional influence. For new energy power generation power, load power and other feature data, it is mapped to the [0, 1] interval, and the specific calculation formula is as follows:

[0064] ;

[0065] wherein, represents the current data value, represents the minimum value in the sample, represents the maximum value in the sample, represents the normalized value, ranging in the [0, 1] interval.

[0066] Based on the pre-processed raw data, the net power historical sequence data is determined. The net power historical sequence data is analyzed to identify and determine the specific scenario type in which the current power system operates. The scenario type may include but is not limited to normal operation state, load peak, renewable energy high output, weather change influence, etc. Determining the scenario type is crucial for the optimization of signal decomposition model parameters, as different scenarios may require different parameter settings to adapt to their specific data characteristics.

[0067] S202, based on the scenario type, combining the swarm intelligence optimization algorithm, the parameters of the signal decomposition model are optimized to obtain the target parameter combination.

[0068] This step involves optimizing the parameters of the signal decomposition model in combination with the scenario type and the swarm intelligence optimization algorithm. Among them, the fitness function of the swarm intelligence optimization algorithm can be defined according to the scenario type.

[0069] Swarm intelligence optimization algorithms, such as the Sparrow Search Algorithm (SSA), are optimization methods that simulate the collective behavior of natural groups to find optimal solutions. They mimic the social behavior of sparrows to search for the best solution.

[0070] By using swarm intelligence optimization algorithms such as SSA, the key parameters of the VMD (Variational Modal Decomposition) model, such as the number of modes and the penalty factor, are dynamically searched and optimized. These parameters are crucial for achieving adaptive signal decomposition of the net power sequence. Through optimization, it can be ensured that the signal decomposition model can adapt to data changes in different scenarios, thereby improving the accuracy of net power prediction.

[0071] In one possible implementation, the swarm intelligence optimization algorithm includes the Sparrow Search Algorithm. By using the swarm intelligence optimization algorithm to optimize the parameters of the signal decomposition model, the target parameter combination can be determined, which can specifically include the following steps:

[0072] Initialize the position parameters of the sparrow population, which correspond to the parameters to be optimized in the signal decomposition model;

[0073] Update and iterate the positions of the explorers and followers in the sparrow population;

[0074] When the preset iteration end condition is met, output the position parameters of the sparrow individual with the optimal fitness value in the sparrow population to determine the target parameter combination.

[0075] In this implementation, a group of sparrows (solutions) is initialized, and the position parameters of each sparrow represent a set of parameters to be optimized in the signal decomposition model. By iteratively updating the positions of the sparrows, the optimal solution is gradually approached. When the preset number of iterations is reached or other stopping conditions are met (such as the fitness value being less than a certain threshold), the algorithm stops iterating. At this time, the sparrow individual with the optimal fitness value is selected from the sparrow population, and its position parameters are the optimal parameter combination sought.

[0076] In one possible implementation, the positions of the explorers and followers in the sparrow population are updated and iterated, which can specifically include the following steps:

[0077] In each iteration process, based on the environmental safety information of the explorer in that iteration process, determine whether to use a global search strategy or a random search strategy to update the position of the explorer, and update the position of the follower based on the updated explorer position.

[0078] In this implementation, in each iteration, based on the environmental safety information of the explorer, it is decided whether to use a global search strategy or a random search strategy to update the position of the explorer. The global search strategy is used for extensive exploration of the solution space, while the random search strategy is used for local search. The follower updates its own position based on the new position of the explorer in order to find a better solution.

[0079] Specifically, first, the possible value range of the parameters to be optimized in the signal decomposition model needs to be determined. After determining the value range of the parameters, the parameter initialization of the initial population is performed.

[0080] A position parameter is randomly assigned to each sparrow (i.e., each potential solution), and the position parameter is within the preset parameter value range.

[0081] The fitness value of each sparrow in the initial population is calculated to evaluate the performance of the parameter combination it represents. The calculation formula of the fitness value is as follows:

[0082] ;

[0083] ;

[0084] wherein, contains the initial position information of the entire population, wherein each row represents an individual, each column represents a parameter dimension, and each element represents the value of the ith individual at the jth parameter dimension. , respectively represent the lower bound and the upper bound of each parameter dimension, which can be a scalar or a vector. is a random number in the interval [0, 1] following a uniform distribution, or a D-dimensional random vector.

