Distributed photovoltaic power generation power prediction method, device, equipment and storage medium

Abstract

The present invention discloses a distributed photovoltaic power generation power prediction method, device, equipment and storage medium. The method includes dynamically clustering historical power generation parameter information according to an improved clustering algorithm to obtain clustered target time series data; performing variational modal decomposition on each target time series data based on the optimal decomposition parameter corresponding to the target time series data to obtain at least two sub-modal components; using an improved kernel principal component analysis method to perform high-dimensional feature space mapping on each sub-modal component to obtain the contribution of historical power generation weather data to distributed photovoltaic power generation; inputting historical power generation parameter information into a long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; the embodiment of the present invention, through the above technical solution, can improve the stability and accuracy of photovoltaic power generation prediction, realize the accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation, and provide a prerequisite guarantee for supporting the power grid to quickly adjust its power.

Description

Technical Field

[0001] The present invention relates to the technical field of ultra-short-term prediction of renewable energy power generation, and in particular to a distributed photovoltaic power generation power prediction method, device, equipment and medium. Background Art

[0002] Under the guidance of a series of goals such as "dual carbon" and "building a new power system", it is necessary to support the large-scale grid connection of distributed photovoltaics. However, due to the peak and intermittent nature of distributed photovoltaic power generation, its large-scale access will lead to problems such as over / under voltage in the distribution network and increased load peak-to-valley differences, which will not only increase the power dispatch burden of the main grid, but also affect the power supply quality of the distribution network and even threaten the safe operation of the system. Therefore, it is urgent to explore and utilize the regulation potential of distributed photovoltaic aggregate power in the distribution network to effectively control the voltage and reduce the peak-to-valley difference. In addition, the ultra-short-term prediction of distributed photovoltaic aggregate power generation is a prerequisite for supporting its rapid power regulation.

[0003] The prediction methods in the prior art have the following disadvantages: 1) Distributed photovoltaic power generation is completely different from centralized photovoltaic power stations. Its photovoltaic arrays are widely distributed, and there are great differences in the surrounding environment and weather. Different time and space distributions lead to large differences in the timing characteristics of its power generation. Therefore, the existing prediction methods for centralized power stations are difficult to apply to distributed photovoltaic power generation prediction; 2) Most of the existing prediction methods are aimed at short-term, medium-term and long-term power generation prediction of photovoltaic power generation, and rarely consider the non-stationary characteristics of photovoltaic power generation. This is mainly because these non-stationary fluctuations are small in amplitude and short in period. On a longer time scale, they can be regarded as weak disturbances that can be ignored, but on an ultra-short time granularity, these non-stationary characteristics cannot be ignored; 3) Existing prediction methods rarely consider the inducing factors of distributed photovoltaic power generation non-stationarity (such as meteorological factors), but if these non-stationary inducing factors are not clearly understood on an ultra-short time scale, they cannot be accurately introduced into the prediction model, and it is difficult to improve the prediction accuracy. Therefore, there is an urgent need for a photovoltaic power generation prediction method to solve the bottleneck problem of the difficulty in accurately predicting massive distributed photovoltaic aggregate power generation with small capacity, wide distribution and strong non-stationarity, and to improve the stability and accuracy of photovoltaic power generation prediction. Summary of the invention

[0004] In view of this, the present invention provides a distributed photovoltaic power generation power prediction method, device, equipment and storage medium, which can solve the bottleneck problem of difficult to accurately predict massive distributed photovoltaic aggregate power generation with small capacity, wide distribution and strong non-stationarity, improve the stability and accuracy of photovoltaic power generation prediction, and realize accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation, providing a prerequisite guarantee for supporting the power grid to quickly adjust its power.

[0005] According to one aspect of the present invention, an embodiment of the present invention provides a distributed photovoltaic power generation power prediction method, the method comprising:

[0006] Acquire historical power generation parameter information of distributed photovoltaic power generation units in a target area within a preset time period, and dynamically cluster the historical power generation parameter information based on an initial cluster center selected from the historical power generation parameter information and a Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain clustered target time series data;

[0007] Based on the optimal decomposition parameters corresponding to the target time series data, each of the target time series data is subjected to variational modal decomposition to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor;

[0008] Using at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components, the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power is obtained;

[0009] The historical power generation parameter information is input into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree.

[0010] According to another aspect of the present invention, an embodiment of the present invention further provides a distributed photovoltaic power generation power prediction device, the device comprising:

[0011] A clustering module is used to obtain historical power generation parameter information of distributed photovoltaic power generation units in a target area within a preset time period, and dynamically cluster the historical power generation parameter information based on an initial clustering center selected from the historical power generation parameter information and a Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain clustered target time series data;

[0012] A decomposition module, configured to perform variational modal decomposition on each of the target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor;

[0013] A contribution determination module, used to use at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components to obtain the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power;

[0014] A prediction module is used to input the historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree.

[0015] According to another aspect of the present invention, an embodiment of the present invention further provides an electronic device, the electronic device comprising:

[0016] at least one processor; and

[0017] a memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the distributed photovoltaic power generation power prediction method described in any embodiment of the present invention.

[0019] According to another aspect of the present invention, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the distributed photovoltaic power generation power prediction method described in any embodiment of the present invention when executed.

[0020] The technical effect of the present invention is that by acquiring the historical power generation parameter information of the distributed photovoltaic power generation units in the target area within a preset time period, and dynamically clustering the historical power generation parameter information based on the initial clustering center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain the clustered target time series data, the problems that the distributed power generation time series characteristics with different time and space distributions are very different and it is difficult to select the benchmark photovoltaic array for aggregation prediction can be considered, and more complex nonlinear data can be processed. There is no need to randomly select K sample points in the initialization stage, and the interpretability is stronger; at least two sub-modal components are obtained by performing variational mode decomposition on each target time series data based on the optimal decomposition parameters corresponding to the target time series data, and the optimized variational mode decomposition algorithm Compared with the traditional variational mode decomposition algorithm, it reduces the influence of artificially set related parameters, reduces the possibility of under-decomposition or over-decomposition, and more effectively suppresses modal aliasing and realizes accurate decomposition of modes; by using the preset improved kernel principal component analysis method to perform high-dimensional feature space mapping on each sub-modal component, the contribution of historical power generation weather data to distributed photovoltaic power generation power is obtained, and this contribution is introduced into the neural network in the form of initial weight setting of forgetting gate, and the historical power generation parameter information is input into the long short-term memory neural network to obtain the distributed photovoltaic power generation power prediction result, which can have a clear understanding of the non-stationary inducing factors on the ultra-short time scale, thereby improving the stability and accuracy of photovoltaic power generation prediction, and realizing the accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation, providing the premise guarantee for supporting the power grid to quickly adjust its power.

[0021] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0023] Figure 1 A flow chart of a distributed photovoltaic power generation power prediction method provided by an embodiment of the present invention;

[0024] Figure 2 A flow chart of another distributed photovoltaic power generation power prediction method provided by an embodiment of the present invention;

[0025] Figure 3A flow chart of another distributed photovoltaic power generation power prediction method provided by an embodiment of the present invention;

[0026] Figure 4 A schematic diagram of a distributed photovoltaic power generation power prediction architecture provided by an embodiment of the present invention;

[0027] Figure 5 A structural block diagram of a distributed photovoltaic power generation prediction device provided by an embodiment of the present invention;

[0028] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0030] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0031] In one embodiment, Figure 1 A flow chart of a distributed photovoltaic power generation power prediction method provided for one embodiment of the present invention. This embodiment can be applied to the situation of ultra-short-term distributed photovoltaic power generation power prediction based on consideration of non-stationary influencing factors. The method can be executed by a distributed photovoltaic power generation power prediction device. The distributed photovoltaic power generation power prediction device can be implemented in the form of hardware and / or software. The distributed photovoltaic power generation power prediction device can be configured in an electronic device.

[0032] like Figure 1 As shown, the distributed photovoltaic power generation power prediction method in this embodiment includes the following specific steps:

[0033] S110, obtaining historical power generation parameter information of distributed photovoltaic power generation units in the target area within a preset time period, and dynamically clustering the historical power generation parameter information based on the initial clustering center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain clustered target time series data.

[0034] The target area can be understood as a relevant area selected for photovoltaic power generation prediction, which may include multiple distributed photovoltaic power generation units. The preset time period can be understood as an ultra-short time period, exemplarily a 4-6 hour time period, and of course, other time periods may also be used, which is not limited in this embodiment.

[0035] In some embodiments, the historical power generation parameter information may include historical power generation data and historical power generation weather data; wherein the historical power generation weather data may include but is not limited to surface air pressure, temperature, humidity, cloud cover, and light intensity. In this embodiment, the historical power generation data and the historical power generation weather data may be clustered by a preset improved clustering algorithm to obtain the corresponding target time series data.

[0036] In this embodiment, the target time series data can be understood as the power generation time series data obtained by dynamically clustering the historical power generation parameter information, which is represented as power generation data in different spaces and at different times.