[0085] In the sparrow search algorithm (SSA), the explorer in the iterative optimization process updates the position according to the following formula, and the update rule depends on the comparison result of the environmental safety information, i.e., the warning value and the safety threshold :

[0086] ;

[0087] wherein, represents the position of the ith sparrow at the tth generation, when , the sparrow performs global search foraging. When , the sparrow performs random walk with normal distribution. is a constant that controls the step attenuation speed; is the maximum number of iterations preset in advance; represents a random number following a normal distribution.

[0088] In the sparrow search algorithm (SSA), the follower updates the position according to the following formula:

[0089] ;

[0090] Random perturbation is performed on the individual with poor fitness to avoid local stagnation, and position resetting is performed on the marginal individual to jump out of local optimum. Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation;

[0091] In addition, position updating (early warning mechanism) can also be performed based on the comparison of the fitness value of the sparrow and the global optimal fitness value, and the specific formula is as follows:

[0092]

[0093] Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation;

[0094] Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation;

[0095] Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation;

[0096] Xi(t+1) represents the position of the i-th sparrow in the t+1 generation; Xi(t+1) represents the position of the i-th sparrow in the t+1 generation;

[0097] Reference Figure 3 The diagram illustrates the application process of the sparrow search algorithm provided in this application. First, the algorithm starts from the "Start" node and enters the "Initialize Population" step, randomly generating an initial population. Next, the individuals in the population are divided into "Explorers" and "Followers." During iteration, the positions of explorers and followers are updated based on environmental safety information. Then, a "Warning Mechanism" is randomly triggered. If the preset iteration termination condition is not met, the algorithm will continue iterating; once met, the algorithm will terminate at the "End" node, outputting the optimal solution.

[0098] S203. Based on the determined combination of target parameters, perform variational mode decomposition on the net power historical sequence data to obtain multiple mode components.

[0099] In this step, variational mode decomposition is performed on the net power historical sequence data using the optimal parameter combination determined by a swarm intelligence optimization algorithm (such as the Sparrow Search Algorithm, SSA). These optimal parameters may include the number of modes and a penalty factor.

[0100] Specifically, a constrained variational model can be constructed. First, an objective function is defined, the purpose of which is to minimize the sum of the bandwidths of all modal components. This is achieved by optimizing the center frequency of each modal component. The objective function and the constraint condition (the sum of all modal components equals the original signal) can be expressed as:

[0101] ;

[0102] in, Indicates the total number of modal components; It is the Dirac function; This represents the convolution operation; This indicates taking the derivative with respect to time t; This represents the k-th modal component; This represents the center frequency of the k-th modal component; This represents the original signal; j is the imaginary unit.

[0103] For the constrained variational model described above, Lagrange multipliers and a quadratic penalty term can be introduced, and the solution can be obtained iteratively using the alternating direction multiplier method (ADMM). After the iteration converges, K modal components are obtained.

[0104] S204. For multiple modal components, high-frequency mutation features and low-frequency trend features are extracted to obtain the modal component feature vectors corresponding to each modal component.

[0105] In one possible implementation, high-frequency abrupt change features and low-frequency trend features are extracted for multiple modal components to obtain modal component feature vectors corresponding to each modal component. Specifically, this may include the following steps:

[0106] The high-frequency mutation features and low-frequency trend features of the plurality of modal components are extracted in parallel based on multi-scale convolution kernels to obtain modal component feature vectors corresponding to the respective modal components.

[0107] In the present embodiment, in order to capture high-frequency mutations and low-frequency trend features, convolution kernels of different sizes are designed. Narrow convolution kernels (such as 3-5 in size) are used to capture high-frequency mutation features, while wide convolution kernels (such as 7-11 in size) are used to extract low-frequency trend features. Multi-scale convolution kernels are applied in parallel to each modal component. This means that each modal component will be processed by narrow and wide kernels, thereby extracting high-frequency and low-frequency features at the same time. The modal component is convolved using a narrow convolution kernel to identify and extract short-term mutation features. The modal component is convolved using a wide convolution kernel to capture long-term trends and change patterns. The features extracted by the convolution operation are pooled and dimensionally reduced to generate a feature vector for each modal component. These feature vectors contain key information of the modal component, which is used for subsequent prediction models.

[0108] Specifically, each modal component can be converted into a three-dimensional tensor with a shape of (number of samples, time step, number of features), so that the time series data can be converted into a format that can be processed by CNN.

[0109] The convolutional neural network includes two convolutional layers, where the first convolutional layer uses 1D convolution kernels of size 3-5, with a number of 32, to extract high-frequency mutation features, and outputs a shape of (number of samples, time step-2, 32). The second convolutional layer uses 1D convolution kernels of size 7-11, with a number of 64, to extract low-frequency trend features, and outputs a shape of (number of samples, time step-8, 64).

[0110] The features extracted by the convolution operation are pooled and dimensionally reduced. The pooled multi-channel features can then be concatenated into a one-dimensional vector as input for subsequent models.

[0111] S205, based on the modal component feature vectors, combining a bidirectional time series modeling model to predict the net power, to obtain the prediction results corresponding to the respective modal components.

[0112] In one possible implementation, based on the modal component feature vectors, combining a bidirectional time series modeling model to predict the net power, to obtain the prediction results corresponding to the respective modal components, which can specifically include the following steps:

[0113] The modal component feature vectors are input into the bidirectional time series modeling model;

[0114] The bidirectional time sequence feature is extracted by a bidirectional time sequence modeling model with a double-channel GRU structure, wherein a forward GRU processes the feature sequence in a forward time sequence to capture forward time sequence dependent features, and a backward GRU processes the feature sequence in a reverse time sequence to capture reverse time sequence dependent features;

[0115] The forward time sequence dependent features and the reverse time sequence dependent features output by the forward GRU and the backward GRU are spliced at each time step to generate a feature representation that fuses bidirectional time sequence information;

[0116] Based on the feature representation that fuses bidirectional time sequence information, a net power prediction result of the corresponding modal component is determined.

[0117] In the embodiment, the feature vectors obtained by feature extraction on the modal components obtained from the variational modal decomposition are input into a bidirectional time sequence modeling model. The model adopts a bidirectional gated recurrent unit (BiGRU) structure, which includes two GRU channels, one for forward processing and the other for reverse processing. The forward GRU processes the feature sequence in a forward time sequence to capture forward time sequence dependent features in the time sequence, i.e., the influence of earlier time points on the current time point. The backward GRU processes the feature sequence in a reverse time sequence to capture reverse time sequence dependent features, i.e., the influence of later time points on the current time point. At each time step, the outputs of the forward GRU and the backward GRU are spliced to fuse the forward and reverse time sequence dependent features and generate a feature representation that contains bidirectional time sequence information. Based on the feature representation that fuses bidirectional time sequence information, the model determines a net power prediction result of the corresponding modal component. This step utilizes the in-depth understanding of the model of the time sequence, including short-term mutations and long-term trends, to produce accurate net power predictions.

[0118] GRU (gated recurrent unit) is the basic building block of BiGRU, which includes two main gating mechanisms: update gate and reset gate.

[0119] The update gate is used to control the degree of preservation of the state of the current state , and the formula is:

[0120] ;

[0121] wherein, Sigmoid activation function is denoted as is denoted as the weight matrix input to the update gate; is denoted as the bias term of the update gate; is denoted as the input vector at time step t; is denoted as the output of the update gate.

[0122] The reset gate is used to control the state For the current state The degree of forgetting is calculated using the following formula:

[0123] ;

[0124] in, This represents the Sigmoid activation function; This represents the weight matrix input to the reset gate; This indicates the offset of the door being reset; This represents the input vector at time step t; This indicates that the output of the gate is being reset.

[0125] Candidate states can be calculated by combining the current input and the reset state from the previous time step, using the following formula:

[0126] ;

[0127] in, Represents the hyperbolic tangent activation function; This represents the Hadamard product, element-wise multiplication. This represents the candidate state weight matrix, used to weight the previous state after reset. and current input Combined, to calculate candidate states; Indicates the candidate state bias term; Indicates that the output of the reset gate is being reset; This represents the state vector at the previous moment.

[0128] The current state can be calculated by combining the previous state, the candidate state at the current time step, and the update gate output. The calculation formula is as follows:

[0129] ;

[0130] in, This represents the state vector at the current time t; This indicates the output of the update gate at time step t; This represents the state vector at the previous moment; This represents the candidate state vector at the current time t; This represents the Hadamard product, element-wise multiplication.

[0131] Reference Figure 4 The diagram shown is a structural schematic of the gated recurrent unit (GRU) provided in this application. The current input... and the hidden state of the previous moment After the door was updated and reset door The processing yields the hidden state at the current moment. Update the gate control information retention ratio, reset the gate control forgetting level, and use the tanh activation function to generate candidate states.