[0037] In this embodiment, for the historical power generation parameter information of the distributed photovoltaic power generation units in the target area within a preset time period, the improved K-means++ algorithm can be used to construct a dynamic clustering partition model of distributed photovoltaic power generation, which can support the aggregation of distributed photovoltaics belonging to the same cluster cluster according to a unified specimen to obtain the time series data of distributed photovoltaic aggregated power generation power; in some embodiments, the spectral clustering algorithm can also be used to perform time series clustering on the original data set to obtain a classified data set for each time interval to form the time series data of distributed photovoltaic aggregated power generation power; specifically, a historical sample data can be selected as the initial clustering center, and the Gaussian kernel distance from other historical sample data in the historical data set to the initial clustering center can be determined, and the Gaussian kernel distance of each historical sample data in the historical data set can be determined according to the Gaussian kernel distance. The probability of the historical sample data being the next cluster center is calculated, and the historical sample data with the largest probability is used as the next cluster center, until the number of cluster centers reaches the preset cluster center point threshold, and K cluster centers are obtained after division, and each historical sample data is divided into the cluster center with the closest core distance to form K cluster clusters and aggregate them to obtain the corresponding target time series data; in other embodiments, at least two historical sample data can also be arbitrarily selected from the historical data set as the initial cluster center, and the Euclidean distance between the historical data and the initial cluster center is calculated, and it is classified into the cluster where the nearest cluster center is located, and the average value of all data in each cluster is calculated to form a new cluster center, and the execution is repeated until the stopping condition is met or the maximum number of iterations is reached to obtain the corresponding target time series data, which is not limited in this embodiment.

[0038] S120. Perform variational modal decomposition on each target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor.

[0039] Among them, the optimal decomposition parameters can be understood as the number and penalty factor of the optimal sub-modal components for variational modal decomposition of the target time series data.

[0040] In this embodiment, there are multiple ways of variational modal decomposition. The artificial rabbit algorithm can be used to determine the target number of sub-modal components and the target penalty factor required for variational modal decomposition of the target time series data. The target time series data is subjected to variational modal decomposition according to the target number of sub-modal components and the target penalty factor to obtain at least two sub-modal components. It can be understood that the artificial rabbit algorithm is used to optimize the number of sub-modal components and the penalty factor of variational modal decomposition with the minimum mean envelope entropy and sample entropy as the optimization objectives, and the optimized number of sub-modal components and the penalty factor are substituted into the variational modal decomposition method to perform modal decomposition on each target time series data. In some embodiments, the classified data set of each time interval can also be normalized and then subjected to variational modal extraction and decomposition to obtain the target number of sub-modal components and the target penalty factor corresponding to the classified data set of each time interval. This embodiment does not limit this.

[0041] S130. Use at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each sub-modal component to obtain the contribution of historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power.

[0042] Among them, the contribution degree can characterize the impact of historical power generation weather data, namely temperature, humidity, light, cloud cover and other data on the non-stationarity of distributed photovoltaic aggregate power generation, thereby benchmarking and analyzing the quantitative impact of multiple influencing factors on distributed photovoltaic power generation.

[0043] In this embodiment, the improved kernel principal component analysis method can be used to perform high-dimensional feature space mapping on each sub-modal component to obtain the contribution of historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power. It can be understood that the contribution of historical power generation weather data, i.e., temperature, humidity, light, cloud cover and other data to the distributed photovoltaic power generation power is obtained by using the preset improved kernel principal component analysis method, so as to benchmark and analyze the quantitative impact of multiple types of influencing factors on non-stationarity. In some embodiments, by determining the input variable matrix corresponding to each sub-modal component, in order to reduce the influence of noise data on the kernel principal component analysis, at least two kernel functions are used to perform high-dimensional feature space mapping on the input variable matrix to obtain at least two corresponding high-dimensional mapping matrices, and the kernel matrix is ​​obtained by averaging multiple high-dimensional mapping matrices, and the sharing degree and the principal component component are benchmarked and analyzed by the eigenvalue and eigenvector of the kernel matrix, and the contribution of the historical power generation weather data to the distributed photovoltaic power generation power is determined by the similarity between the principal component component and the historical power generation weather data in the historical power generation parameter information. This embodiment is not limited here.

[0044] S140, inputting historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein, the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree.

[0045] In this embodiment, each memory unit in the long short-term memory neural network includes a forget gate f t , input gate i t , output gate o t , the input of the memory unit at time t is x t , the output is h t , the state of the memory cell is c t ; where X=[x1,...x t ,...,x n ], which represents the historical power generation parameter information.

[0046] In this embodiment, the distributed photovoltaic power generation power can be predicted based on the multi-source data fusion method of the pre-trained long short-term memory neural network. It should be noted that when using the pre-trained long short-term memory neural network to predict the distributed photovoltaic power generation power, the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power is required to set the initial weight value of the forgetting gate in the long short-term memory neural network. It can be understood that the quantitative influence of the historical power generation weather data, i.e., temperature, humidity, light, cloud cover and other multi-source data on the non-stationarity of distributed photovoltaic aggregate power generation is introduced into the neural network in the form of the initial weight of the forgetting gate, and then the historical power generation parameter information is input into the input gate of the pre-trained long short-term memory neural network to obtain the distributed photovoltaic power generation power prediction result.

[0047] In this embodiment, the normalized historical power generation data and historical weather data can be used as labels, and the long short-term memory neural network is iteratively trained through the labeled historical power generation data and historical weather data to obtain a pre-trained long short-term memory neural network. The training target is the loss function Minimize, and back-propagate the loss function using the gradient descent algorithm to update the weights and bias ω parameters of the prediction model, expressed as: Among them, n represents the number of samples in the historical data set, y i is the output of the ith training data after the long short-term memory neural network, is the true output of the i-th training data, and α is the learning rate.

[0048] The technical solution of the embodiment of the present invention obtains the historical power generation parameter information of the distributed photovoltaic power generation unit in the target area within a preset time period, and dynamically clusters the historical power generation parameter information based on the initial clustering center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain the clustered target time series data. It can take into account the large differences in the time series characteristics of distributed power generation with different time and space distributions, and the difficulty in selecting a benchmark photovoltaic array for aggregate prediction. It can handle more complex nonlinear data, does not need to randomly select K sample points in the initialization stage, and has stronger interpretability; by performing variational mode decomposition on each target time series data based on the optimal decomposition parameter corresponding to the target time series data to obtain at least two sub-modal components, the optimized variational mode decomposition algorithm Compared with the traditional variational mode decomposition algorithm, this method reduces the influence of artificially set related parameters, reduces the possibility of under-decomposition or over-decomposition, and more effectively suppresses mode aliasing and realizes accurate mode decomposition; by using the preset improved kernel principal component analysis method to perform high-dimensional feature space mapping on each sub-modal component, the contribution of historical power generation weather data to distributed photovoltaic power generation power is obtained. This contribution is introduced into the neural network in the form of initial weight setting of the forgetting gate, and the historical power generation parameter information is input into the long short-term memory neural network to obtain the distributed photovoltaic power generation power prediction result. It can have a clear understanding of the non-stationary inducing factors on the ultra-short time scale, thereby improving the stability and accuracy of photovoltaic power generation prediction, realizing the accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation, and providing the premise guarantee for supporting the power grid to quickly adjust its power.

[0049] In one embodiment, Figure 2 A flow chart of another distributed photovoltaic power prediction method provided for an embodiment of the present invention. Based on the above embodiments, this embodiment dynamically clusters the historical power generation parameter information based on the initial clustering center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain clustered target time series data; performs variational modal decomposition on each target time series data based on the optimal decomposition parameter corresponding to the target time series data to obtain at least two sub-modal components; uses at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each sub-modal component to obtain the contribution of historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power; and inputs the historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result, which is further refined.

[0050] like Figure 2 As shown, the distributed photovoltaic power generation power prediction method in this embodiment may specifically include the following steps:

[0051] S210, obtaining historical power generation parameter information of distributed photovoltaic power generation units in a target area within a preset time period, normalizing the historical power generation parameter information to obtain target historical power generation parameter information, and forming the target historical power generation parameter information into a historical data set.

[0052] In this embodiment, the historical power generation data and historical weather data (including temperature, humidity, cloud cover, light intensity, etc.) of each distributed photovoltaic power generation unit are collected, the historical power generation data and historical weather data are normalized to obtain the target historical power generation parameter information, and the target historical power generation parameter information is formed into a historical data set, wherein the historical data set can be expressed by the formula X=[x1,...,x n ], x i Represents one of the historical sample data, and n represents the total number of samples in the historical data set; in some embodiments, the way in which the historical power generation parameter information is normalized is expressed by the formula: x′=x / |Max|, wherein |Max| represents the maximum absolute value of the historical power generation parameter information, x represents the historical power generation parameter information before normalization, and x′ represents the target historical power generation parameter information after normalization.

[0053] S220: Randomly select a historical sample data from the historical data set as an initial cluster center, and put the initial cluster center into the cluster center set.