[0132] At each time step, the forward GRU and backward GRU respectively compute their hidden states. The forward and backward hidden states are then fused using a matrix concatenation operation, as shown in the following formula:

[0133] ;

[0134] Where concat represents the matrix concatenation operation; Indicates a forward-hidden state; This indicates a backward hidden state.

[0135] Reference Figure 5 The diagram shows a schematic of the bidirectional gated recurrent unit (GRU) provided in this application, used for feature extraction from time series data. The diagram includes GRU units in two directions: a forward GRU processes the sequence from... arrive The backward GRU processing sequence is from arrive .

[0136] The hidden state at the last time step is fully connected and mapped, and a single-mode prediction value for each modal component is generated by a linear activation function.

[0137] S206. Based on the prediction results corresponding to each modal component, generate the final net power prediction result.

[0138] In one possible implementation, the final net power prediction result is generated based on the prediction results corresponding to each modal component, which may specifically include the following steps:

[0139] The prediction results corresponding to each modal component are weighted and fused to generate the final net power prediction result.

[0140] In practical implementation, the predicted value of each modal component can be... Normalization is performed, and the normalization formula is shown below:

[0141] ;

[0142] in, , This step ensures that the predicted values ​​are within the correct range by recording the maximum and minimum values ​​of the original data.

[0143] Next, the final net power prediction can be calculated using the following formula, which is the sum of all inversely normalized single-mode predictions:

[0144] ;

[0145] wherein K is the total number of modal components; is the kth modal component after de-normalization.

[0146] Referring to Figure 6 , an implementation process diagram of the net power prediction method provided by the present application is shown. The method extracts the net power historical sequence data from the preprocessed original data, then initializes the population parameters of the SSA algorithm, and then optimizes the parameters of VMD using the SSA algorithm to determine the optimal modal number and penalty factor. The optimized parameters are used to perform variational modal decomposition on the net power historical sequence data to obtain a plurality of intrinsic modal function components. After these IMF components are fused with other features (such as environmental features), the prediction value of each modal component is obtained through the CNN-BiGRU model. Finally, the prediction results of each modal component are superimposed to generate the final net power prediction result, and the prediction performance of the model is evaluated. The whole process aims to significantly improve the net power prediction accuracy in high fluctuation scenarios through dynamic optimization of signal decomposition parameters, deep fusion of cross-modal features, and network structure perception aggregation mechanism, and to provide support for real-time scheduling, voltage control and energy storage optimization decision of distribution network.

[0147] Through the above steps, end-to-end modeling from signal decomposition to feature fusion to prediction output is realized. The net power high-precision real-time prediction method based on SSA-VMD-CNN-BiGRU provided by the present application dynamically optimizes the modal number and bandwidth parameters of variational modal decomposition (VMD) using sparrow search algorithm (SSA), accurately separates high-frequency noise and low-frequency trend in the net power signal through the role division of explorers and followers and the dynamic early warning mechanism, and realizes the adaptive frequency domain decomposition of the net power signal. Multi-scale convolutional neural network (CNN) is used to extract local features of each modal component, different scale convolution kernels are used to process each modal component in parallel, and local convolution and pooling operations are used to compress redundant information. The bidirectional gated recurrent unit (BiGRU) integrates the spatial and temporal features through the bidirectional gating mechanism, and outputs the hidden state vector containing rich spatial and temporal information. SSA-VMD solves the poor adaptability problem caused by the fixed parameters of the traditional model, and improves the flexibility and accuracy of the model through dynamic optimization. The CNN-BiGRU model solves the problems of insufficient accuracy and response delay caused by feature fragmentation, improves the accuracy and real-time performance of the prediction by deeply fusing the features of different modal components.

[0148] Figure 7 The structure diagram of the net power prediction device provided by the present application is shown in Figure 7As shown, the net power prediction device 70 provided by the embodiment includes:

[0149] The determination module 701 is configured to adopt a swarm intelligence optimization algorithm to optimize parameters of the signal decomposition model, and determine a target parameter combination.

[0150] The decomposition module 702 is configured to perform variational mode decomposition on the net power historical sequence data based on the determined target parameter combination, to obtain a plurality of modal components.

[0151] The extraction module 703 is configured to perform high-frequency mutation feature and low-frequency trend feature extraction on the plurality of modal components, to obtain a modal component feature vector corresponding to each modal component.

[0152] The prediction module 704 is configured to perform net power prediction based on the modal component feature vector and in combination with a bidirectional time series modeling model, to obtain a prediction result corresponding to each modal component.