[0054] In one embodiment, a historical sample data can be randomly selected from the historical data set as the initial cluster center, and the initial cluster center is placed in the cluster center set. The historical data set is represented as: X = [x1, ..., x n ], where x i Represented as the i-th historical sample data.

[0055] S230. Determine the first Gaussian kernel distance from other historical sample data in the historical data set to the initial cluster center, determine the probability of each historical sample data in the historical data set being the next cluster center based on the first Gaussian kernel distance, and use the historical sample data with the largest probability as the next cluster center and store it in the cluster center set. Repeat this step until the number of cluster center points in the cluster center set reaches the preset cluster center point number threshold, and obtain K cluster centers after division.

[0056] Among them, the preset cluster center point number threshold can be understood as the K cluster centers finally formed, and the value of K is at least 2. The preset cluster center point number threshold can be set according to experience, needs, etc., and this embodiment does not limit it here.

[0057] In this embodiment, from the historical data set X = [x1, ..., x n] randomly selects one data sample as the initial cluster center c, and calculates the kernel distance between each sample and the initial cluster center Among them, x i represents the historical data of the i-th sample, Π(·,·) represents the Gaussian kernel function, which is specifically expressed as: Where a and b represent two data samples, and σ is the bandwidth. Calculate the probability P(i) of each sample being selected as the next cluster center, and select the sample point corresponding to the maximum probability value as the next cluster center. The probability of being selected as the next cluster center is expressed as: Repeat the above steps until K cluster centers are selected, and calculate the nuclear distance between each sample and all cluster centers, put the sample into the aggregation point with the smallest distance, and finally form K clusters. The distributed photovoltaics belonging to the same cluster are aggregated according to the unified specimen to obtain the time series data of the distributed photovoltaic aggregated power generation power.

[0058] It can be understood that the distance calculation formula of the Gaussian kernel distance in this embodiment is expressed as: Among them, c represents the cluster center, Π(·,·) is the Gaussian kernel function, and the Gaussian kernel function is expressed by the formula: Wherein, a and b represent two historical sample data in the historical data set, respectively, ||ab|| 2 It is expressed as the Euclidean distance between a and b, and σ is the bandwidth. In one embodiment, the probability is expressed as: in, It is expressed as the Gaussian kernel distance of the i-th historical sample data.

[0059] S240, determining the Gaussian kernel distance from each historical sample data in the historical data set to K cluster centers, and dividing each historical sample data to the cluster center with the closest kernel distance according to the Gaussian kernel distance, to form K clusters.

[0060] In this embodiment, the Gaussian kernel distance from each historical sample data in the historical data set to the K cluster centers is determined, and each historical sample data is divided into the cluster center with the closest kernel distance according to the Gaussian kernel distance to form K cluster clusters. The distance calculation formula of the Gaussian kernel distance is expressed as: Among them, c represents the cluster center, Π(·,·) is the Gaussian kernel function, and the Gaussian kernel function is expressed by the formula: Wherein, a and b represent two historical sample data in the historical data set, respectively, ||ab|| 2 It is expressed as the Euclidean distance between a and b, and σ is the bandwidth.

[0061] S250: Aggregate historical sample data belonging to the same cluster to obtain corresponding target time series data.

[0062] In this embodiment, in the K clusters formed, the distributed photovoltaics belonging to the same cluster are aggregated according to a unified specimen to obtain the time series data of the distributed photovoltaic aggregated power generation power. In this embodiment, the historical power generation parameter information is dynamically clustered based on the initial cluster center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain the clustered target time series data. This method improves the traditional kmeans++ algorithm. Compared with the traditional Kmeans++ algorithm, the kernel distance is introduced to replace the original Euclidean distance, which can handle more complex nonlinear data; compared with the traditional Kmeans algorithm, the improved kmeans++ algorithm does not need to randomly select K sample points in the initialization stage, and is more interpretable.

[0063] S260. Determine the optimal decomposition parameters required for variational modal decomposition of the target time series data using an artificial rabbit algorithm; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor.

[0064] In this embodiment, the artificial rabbit algorithm can be used to determine the optimal decomposition parameters required for variational modal decomposition of the target time series data; it can be understood that the artificial rabbit algorithm is used to optimize the number of sub-modes and penalty factors of variational modal decomposition with the minimum average envelope entropy and sample entropy as optimization targets, and the optimized number of sub-modals and penalty factors are substituted into the variational modal decomposition method to perform modal decomposition on the normalized power generation data.

[0065] In one embodiment, S260 may further include: S2601-S2603:

[0066] S2601, randomly initialize the current position of the artificial rabbit; wherein the current position is represented as the initial decomposition parameter of the target time series data, and the initial decomposition parameter is represented as L i =[L i1 ,L i2 ], L i1 represents the number of sub-modes, L i2 Represents the penalty factor.

[0067] S2602. Determine the value of the energy factor corresponding to the rabbit in the tth iteration, and determine the candidate position of the rabbit in the next iteration based on the value.

[0068] In some embodiments, the candidate positions include a first candidate position and a second candidate position. Accordingly, determining the candidate position of the rabbit in the next iteration according to the value situation further includes: when the value situation is that the energy factor value is greater than 1, determining the candidate position of the i-th rabbit in the next iteration as the first candidate position; wherein the first candidate position is the candidate position that the i-th rabbit will choose to move to the position of the j-th rabbit when foraging, which is expressed by the formula: Among them, i,j=1,…,n,i≠j, It is expressed as the ratio of the position of rabbit i moving to rabbit j; L i (t) represents the current position of the i-th rabbit at the t-th iteration, L j (t) represents the current position of the jth rabbit at the tth iteration; r1, r2 are both random numbers between (0, 1), round represents rounding, o is the mutation operator, which takes a value of 0 or 1, n1 is a random number that obeys the normal distribution, and n is the number of rabbits; when the energy factor is less than or equal to 1, the candidate position of the rabbit at the next iteration is determined to be the second candidate position; the second candidate position is the candidate position when the i-th rabbit randomly selects a cave to hide, which is expressed by the formula: in, It represents the location of the cave that the i-th rabbit will randomly choose; Represented as rabbit i heading to cave B i The ratio of the moved position; r3 and r4 are random numbers between (0, 1), o and g are mutation operators, and their values ​​are 0 or 1.

[0069] S2603. Determine the target fitness function of the candidate position and the current position, and update the rabbit position according to the target fitness function to obtain the updated target position, and return to the step of determining the value of the energy factor corresponding to the rabbit in the current iteration until the maximum number of iterations is reached or the target position with the minimum target fitness function is found, and use the target position as the target sub-modal component number and target penalty factor of variational mode decomposition; wherein the target fitness function is expressed as the sum of the average envelope entropy and sample entropy of at least two sub-modal components.

[0070] In some embodiments, the target fitness function is expressed as: Among them, K represents the number of sub-modal components corresponding to the variational mode decomposition, E i represents the envelope entropy of the ith submodal component, B m (r) is represented as sample entropy, which represents the total number of time series with intervals less than r in the m dimension, and is expressed as: Where N represents the total data length corresponding to each modal component, m represents the data length selected from the total data length, r represents the similarity tolerance, X(i) and X(j) represent the i-th and j-th submodal components respectively, and D[X(i), X(j)] represents the maximum difference between the corresponding elements of X(i) and X(j); the energy factor is expressed as follows: Where t represents the tth iteration; T represents the maximum number of iterations, and f is a random number between (0, 1).

[0071] In one embodiment, the target position is expressed by the formula: Among them, L i (t) represents the current position, Indicated as candidate location.

[0072] S270. Perform variational modal decomposition on the target time series data according to the optimal decomposition parameters to obtain at least two sub-modal components.

[0073] In this embodiment, the target time series data can be subjected to variational modal decomposition according to the optimal decomposition parameters to obtain at least two sub-modal components. Specifically, the variational modal decomposition problem of the target time series data can be converted into a variational function problem containing constraints, and then converted into an unconstrained variational problem through the Lagrangian multiplication operator λ and the penalty factor α, and the Lagrangian multiplication operator, the center frequency and the finite bandwidth, and the desired sub-modal component are iteratively updated through the alternating direction multiplier algorithm to obtain at least two sub-modal components.

[0074] In one embodiment, S270 may further include: S2701-S2703:

[0075] S2701. Convert the variational mode decomposition problem of the target time series data into a variational function problem containing constraints.

[0076] In this embodiment, the variational function problem is expressed as follows: Among them, {u k}:={u1,...,u K} represents K sub-modal components, {ω k}:={ω1,...,ω K} represents the center frequency of each submodal component, K represents the number of submodal components, x(t) represents the target time series data, δ(t) is the pulse function, represents the partial derivative with respect to time t, * represents the convolution calculation; u k (t) is represented by the kth submodal component at time t, which is expressed as u k (t) = A k (t)cos(φ k (t)), where Ak (t) is u k (t) envelope amplitude; φ k (t) is expressed as phase, φ k ′(t)≥0; Expressed as a one-sided spectrum function; It is expressed as baseband; Represents the preset estimated center frequency.

[0077] S2702. Use a Lagrange multiplication operator and a target penalty factor to convert the variational function problem into an unconstrained variational function problem.

[0078] In this embodiment, the unconstrained variational problem is expressed as follows:

[0079]

[0080] Among them, ||·||2 represents the L2 norm, <·> represents the inner product, λ represents the Lagrange multiplication operator, and α represents the target penalty factor.

[0081] S2703, using the alternating direction multiplier algorithm to iteratively update the Lagrange multiplication operator, the center frequency and the finite bandwidth, and the desired submodal component to obtain the saddle point, and terminating the iterative process when the stopping condition is met to obtain K desired submodal components.

[0082] In this embodiment, the alternating direction multiplier algorithm is used to iteratively update u n+1 ,ω n+1 ,λ n+1 We can find the saddle point where u n+1 ,ω n+1 ,λ n+1 Respectively expressed as:

[0083]

[0084]

[0085]

[0086] Among them, ρ represents the learning rate. Convert the above formula to the frequency domain and solve it, and finally get: That is, the update formula of the Lagrange multiplication operator is expressed as: Among them, ρ represents the learning rate; the update of the center frequency and the limited bandwidth is expressed by the formula: The update of the expected submodal component is formulated as: Among them, x(ω), u i (ω) and λ(ω) are the Parseval Fourier transforms of their respective variables; the stopping condition is expressed by the formula: Among them, ε represents the iterative convergence value.

[0087] S280. Determine an input variable matrix corresponding to each sub-modal component; wherein the input variable matrix includes envelope amplitude data of M time dimensions.

[0088] In this embodiment, based on the obtained K sub-modal components u k (t) = A k (t)cos(φ k (t)), k = 1, 2, ... K, where A k (t) is u k (t) envelope amplitude; phase φ k (t) is a non-decreasing function, that is, φ k ′(t)≥0. Assume that the number of submodal components is K; each submodal component includes the envelope amplitude data of M time dimensions, and the amplitude of the jth frequency band in the i-th time period is expressed as A ij , the input variable matrix is ​​expressed as:

[0089]

[0090] S290. Use at least two kernel functions to perform high-dimensional feature space mapping on the input variable matrix to obtain at least two target variable matrices.

[0091] In this embodiment, the input variable matrix is ​​mapped into a D-dimensional high-dimensional feature space. First, a kernel function Φ(·) is used for nonlinear mapping to obtain the corresponding variable matrix, which is denoted as Φ(A K×M ) D×M , expressed as: In order to reduce the influence of noise data on kernel principal component analysis, S different kernel functions are selected to perform high-dimensional mapping on the input variable matrix, and finally S high-dimensional mapping matrices, i.e., the target variable matrix in this embodiment, can be obtained, which is expressed as: Φ1(A K×M ) D×M ,Φ2(A K×M ) D×M ,…Φ S (A K×M ) D×M ; Among them, (A K×M ) D×M Represented as the target variable matrix after mapping into high-dimensional feature space,

[0092] S2100. Calculate the average value of at least two target variable matrices to obtain a corresponding new variable matrix, and determine the eigenvalues ​​of the new variable matrix and its corresponding orthogonal unit eigenvector.

[0093] In this embodiment, the average value of at least two target variable matrices is calculated to obtain the corresponding new variable matrix, and the eigenvalues ​​of the new variable matrix and the corresponding orthogonal unit eigenvectors are determined. In this embodiment, the new variable matrix Φ'(A K×M ) D×M , expressed as:

[0094]

[0095] S2110. Determine the contribution of each eigenvalue according to a preset contribution formula and sort them according to the contribution to select G eigenvalues ​​and their corresponding G orthogonal unit eigenvectors.

[0096] In this embodiment, the contribution of each eigenvalue is determined according to a preset contribution formula and sorted according to the contribution to select G eigenvalues ​​(λ1, λ2, ..., λ G ) and its corresponding G orthogonal unit eigenvectors (e1, e2, …, e G ), the preset contribution is expressed as: Among them, the eigenvalue λ i That is, the kernel matrix Φ'(A K×M ) D×M The characteristic value of .

[0097] S2120. Extracting principal component components from the input variable matrix corresponding to each sub-modal component based on each orthogonal unit eigenvector.

[0098] In this embodiment, the principal component is extracted from the input variable matrix corresponding to each sub-modal component according to each orthogonal unit eigenvector, which is expressed as follows:

[0099] S2130. Determine the similarity between the principal component and the historical power generation weather data in the historical power generation parameter information based on a preset similarity formula, determine the correspondence between the principal component and the historical power generation weather data based on the similarity, and determine the contribution of the historical power generation weather data to the distributed photovoltaic power generation power based on the correspondence.

[0100] In this embodiment, according to the obtained principal component Y i , calculate the principal components and the normalized weather data Z j , (j=1,…Q) (temperature, humidity, light intensity, cloud cover, etc.), where Q represents the number of types of weather data. The specific calculation formula is expressed as: Among them, cov(Z,Y) represents the covariance between historical power generation weather data and the principal component, σ Z represents the standard deviation of historical power generation weather data, σ Yrepresents the standard deviation of the principal component, and E represents the expected value; when β is closer to 1, the similarity between the historical power generation weather data and the principal component is higher. First, for the first principal component Y1, calculate Z j , (j=1,…Q) and Y1, select the weather data with the largest similarity as the benchmark analysis data of Y1, and then calculate Z for the second principal component Y2 j The remaining weather data in the above table are similar to the data of Y2, and the one with the highest similarity is selected as the benchmark analysis data of Y2. i With Z j By making a one-to-one correspondence, we can analyze the quantitative impact (i.e. contribution) of various types of weather data on power generation data; the closer β is to 1, the more similar the two sets of data are.

[0101] S2140, historical power generation parameter information x t Input to the memory unit in the long short-term memory neural network, each input corresponds to the forget gate f of the memory unit t , input gate i t , output gate o t .

[0102] In this embodiment, each memory unit in the long short-term memory neural network includes a forget gate f t , input gate i t , output gate o t , the input of the memory unit at time t is x t , the output is h t , the state of the memory cell is c t ; where X=[x1,...x t ,...,x n ], which represents the historical power generation parameter information;

[0103] In this embodiment, the forget gate f t Expressed as: f t =σ(W f h t-1 +U f x t +b f ), input gate i t Expressed as: i t =σ(W i h t-1 +U i x t +b i ), output gate o t Expressed as: o t =σ(W o h t-1 +U o xt +b o ), where σ represents the activation function, b f ,b i ,b o is the bias term, h t-1 is the output of the memory unit at time t-1, W f ,U f ,W i ,U i ,W o ,U o are all weight matrices, where W f ,W i ,U i ,W o ,U o The weight matrix is ​​randomly initialized weights, and the forget gate U f The initial weight value of is determined by the contribution, and the forget gate U f Expressed as: U f =[0,ρ1,ρ2,…,ρ G ].

[0104] S2150, calculate the state c of the memory unit at each moment t .

[0105] In this embodiment, state c t It is expressed as: Among them, c t-1 is the state of the memory unit at time t-1, ⊙ is the Harmand product operator, is the candidate state of the memory unit at time t, and the candidate state is expressed by the formula: Among them, tanh() is the activation function, W c and U c Expressed as a weight matrix, b c is the bias term.

[0106] S2160, calculate the output h of the memory unit at each moment t , will output h t Denoted as the output y of the long short-term memory neural network, it is the prediction result of distributed photovoltaic power generation.

[0107] Among them, the output h t Expressed as: h t =o t ⊙tanh(c t ).

[0108] The technical solution of the embodiment of the present invention aggregates the distributed photovoltaic power generation power belonging to the same cluster cluster by selecting the initial cluster center from the historical power generation parameter information and the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information. Compared with the traditional Kmeans++ algorithm, the kernel distance is introduced to replace the original Euclidean distance, and more complex nonlinear data can be processed; compared with the traditional Kmeans++ algorithm, the improved kmeans++ algorithm does not need to randomly select K sample points in the initialization stage, and has stronger interpretability; the artificial rabbit algorithm is used to determine the optimal decomposition parameters required for variational mode decomposition of the target time series data, and the target time series data is subjected to variational mode decomposition according to the optimal decomposition parameters to obtain at least two sub-modal components, which reduces the influence of artificially set related parameters compared with the traditional variational mode decomposition algorithm, and reduces the possibility of under-decomposition or over-decomposition; at the same time, the optimization algorithm is optimized with average envelope entropy and Minimizing sample entropy can more effectively suppress modal aliasing and achieve accurate modal decomposition; in addition, the artificial rabbit algorithm has stronger optimization ability and faster convergence speed than other optimization algorithms; in order to clarify the influencing factors of the non-stationarity of distributed photovoltaic power generation, it is proposed to use the improved kernel principal component analysis method to quantitatively analyze the influencing factors that cause the non-stationarity of photovoltaic power generation, namely, multi-source data such as temperature, humidity, light, and cloud cover. Compared with the traditional kernel principal component analysis method, the improved kernel principal component analysis method can effectively reduce the impact of noise data or abnormal data; on this basis, further, a multi-source data fusion method based on long short-term memory neural network is proposed, and the quantitative influence of multi-source data on the non-stationarity of distributed photovoltaic aggregate power generation is introduced into the neural network in the form of the initial weight setting of the forgetting gate, and then combined with the distributed photovoltaic aggregate power generation historical data and historical weather data introduced through the input gate, the accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation is realized.