[0153] The generation module 705 is configured to generate a final net power prediction result based on the prediction result corresponding to each modal component.

[0154] In a possible implementation, the determination module is specifically configured to:

[0155] determine a current scene type based on the net power historical sequence data;

[0156] optimize parameters of the signal decomposition model based on the scene type and in combination with a swarm intelligence optimization algorithm, to obtain the target parameter combination.

[0157] In a possible implementation, the swarm intelligence optimization algorithm includes a sparrow search algorithm, and the determination module is specifically configured to:

[0158] initialize position parameters of a sparrow population, the position parameters corresponding to parameters to be optimized in the signal decomposition model;

[0159] update and iterate positions of explorers and positions of followers in the sparrow population;

[0160] when a preset iteration end condition is met, output a position parameter of a sparrow individual with an optimal fitness value in the sparrow population, to determine the target parameter combination.

[0161] In a possible implementation, the determination module is specifically configured to:

[0162] in each iteration process, determine whether to update the position of the explorer by using a global search strategy or a random search strategy according to environmental safety information of the explorer in the iteration process, and update the position of the follower based on the updated position of the explorer.

[0163] In a possible implementation, the extraction module is specifically configured to:

[0164] The high-frequency mutation features and the low-frequency trend features of the plurality of modal components are extracted in parallel based on the multi-scale convolution kernel, to obtain modal component feature vectors corresponding to the plurality of modal components respectively.

[0165] In a possible implementation, the prediction module is specifically configured to:

[0166] input the modal component feature vectors into the bidirectional time series modeling model;

[0167] extract bidirectional time series features through the double-channel GRU structure in the bidirectional time series modeling model, wherein the forward GRU processes the feature sequence in a forward time sequence to capture forward time sequence dependent features, and the backward GRU processes the feature sequence in a reverse time sequence to capture reverse time sequence dependent features;

[0168] concatenate the forward time sequence dependent features and the reverse time sequence dependent features output by the forward GRU and the backward GRU at each time step to generate feature representations that fuse bidirectional time sequence information;

[0169] determine the net power prediction result of the corresponding modal component based on the feature representations that fuse bidirectional time sequence information.

[0170] In a possible implementation, the generation module is specifically configured to:

[0171] weight fuse the prediction results corresponding to the plurality of modal components to generate a final net power prediction result.

[0172] The net power prediction apparatus provided in this embodiment can perform the method provided in the method embodiment, and has similar implementation principles and technical effects, which will not be described here again in this embodiment.

[0173] Figure 8 A structural schematic diagram of a net power prediction device provided in this application is shown in FIG. 8. Figure 8 As shown in FIG. 8, the net power prediction device 80 provided in this embodiment includes at least one processor 801 and a memory 802. Optionally, the device 80 further includes a communication component 803. The processor 801, the memory 802, and the communication component 803 are connected through a bus.

[0174] In the specific implementation process, the at least one processor 801 executes the computer execution instructions stored in the memory 802, so that the at least one processor 801 performs the method described above.

[0175] The specific implementation process of the processor 801 can refer to the method embodiment described above, and has similar implementation principles and technical effects, which will not be described here again in this embodiment.

[0176] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in combination with the application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0177] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), such as at least one disk memory.

[0178] The bus can be an industry standard architecture (ISA) bus, a peripheral component (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0179] The present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above method.

[0180] The present application also provides a computer readable storage medium, which stores computer execution instructions, and when a processor executes the computer execution instructions, the above method is implemented.

[0181] The above readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0182] An example readable storage medium is coupled to the processor such that the processor can read information from the readable storage medium and can write information to the readable storage medium. Of course, the readable storage medium can also be a part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist as discrete components in the device.

[0183] The division of units is only a logical functional division, and in actual implementation, there can be another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0184] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment scheme.

[0185] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit.

[0186] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application or the part of the present application that essentially contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various program code storage media.

[0187] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware. The foregoing program can be stored in a computer readable storage medium. The program executes to perform the steps of the above-mentioned method embodiments; and the foregoing storage medium includes various media capable of storing program codes, such as ROM, RAM, magnetic disk, or optical disk.

[0188] Finally, it should be noted that other embodiments of the present application will readily occur to those skilled in the art upon consideration of the specification and practice of the present application disclosed herein. The present application is intended to include all such variations, uses, or adaptations of the application in which the general principles of the application are used to best advantage and encompassed within its scope. The present application is not limited to the precise structures described and shown in the accompanying drawings and figures, and can be practiced with variation of modifications and alterations without departing from the scope of the present application. The scope of the present application is limited only by the claims appended hereto.