[0109] In one embodiment, in order to better understand the distributed photovoltaic power generation power prediction method, Figure 3 A flow chart of another distributed photovoltaic power generation power prediction method provided by an embodiment of the present invention is as follows: Figure 3 As shown, the distributed photovoltaic power generation power prediction method specifically includes the following steps:

[0110] a1. Based on the initial cluster center selected from the historical power generation parameter information and the improved Kmeans++ clustering analysis method of the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information, the distributed photovoltaic aggregate power generation data and historical weather data are obtained, and the distributed photovoltaics belonging to the same cluster are aggregated according to a unified sample to obtain the target time series data of the distributed photovoltaic aggregate power generation power.

[0111] a2. Use the optimized variational mode decomposition method to perform modal decomposition on the target time series data. The optimized variational mode decomposition is mode 1, mode 2, ..., mode N.

[0112] a3. Analyze the non-stationary influence of multiple factors on distributed photovoltaic power generation according to modal feature comparison, that is, use at least two kernel functions in the preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each sub-modal component to obtain at least two target variable matrices, calculate the average value of at least two target variable matrices to obtain the corresponding new variable matrix, and determine the eigenvalues ​​of the new variable matrix and their corresponding orthogonal unit eigenvectors, determine the contribution of each eigenvalue according to a preset contribution formula and sort them according to the contribution, so as to select G eigenvalues ​​and their corresponding G orthogonal unit eigenvectors; extract the principal component components from the input variable matrix corresponding to each sub-modal component according to each orthogonal unit eigenvector, and determine the principal component components according to the preset similarity formula. The similarity between the component components and the historical power generation weather data in the historical power generation parameter information, the correspondence between the principal component components and the historical power generation weather data is determined based on the similarity, and the contribution of the historical power generation weather data to the distributed photovoltaic power generation power is determined based on the correspondence. It can be understood that the improved kernel principal component analysis method is used to obtain the contribution of weather factors, such as temperature, humidity, light, cloud cover and other data, to the non-stationarity of distributed photovoltaic aggregate power generation power, so as to benchmark and analyze the quantitative impact of multiple types of influencing factors on non-stationarity: by constructing the kernel principal component analysis input matrix, the principal component components (contribution) are obtained - that is, the quantitative impact of multi-source data of multiple types of factors on the non-stationarity of power generation is analyzed, and the quantitative impact degree is introduced into the neural network in the form of the initial weight of the forgetting gate.

[0113] a4. Input the historical power generation parameter information with label information into the long short-term memory neural network for model training, update the weight value of the model until the loss function reaches the minimum or reaches the preset maximum number of iterations to output the trained long short-term memory neural network model, and input the historical power generation parameter information into the pre-trained long short-term memory neural network to obtain the distributed photovoltaic power generation power prediction result.

[0114] The embodiment of the present invention deeply mines and fully considers the quantitative impact of multi-source data such as temperature, humidity, light intensity, cloud cover, etc. on the non-stationarity of photovoltaic power generation, and integrates multi-source data to improve the ultra-short-term prediction accuracy to 95%.

[0115] In one embodiment, to facilitate a better understanding of the architecture diagram of distributed photovoltaic power generation prediction, Figure 4 A schematic diagram of a distributed photovoltaic power generation power prediction architecture provided by an embodiment of the present invention is shown in FIG. Figure 4As shown in the figure, based on the initial clustering center selected from the historical power generation parameter information and the improved K-means++ algorithm of the Gaussian kernel distance between the initial clustering center and the historical power generation parameter information, a dynamic clustering partition model of distributed photovoltaic power generation is constructed, which can support the aggregation of distributed photovoltaics belonging to the same cluster cluster according to a unified specimen to obtain the time series data of distributed photovoltaic aggregated power generation power; the distributed photovoltaic aggregated power generation power obtained in step 1 is modally decomposed by using the optimized variational mode decomposition method; the contribution of weather factors such as temperature, humidity, light, cloud cover and other data to the non-stationarity of distributed photovoltaic aggregated power generation power is obtained by using the improved kernel principal component analysis method, so as to benchmark and analyze the quantitative influence of multiple types of influencing factors on the non-stationarity; the quantitative influence of multi-source data on the non-stationarity of distributed photovoltaic aggregated power generation is introduced into the neural network in the form of the initial weight of the forgetting gate by using the long short-term memory neural network, and then combined with the historical data of distributed photovoltaic aggregated power generation and historical weather data, a multi-source data fusion method is created, and the accurate prediction of ultra-short-term distributed photovoltaic aggregated power generation is realized.

[0116] In one embodiment, Figure 5 This is a structural block diagram of a distributed photovoltaic power generation prediction device provided by an embodiment of the present invention. The device is suitable for predicting ultra-short-term distributed photovoltaic power generation based on the consideration of non-stationary factors. The device can be implemented by hardware / software. It can be configured in an electronic device to implement a distributed photovoltaic power generation prediction processing method in an embodiment of the present invention.

[0117] like Figure 5 As shown, the device includes: a clustering module 510, a decomposition module 520, a contribution determination module 530 and a prediction module 540;

[0118] The clustering module 510 is used to obtain the historical power generation parameter information of the distributed photovoltaic power generation units in the target area within a preset time period, and dynamically cluster the historical power generation parameter information based on the initial cluster center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain the clustered target time series data;

[0119] A decomposition module 520 is used to perform variational modal decomposition on each of the target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor;

[0120] A contribution determination module 530 is used to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components using at least two kernel functions in a preset kernel principal component analysis method to obtain the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power;

[0121] The prediction module 540 is used to input the historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree.

[0122] In the embodiment of the present invention, the clustering module is used to dynamically cluster the historical power generation parameter information based on the initial cluster center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain the clustered target time series data. The problem that the time series characteristics of distributed power generation with different time and space distributions are very different and it is difficult to select a benchmark photovoltaic array for aggregate prediction can be considered. The present invention can process more complex nonlinear data and does not require random selection of K sample points in the initialization phase, which makes it more interpretable. The decomposition module is used to perform variational modal decomposition on each target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components. The optimized variational modal decomposition algorithm reduces the number of manual settings compared to the traditional variational modal decomposition algorithm. The influence of relevant parameters is reduced, the possibility of under-decomposition or over-decomposition is reduced, and modal aliasing is more effectively suppressed to achieve accurate modal decomposition; through the contribution determination module, the preset improved kernel principal component analysis method is used to perform high-dimensional feature space mapping on each sub-modal component to obtain the contribution of historical power generation weather data to distributed photovoltaic power generation power. This contribution is introduced into the neural network in the form of initial weight setting of the forgetting gate. Through the prediction module, the historical power generation parameter information is input into the long short-term memory neural network to obtain the distributed photovoltaic power generation power prediction result, which can have a clear understanding of the non-stationary inducing factors on the ultra-short time scale, thereby improving the stability and accuracy of photovoltaic power generation prediction, realizing the accurate prediction of ultra-short-term distributed photovoltaic aggregate power generation, and providing the premise guarantee for supporting the power grid to quickly adjust its power.

[0123] In one embodiment, the historical power generation parameter information includes: historical power generation data and historical power generation weather data; the historical power generation weather data at least includes: surface air pressure, temperature, humidity, cloud cover and light intensity.

[0124] In one embodiment, the clustering module 510 further includes:

[0125] a forming unit, configured to perform normalization processing on the historical power generation parameter information to obtain target historical power generation parameter information, and form the target historical power generation parameter information into a historical data set;

[0126] A selection unit, used to randomly select a historical sample data from the historical data set as an initial cluster center, and put the initial cluster center into a cluster center set;

[0127] The first division unit is used to determine the Gaussian kernel distance from other historical sample data in the historical data set to the initial cluster center, determine the probability of each historical sample data in the historical data set being the next cluster center according to the Gaussian kernel distance, and use the historical sample data with the largest probability as the next cluster center and store it in the cluster center set, repeat this step until the number of cluster center points in the cluster center set reaches the preset cluster center point number threshold, and obtain K cluster centers after division;

[0128] A second division unit is used to determine the Gaussian kernel distance from each historical sample data in the historical data set to the K cluster centers, and divide each of the historical sample data into the cluster center with the closest kernel distance according to the Gaussian kernel distance to form K cluster clusters;

[0129] The aggregation unit is used to aggregate the historical sample data belonging to the same cluster to obtain the corresponding target time series data.