Claims

1. A net power prediction method, characterized in that, include: A swarm intelligence optimization algorithm is used to optimize the parameters of the signal decomposition model and determine the target parameter combination; Based on the determined combination of target parameters, variational mode decomposition is performed on the net power historical sequence data to obtain multiple mode components; For the multiple modal components, high-frequency mutation features and low-frequency trend features are extracted to obtain the modal component feature vectors corresponding to each modal component. Based on the modal component feature vectors, net power prediction is performed using a bidirectional time series modeling model to obtain the prediction results for each modal component. Based on the prediction results corresponding to each modal component, the final net power prediction result is generated.

2. The method according to claim 1, characterized in that, The method employs a swarm intelligence optimization algorithm to optimize the parameters of the signal decomposition model and determine the target parameter combination, including: The current scenario type is determined based on the net power historical sequence data; Based on the scenario type, and combined with the swarm intelligence optimization algorithm, the parameters of the signal decomposition model are optimized to obtain the target parameter combination.

3. The method according to claim 1 or 2, characterized in that, The swarm intelligence optimization algorithm includes a sparrow search algorithm. The use of the swarm intelligence optimization algorithm to optimize the parameters of the signal decomposition model and determine the target parameter combination includes: Initialize the location parameters of the sparrow population, which correspond to the parameters to be optimized in the signal decomposition model; The positions of explorers and followers in the sparrow population are updated and iterated. When the preset iteration termination condition is met, the location parameters of the sparrow individual with the best fitness value in the sparrow population are output to determine the target parameter combination.

4. The method according to claim 3, characterized in that, The updating and iteration of the positions of explorers and followers in the sparrow population includes: During each iteration, based on the explorer's environmental safety information during that iteration, it is determined whether to use a global search strategy or a random search strategy to update the explorer's position, and the follower's position is updated based on the updated explorer's position.

5. The method according to claim 1 or 2, characterized in that, The process involves extracting high-frequency abrupt change features and low-frequency trend features for the multiple modal components to obtain modal component feature vectors corresponding to each modal component, including: Based on the parallel extraction of high-frequency mutation features and low-frequency trend features of the multiple modal components using multi-scale convolution kernels, the modal component feature vectors corresponding to each modal component are obtained.

6. The method according to claim 1 or 2, characterized in that, The net power prediction based on the modal component feature vectors, combined with a bidirectional time-series modeling model, yields prediction results for each modal component, including: The modal component feature vectors are input into the bidirectional temporal modeling model; Bidirectional temporal feature extraction is performed using the dual-channel GRU structure in the bidirectional temporal modeling model. The forward GRU processes the feature sequence in the forward temporal order to capture forward temporal dependency features, while the backward GRU processes the feature sequence in the reverse temporal order to capture reverse temporal dependency features. At each time step, the forward and backward temporal dependency features output by the forward GRU and backward GRU are concatenated to generate a feature representation that integrates bidirectional temporal information. Based on the feature representation of the fused bidirectional time-series information, the net power prediction result of the corresponding modal component is determined.

7. The method according to claim 1 or 2, characterized in that, The generation of the final net power prediction result based on the prediction results corresponding to each modal component includes: The prediction results corresponding to each modal component are weighted and fused to generate the final net power prediction result.

8. A net power prediction device, characterized in that, include: The determination module is used to optimize the parameters of the signal decomposition model using a swarm intelligence optimization algorithm and determine the target parameter combination; The decomposition module is used to perform variational mode decomposition on the net power historical sequence data based on the determined target parameter combination to obtain multiple mode components; The extraction module is used to extract high-frequency mutation features and low-frequency trend features for the multiple modal components, and obtain the modal component feature vectors corresponding to each modal component. The prediction module is used to perform net power prediction based on the modal component feature vectors and combined with a bidirectional time series modeling model to obtain the prediction results corresponding to each modal component. The generation module is used to generate the final net power prediction result based on the prediction results corresponding to each modal component.

9. A net power prediction device, characterized in that, include: Memory, processor; The memory stores instructions that the computer executes; The processor executes computer execution instructions stored in memory, causing the processor to perform the method as claimed in any one of claims 1-7.

10. A computer-readable storage medium or computer program product, characterized in that, A computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as claimed in any one of claims 1-7; or, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images