[0130] In one embodiment, the historical data set is represented as: X = [x1, ..., x n ], where x i Represented as the i-th historical sample data;

[0131] The distance calculation formula of the Gaussian kernel distance is expressed as: Wherein, c represents the cluster center, Π(·,·) is the Gaussian kernel function, and the Gaussian kernel function is expressed by the formula: Wherein, a and b represent two historical sample data in the historical data set, respectively, ||ab|| 2 It is expressed as the Euclidean distance between a and b, and σ is the bandwidth;

[0132] The probability is expressed by the formula: in, It is expressed as the Gaussian kernel distance of the i-th historical sample data.

[0133] In one embodiment, the decomposition module 520 includes:

[0134] A parameter determination unit, used for determining the optimal decomposition parameters required for performing variational modal decomposition of the target time series data by using an artificial rabbit algorithm; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor;

[0135] A decomposition unit is used to perform variational modal decomposition on the target time series data according to the optimal decomposition parameters to obtain at least two sub-modal components.

[0136] In one embodiment, the parameter determination unit includes:

[0137] The first position determination subunit is used to randomly initialize the current position of the artificial rabbit; wherein the current position is represented by the initial decomposition parameter of the target time series data, and the initial decomposition parameter is represented by L i =[L i1 ,L i2 ], L i1 represents the number of sub-modes, L i2 represents the penalty factor;

[0138] The second position determination subunit is used to determine the value of the energy factor corresponding to the rabbit at the t-th iteration, and determine the candidate position of the rabbit at the next iteration according to the value;

[0139] A parameter determination subunit is used to determine the target fitness function of the candidate position and the current position, and update the rabbit position according to the target fitness function to obtain an updated target position, and return to the step of determining the value of the energy factor corresponding to the rabbit in the current iteration until the maximum number of iterations is reached or the target position with the minimum target fitness function is found, and the target position is used as the target sub-modal component number and target penalty factor of variational mode decomposition; wherein the target fitness function is expressed as the sum of the average envelope entropy and sample entropy of at least two sub-modal components.

[0140] In one embodiment, the target fitness function is expressed as: Among them, K represents the number of sub-modal components corresponding to the variational mode decomposition, E i represents the envelope entropy of the ith submodal component, B m (r) is represented as sample entropy, which represents the total number of time series with intervals less than r in the m dimension, and is expressed as: Wherein, N represents the total data length corresponding to each of the submodal components, m represents the data length selected from the total data length, r represents the similarity tolerance, X(i), X(j) represent the i-th and j-th submodal components respectively, and D[X(i), X(j)] represents the maximum difference between the corresponding elements of X(i), X(j);

[0141] The energy factor is expressed by the formula: Where t represents the tth iteration; T represents the maximum number of iterations, and f is a random number between (0, 1);

[0142] The target position is expressed by the formula: Among them, L i (t) represents the current position, Indicated as candidate location.

[0143] In one embodiment, the candidate position includes a first candidate position and a second candidate position, and accordingly, the second position determination subunit is specifically configured to:

[0144] When the energy factor is greater than 1, the candidate position of the i-th rabbit in the next iteration is determined to be the first candidate position; wherein the first candidate position is the candidate position that the i-th rabbit will choose to move to the j-th rabbit's position when foraging, and is expressed by the formula: Where i, j = 1, ..., n, i ≠ j, It is expressed as the ratio of the position of rabbit i moving to rabbit j; L i (t) represents the current position of the i-th rabbit at the t-th iteration, L j (t) represents the current position of the jth rabbit at the tth iteration; r1, r2 are both random numbers between (0, 1), round represents rounding, o is the mutation operator, which takes a value of 0 or 1, n1 is a random number that follows a normal distribution, and n is the number of rabbits;

[0145] When the value of the energy factor is less than or equal to 1, the candidate position of the rabbit in the next iteration is determined to be the second candidate position; wherein the second candidate position is the candidate position when the i-th rabbit randomly selects a cave to hide, which is expressed by the formula: in, It is represented as the location of the cave that the i-th rabbit will randomly choose; Represented as rabbit i heading to cave B i The ratio of the moved position; r3 and r4 are random numbers between (0, 1), o and g are mutation operators, and their values ​​are 0 or 1.

[0146] In one embodiment, the decomposition unit comprises:

[0147] The first problem conversion subunit is used to convert the variational mode decomposition problem of the target time series data into a variational function problem containing constraints; wherein the variational function problem is expressed by the formula: Among them, {u k}:={u1,...,u K} represents K sub-modal components, {ω k}:={ω1,...,ω K} represents the center frequency of each submodal component, K represents the number of submodal components, x(t) represents the target time series data, δ(t) is the pulse function, represents the partial derivative with respect to time t, * represents the convolution calculation; u k (t) is represented by the kth submodal component at time t, which is expressed as u k (t) = A k (t)cos(φ k (t)), where A k (t) is u k (t) envelope amplitude; φ k (t) is expressed as phase, φ k ′(t)≥0; Expressed as a one-sided spectrum function; It is expressed as baseband; Represented as a preset estimated center frequency;

[0148] The second problem conversion subunit is used to convert the variational function problem into an unconstrained variational function problem by using a Lagrange multiplication operator and a target penalty factor; wherein the unconstrained variational problem is expressed by a formula: Among them, ||·||2 represents the L2 norm, <·> represents the inner product, λ represents the Lagrange multiplication operator, and α represents the target penalty factor;

[0149] The updating subunit is used to iteratively update the Lagrangian multiplication operator, the center frequency and the finite bandwidth and the desired submodal component by using the alternating direction multiplier algorithm to obtain the saddle point, and terminate the iterative process when the stop condition is met to obtain K desired submodal components; wherein the update of the Lagrangian multiplication operator is expressed by the formula: Wherein, ρ represents the learning rate; the update of the center frequency and the limited bandwidth is expressed by the formula: The update of the desired submodal component is expressed by the formula: Among them, x(ω), u i (ω), λ(ω) are the Parseval Fourier transforms of their respective variables; the stopping condition is expressed by the formula: Among them, ε represents the iterative convergence value.

[0150] In one embodiment, the contribution determination module 530 further includes:

[0151] A determination unit, used to determine an input variable matrix corresponding to each of the sub-modal components; wherein the input variable matrix includes envelope amplitude data of M time dimensions;

[0152] A mapping unit, used for performing high-dimensional feature space mapping on the input variable matrix using at least two kernel functions to obtain at least two target variable matrices;

[0153] Calculating the average value of the at least two target variable matrices to obtain a corresponding new variable matrix, and determining the eigenvalues ​​of the new variable matrix and the corresponding orthogonal unit eigenvectors thereof;

[0154] A sorting unit, used to determine the contribution of each eigenvalue according to a preset contribution formula and sort them according to the contribution to select G eigenvalues ​​and their corresponding G orthogonal unit eigenvectors;

[0155] An extraction unit, used for extracting principal component components from the input variable matrix corresponding to each of the sub-modal components according to each of the orthogonal unit eigenvectors;

[0156] A contribution determination unit is used to determine the similarity between the principal component and the historical power generation weather data in the historical power generation parameter information according to a preset similarity formula, determine the correspondence between the principal component and the historical power generation weather data according to the similarity, and determine the contribution of the historical power generation weather data to the distributed photovoltaic power generation power according to the correspondence.

[0157] In one embodiment, the number of submodal components is K; each submodal component includes envelope amplitude data of M time dimensions, and the amplitude of the jth frequency band in the i-th time period is represented by A ij , the input variable matrix is ​​expressed by the formula:

[0158]

[0159] The kernel function is represented by Φ(·), and the target variable matrix is ​​represented by: Φ1(A K×M ) D×M ,Φ2(A K×M ) D×M ,…Φ S (A K×M ) D×M , where (A K×M ) D×M Represented as the target variable matrix after mapping into high-dimensional feature space,

[0160] The new variable matrix Φ'(A K×M ) D×M , expressed as:

[0161]

[0162] The preset contribution is expressed by the formula:

[0163] The principal component formula is expressed as:

[0164] The preset similarity is expressed by the formula: Wherein, cov(Z,Y) represents the covariance between the historical power generation weather data and the principal component, σ Z represents the standard deviation of the historical power generation weather data, σ Y represents the standard deviation of the principal component, and E represents the expected value; when β is closer to 1, the similarity between the historical power generation weather data and the principal component is higher.

[0165] In one embodiment, each memory unit in the long short-term memory neural network includes a forget gate f t , input gate i t , output gate o t , the input of the memory unit at time t is x t , the output is h t , the state of the memory cell is c t ; where X=[x1,...x t ,...,x n ], represented as the historical power generation parameter information; the prediction module 540 further includes:

[0166] An input unit is used to input the historical power generation parameter information x t Input to the memory unit in the long short-term memory neural network, each input x t The forget gate f of the corresponding memory unit t , input gate i t , output gate o t ; Wherein, the forget gate f t Expressed as: f t =σ(W f h t-1 +U f x t +b f ), the input gate i t Expressed as: i t =σ(W i h t-1 +U i x t +b i ), the output gate o t Expressed as: o t =σ(W o h t-1 +U o x t +b o ), where σ represents the activation function, b f ,b i ,b o is the bias term, h t-1is the output of the memory unit at time t-1, W f ,U f ,W i ,U i ,W o ,U o are all weight matrices, where W f ,W i ,U i ,W o ,U o The weight matrix is ​​randomly initialized weights, and the forget gate U f The initial weight value of is determined by the contribution, and the forget gate U f Expressed as: U f =[0,ρ1,ρ2,…,ρ G ];

[0167] State determination unit, used to calculate the state c of the memory unit at each moment t ; wherein the state c t It is expressed as: Among them, c t-1 is the state of the memory unit at time t-1, ⊙ is the Harmand product operator, is the candidate state of the memory unit at time t, and the candidate state is expressed by the formula: Among them, tanh() is the activation function, W c and U c Expressed as a weight matrix, b c is the bias term;

[0168] Output unit, used to calculate the output h of the memory unit at each moment t , will output h t The output y of the long short-term memory neural network is recorded as the prediction result of distributed photovoltaic power generation; wherein the output h t Expressed as: h t =o t ⊙tanh(c t ).

[0169] The distributed photovoltaic power generation power prediction and processing device provided in the embodiment of the present invention can execute the distributed photovoltaic power generation power prediction and processing method applied to the financial system provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0170] In one embodiment, Figure 6A schematic diagram of the structure of an electronic device provided for an embodiment of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0171] like Figure 6 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0172] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0173] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as a distributed photovoltaic power generation power prediction method.

[0174] In some embodiments, the distributed photovoltaic power generation power prediction processing method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the distributed photovoltaic power generation power prediction method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the distributed photovoltaic power generation power prediction method in any other appropriate manner (for example, by means of firmware).

[0175] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0176] The computer program for implementing the method of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer or other programmable distributed photovoltaic power generation prediction device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program can be executed entirely on the machine, partially on the machine, partially on the machine as an independent software package and partially on a remote machine, or completely on a remote machine or server.

[0177] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0178] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0179] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0180] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0181] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0182] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A distributed photovoltaic power generation power prediction method, characterized in that: The method comprises: Acquire historical power generation parameter information of distributed photovoltaic power generation units in a target area, and dynamically cluster the historical power generation parameter information based on an initial cluster center selected from the historical power generation parameter information and a Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain clustered target time series data; Based on the optimal decomposition parameters corresponding to the target time series data, each of the target time series data is subjected to variational modal decomposition to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor; Using at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components, the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power is obtained; Inputting the historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree; The method of dynamically clustering the historical power generation parameter information based on the initial cluster center selected from the historical power generation parameter information and the Gaussian kernel distance between the initial cluster center and the historical power generation parameter information to obtain the clustered target time series data further includes: Normalizing the historical power generation parameter information to obtain target historical power generation parameter information, and forming the target historical power generation parameter information into a historical data set; Randomly selecting a historical sample data from the historical data set as an initial cluster center, and placing the initial cluster center into a cluster center set; Determine the Gaussian kernel distance from other historical sample data in the historical data set to the initial cluster center, determine the probability of each historical sample data in the historical data set being the next cluster center based on the Gaussian kernel distance, and use the historical sample data with the largest probability as the next cluster center and store it in the cluster center set, repeat this step until the number of cluster center points in the cluster center set reaches a preset cluster center point number threshold, and obtain K cluster centers after division; Determine the Gaussian kernel distance from each historical sample data in the historical data set to the K cluster centers, and divide each of the historical sample data into cluster centers with the closest kernel distance according to the Gaussian kernel distance to form K cluster clusters; Aggregate the historical sample data belonging to the same cluster to obtain the corresponding target time series data; Wherein, performing variational modal decomposition on each of the target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components further includes: Determine the optimal decomposition parameters required for variational modal decomposition of the target time series data using an artificial rabbit algorithm; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor; Performing variational modal decomposition on the target time series data according to the optimal decomposition parameters to obtain at least two sub-modal components; Among them, the use of at least two kernel functions in the preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components to obtain the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power further includes: Determine an input variable matrix corresponding to each of the submodal components; wherein the input variable matrix includes envelope amplitude data of M time dimensions; Using at least two kernel functions to perform high-dimensional feature space mapping on the input variable matrix to obtain at least two target variable matrices; Calculating the average value of the at least two target variable matrices to obtain a corresponding new variable matrix, and determining the eigenvalues ​​of the new variable matrix and the corresponding orthogonal unit eigenvectors thereof; Determine the contribution of each eigenvalue according to a preset contribution formula and sort them according to the contribution to select G eigenvalues ​​and their corresponding G orthogonal unit eigenvectors; Extracting principal component components from the input variable matrix corresponding to each of the sub-modal components according to each of the orthogonal unit eigenvectors; Determine the similarity between the principal component and the historical power generation weather data in the historical power generation parameter information according to a preset similarity formula, determine the corresponding relationship between the principal component and the historical power generation weather data according to the similarity, and determine the contribution of the historical power generation weather data to the distributed photovoltaic power generation power according to the corresponding relationship; The preset similarity is expressed by the formula: Wherein, cov(Z,Y) represents the covariance between the historical power generation weather data and the principal component, σ Z represents the standard deviation of the historical power generation weather data, σ Y represents the standard deviation of the principal component, and E represents the expected value; when β is closer to 1, the similarity between the historical power generation weather data and the principal component is higher.

2. The method according to claim 1, characterized in that The historical power generation parameter information includes: historical power generation data and historical power generation weather data; the historical power generation weather data includes: surface air pressure, temperature, humidity, cloud cover and light intensity.

3. The method according to claim 1, characterized in that The historical data set is represented as: X = [x1, ..., x n ], where x i It is represented as the i-th historical sample data, and n represents the total number of samples in the historical data set; The distance calculation formula of the Gaussian kernel distance is expressed as: Wherein, c represents the cluster center, Π(·,·) is the Gaussian kernel function, and the Gaussian kernel function is expressed by the formula: Wherein, a and b represent two historical sample data in the historical data set, respectively, ||ab|| 2 It is expressed as the Euclidean distance between a and b, and σ is the bandwidth; The probability is expressed by the formula: in, It is expressed as the Gaussian kernel distance of the i-th historical sample data.

4. The method according to claim 1, characterized in that The method of using the artificial rabbit algorithm to determine the optimal decomposition parameters required for performing variational mode decomposition on the target time series data further includes: Randomly initialize the current position of the artificial rabbit; wherein the current position is represented as the initial decomposition parameter of the target time series data, and the initial decomposition parameter is represented as L i =[L i1 ,L i2 ], L i1 represents the number of sub-modes, L i2 represents the penalty factor; Determine the value of the energy factor corresponding to the rabbit at the t-th iteration, and determine the candidate position of the rabbit at the next iteration based on the value; Determine the target fitness function of the candidate position and the current position, and update the rabbit position according to the target fitness function to obtain the updated target position, return to the step of determining the value of the energy factor corresponding to the rabbit in the current iteration, until the maximum number of iterations is reached or the target position with the minimum target fitness function is found, and use the target position as the target sub-modal component number and target penalty factor of variational mode decomposition; wherein the target fitness function is expressed as the sum of the average envelope entropy and sample entropy of at least two sub-modal components.

5. The method according to claim 4, characterized in that The target fitness function is expressed as follows: Among them, K represents the number of sub-modal components corresponding to the variational mode decomposition, E i represents the envelope entropy of the ith submodal component, B m (r) is represented as sample entropy, which represents the total number of time series with intervals less than r in the m dimension, and is expressed as: Wherein, N represents the total data length corresponding to each of the submodal components, m represents the data length selected from the total data length, r represents the similarity tolerance, X(i), X(j) represent the i-th and j-th submodal components respectively, and D[X(i), X(j)] represents the maximum difference between the corresponding elements of X(i), X(j); The energy factor is expressed by the formula: Where t represents the tth iteration; T represents the maximum number of iterations, and f is a random number between (0, 1); The target position is expressed by the formula: Among them, L i (t) represents the current position, Indicated as candidate location.

6. The method according to any one of claims 4 or 5, characterized in that: The candidate positions include a first candidate position and a second candidate position; accordingly, determining the candidate position of the rabbit for the next iteration according to the value situation further includes: When the energy factor is greater than 1, the candidate position of the i-th rabbit in the next iteration is determined to be the first candidate position; wherein the first candidate position is the candidate position that the i-th rabbit will choose to move to the j-th rabbit's position when foraging, which is expressed by the formula: Where i, j = 1, ..., n, i ≠ j, It is expressed as the ratio of the position of rabbit i moving to rabbit j; L i (t) represents the current position of the i-th rabbit at the t-th iteration, L j (t) represents the current position of the jth rabbit at the tth iteration; r1, r2 are both random numbers between (0, 1), round represents rounding, o is the mutation operator, which takes a value of 0 or 1, n1 is a random number that follows a normal distribution, and n is the number of rabbits; When the energy factor value is less than or equal to 1, the candidate position of the rabbit in the next iteration is determined to be the second candidate position; wherein the second candidate position is the candidate position when the i-th rabbit randomly selects a cave to hide, which is expressed by the formula: in, It represents the location of the cave that the i-th rabbit will randomly choose; Represented as rabbit i heading to cave B i The ratio of the moved position; r3 and r4 are random numbers between (0, 1), o and g are mutation operators, and their values ​​are 0 or 1.

7. The method according to claim 1, characterized in that The step of performing variational modal decomposition on the target time series data according to the optimal decomposition parameter to obtain at least two sub-modal components further includes: The variational mode decomposition problem of the target time series data is converted into a variational function problem containing constraints; wherein the variational function problem is expressed by the formula: Among them, {u k }:={u1,...,u K } represents K sub-modal components, {ω k }:={ω1,...,ω K } represents the center frequency of each submodal component, K represents the number of submodal components, x(t) represents the target time series data, δ(t) is the pulse function, represents the partial derivative with respect to time t, * represents the convolution calculation; u k (t) is represented by the kth submodal component at time t, which is expressed as u k (t) = A k (t)cos(φ k (t)), where A k (t) is u k (t) envelope amplitude; φ k (t) is expressed as phase, φ k ′(t)≥0; Expressed as a one-sided spectrum function; It is expressed as baseband; Represented as a preset estimated center frequency; The variational function problem is converted into an unconstrained variational function problem by using a Lagrange multiplication operator and a target penalty factor; wherein the unconstrained variational function problem is expressed by a formula: Among them, ||·||2 represents the L2 norm, <·> represents the inner product, λ represents the Lagrange multiplication operator, and α represents the target penalty factor; The alternating direction multiplier algorithm is used to iteratively update the Lagrangian multiplication operator, the center frequency and the finite bandwidth, and the desired submodal component to obtain the saddle point, and the iterative process is terminated when the stop condition is met to obtain K desired submodal components; wherein the update of the Lagrangian multiplication operator is expressed by the formula: Wherein, ρ represents the learning rate; the update of the center frequency and the limited bandwidth is expressed by the formula: The update of the desired submodal component is expressed by the formula: Among them, x(ω), u i (ω), λ(ω) are the Parseval Fourier transforms of their respective variables; the stopping condition is expressed by the formula: Among them, ε represents the iterative convergence value.

8. The method according to claim 1, characterized in that: The number of submodal components is K; each submodal component includes envelope amplitude data of M time dimensions. Let the amplitude of the jth frequency band in the i-th time period be expressed as A ij , the input variable matrix is ​​expressed by the formula: The kernel function is represented by Φ(·), and the target variable matrix is ​​represented by: Φ1(A K×M ) D×M ,Φ2(A K×M ) D×M ,…Φ S (A K×M ) D×M , where (A K×M ) D×M Represented as the target variable matrix after mapping into high-dimensional feature space, The new variable matrix Φ'(A K×M ) D×M , expressed as: The preset contribution is expressed by the formula: The principal component formula is expressed as:

9. The method according to claim 1, characterized in that: Each memory unit in the long short-term memory neural network includes a forget gate f t , input gate i t , output gate o t , the input of the memory unit at time t is x t , the output is h t , the state of the memory cell is c t ; Where X=[x1,...x t ,...,x n ], represented as the historical power generation parameter information; accordingly, the historical power generation parameter information is input into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result, further comprising: The historical power generation parameter information x t Input to the memory unit in the long short-term memory neural network, each input x t The forget gate f of the corresponding memory unit t , input gate i t , output gate o t ; Wherein, the forget gate f t Expressed as: f t =σ(W f h t-1 +U f x t +b f ), the input gate i t Expressed as: i t =σ(W i h t-1 +U i x t +b i ), the output gate o t Expressed as: o t =σ(W o h t-1 +U o x t +b o ), where σ represents the activation function, b f ,b i ,b o is the bias term, h t-1 is the output of the memory unit at time t-1, W f ,U f ,W i ,U i ,W o ,U o are all weight matrices, where W f ,W i ,U i ,W o ,U o The weight matrix is ​​randomly initialized weights, and the forget gate U f The initial weight value of is determined by the contribution, and the forget gate U f Expressed as: U f =[0,ρ1,ρ2,…,ρ G ]; Calculate the state c of the memory unit at each moment t ; wherein the state c t It is expressed as: Among them, c t-1 is the state of the memory unit at time t-1, ⊙ is the Harmand product operator, is the candidate state of the memory unit at time t, and the candidate state is expressed by the formula: Among them, tanh() is the activation function, W c and U c Expressed as a weight matrix, b c is the bias term; Calculate the output h of the memory unit at each moment t , will output h t The output y of the long short-term memory neural network is recorded as the prediction result of distributed photovoltaic power generation; wherein the output h t Expressed as: h t =o t ☉tanh(c t ).

10. A distributed photovoltaic power generation power prediction device, characterized in that: The device comprises: A clustering module is used to obtain historical power generation parameter information of distributed photovoltaic power generation units in a target area within a preset time period, and dynamically cluster the historical power generation parameter information based on an initial clustering center selected from the historical power generation parameter information and a Gaussian kernel distance between the initial clustering center and the historical power generation parameter information to obtain clustered target time series data; A decomposition module, configured to perform variational modal decomposition on each of the target time series data based on the optimal decomposition parameters corresponding to the target time series data to obtain at least two sub-modal components; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor; A contribution determination module, used to use at least two kernel functions in a preset kernel principal component analysis method to perform high-dimensional feature space mapping on the input variable matrix corresponding to each of the sub-modal components to obtain the contribution of the historical power generation weather data in the historical power generation parameter information to the distributed photovoltaic power generation power; A prediction module, used for inputting the historical power generation parameter information into a pre-trained long short-term memory neural network to obtain a distributed photovoltaic power generation power prediction result; wherein the initial weight value of the forget gate in the long short-term memory neural network is determined by the contribution degree; Wherein, the clustering module further comprises: a forming unit, configured to perform normalization processing on the historical power generation parameter information to obtain target historical power generation parameter information, and form the target historical power generation parameter information into a historical data set; A selection unit, used to randomly select a historical sample data from the historical data set as an initial cluster center, and put the initial cluster center into a cluster center set; The first division unit is used to determine the Gaussian kernel distance from other historical sample data in the historical data set to the initial cluster center, determine the probability of each historical sample data in the historical data set being the next cluster center according to the Gaussian kernel distance, and use the historical sample data with the largest probability as the next cluster center and store it in the cluster center set, repeat this step until the number of cluster center points in the cluster center set reaches the preset cluster center point number threshold, and obtain K cluster centers after division; A second division unit is used to determine the Gaussian kernel distance from each historical sample data in the historical data set to the K cluster centers, and divide each of the historical sample data into the cluster center with the closest kernel distance according to the Gaussian kernel distance to form K cluster clusters; Aggregation unit, used to aggregate historical sample data belonging to the same cluster to obtain corresponding target time series data; Wherein, the decomposition module includes: A parameter determination unit, used for determining the optimal decomposition parameters required for performing variational modal decomposition of the target time series data by using an artificial rabbit algorithm; wherein the optimal decomposition parameters include: the number of target sub-modal components and a target penalty factor; A decomposition unit, configured to perform variational modal decomposition on the target time series data according to the optimal decomposition parameters to obtain at least two sub-modal components; Wherein, the contribution determination module further comprises: A determination unit, used to determine an input variable matrix corresponding to each of the sub-modal components; wherein the input variable matrix includes envelope amplitude data of M time dimensions; A mapping unit, used for performing high-dimensional feature space mapping on the input variable matrix using at least two kernel functions to obtain at least two target variable matrices; Calculating the average value of the at least two target variable matrices to obtain a corresponding new variable matrix, and determining the eigenvalues ​​of the new variable matrix and the corresponding orthogonal unit eigenvectors thereof; A sorting unit, used to determine the contribution of each eigenvalue according to a preset contribution formula and sort them according to the contribution to select G eigenvalues ​​and their corresponding G orthogonal unit eigenvectors; An extraction unit, used for extracting principal component components from the input variable matrix corresponding to each of the sub-modal components according to each of the orthogonal unit eigenvectors; A contribution determination unit, configured to determine the similarity between the principal component and the historical power generation weather data in the historical power generation parameter information according to a preset similarity formula, determine the correspondence between the principal component and the historical power generation weather data according to the similarity, and determine the contribution of the historical power generation weather data to the distributed photovoltaic power generation power according to the correspondence; The preset similarity is expressed by the formula: Wherein, cov(Z,Y) represents the covariance between the historical power generation weather data and the principal component, σ Z represents the standard deviation of the historical power generation weather data, σ Y represents the standard deviation of the principal component, and E represents the expected value; when β is closer to 1, the similarity between the historical power generation weather data and the principal component is higher.

11. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the distributed photovoltaic power generation power prediction method described in any one of claims 1-9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the distributed photovoltaic power generation power prediction method described in any one of claims 1-9 when executed.

Images