Energy storage station system of home substation

Abstract

The invention discloses an energy storage station system of a home substation. A household transformer station energy storage station system comprises a wind power module used for generating power by using wind power; the wind power collection module is used for collecting historical wind power data of the wind power module; the photoelectric module is used for generating power by utilizing solar energy; the photoelectric collection module is used for collecting historical photoelectric data of the photoelectric module; the weather module is used for collecting weather information of the location; the energy storage module is used for storing electric energy; the wind-light-electricity prediction device is used for predicting the future wind power generation amount based on the historical wind power data and the weather information and predicting the future photoelectric power generation amount based on the historical photoelectric data and the weather information; and the grid-connected control module is used for grid-connected power transmission based on future photoelectric generating capacity and residual energy storage of the energy storage module. According to the scheme provided by the invention, the prediction accuracy is higher, so that when the household energy storage station is connected with the power grid, the input or output electric energy is high in regularity, and the utilization of clean energy is facilitated.

Description

Technical Field

[0001] The present application relates to the technical field of substation management, and in particular to a household substation energy storage station system. Background Art

[0002] Small household energy storage stations are an important source of energy that can fully utilize local wind energy and electricity to ease power supply pressure. However, the construction cost of small household energy storage stations is low, and the capacity of the energy storage batteries in household energy storage stations is low. Therefore, during the operation of household energy storage stations, only by accurately predicting the power generation of energy storage stations in the future time period can they operate at the maximum efficiency, so as to avoid large errors in the prediction of regional power shortages by the national grid due to prediction errors, which in turn makes power supply management difficult.

[0003] At present, household energy storage stations are all wind power and photovoltaic integrated power stations. In a region, the photovoltaic power generation of each user is basically the same, but it will also be affected by aging, and the wind power of each user is affected by various factors, and wind power has greater randomness. The main reason for such a large error in wind power is that the prediction basis of wind power comes from local climate conditions, but the local climate conditions are the climate conditions of an administrative division and have no guiding significance for independent wind power. For this reason, neither wind power nor photovoltaic power can be accurately predicted. Summary of the invention

[0004] The content of this application is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this application is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0005] In order to solve the technical problem that a small household energy storage station cannot accurately predict the power generation, some embodiments of the present application provide a household substation energy storage station system including:

[0006] Wind power module, used to generate electricity using wind power;

[0007] A wind power collection module is used to collect historical wind power data of the wind power module;

[0008] Photovoltaic modules, used to generate electricity from solar energy;

[0009] A photoelectric collection module, used to collect historical photoelectric data of the photoelectric module;

[0010] Weather module, used to collect local weather information;

[0011] Energy storage module, used to store electrical energy;

[0012] Wind and photovoltaic power prediction device, which predicts future wind power generation based on historical wind power data and weather information, and predicts future photovoltaic power generation based on historical photovoltaic data and weather information;

[0013] The grid-connected control module determines the grid-connected transmission capacity based on the future photovoltaic power generation and the remaining energy storage of the energy storage module.

[0014] In the technical solution provided by the present application, each household substation energy storage system uses the information collected by itself to predict future power generation. Compared with the method of using weather information in a region to predict the power generation of all energy storage stations, the solution provided by the present application has higher prediction accuracy. Therefore, when the household energy storage station is connected to the power grid, the input or output of electric energy is highly regular, which facilitates the use of clean energy.

[0015] Furthermore, the wind and photovoltaic prediction device includes a wind power prediction module and a photovoltaic prediction module. The wind power prediction module is used to predict future wind power generation, and the photovoltaic prediction module is used to predict future photovoltaic power generation.

[0016] Further, the historical wind power data includes historical power output data;

[0017] Weather information includes historical wind speed, historical wind direction, historical temperature, historical humidity, and historical air pressure.

[0018] When predicting wind power generation, there is a large degree of randomness, but there is also a large degree of seasonality. Therefore, if the weather information of different seasons is used as the same type of data for wind power generation prediction, the model will be overfitted because the data itself lacks actual rules. Therefore, this application provides the following technical solutions:

[0019] Furthermore, the historical wind speed, historical wind direction, historical temperature, historical humidity and historical air pressure are data change curves for 24 hours a day within a preset date each year;

[0020] For each item of weather information, it is organized into an information matrix with hours in the horizontal direction and dates in the vertical direction.

[0021] In the technical solution provided by the present application, for weather changes in different seasons, information within a preset date is collected as a prediction standard, so when making predictions, the impact of different climate change trends in other seasons on prediction accuracy can be removed.

[0022] When predicting wind power generation and photovoltaic power generation, it is necessary to improve the accuracy as much as possible, and the improvement of accuracy depends on the complexity of the model. Large neural network models have stronger learning capabilities than small neural network models, but setting up a large neural network model in each energy storage power station and training it will result in high costs. To this end, this application provides the following technical solutions:

[0023] The wind and solar power prediction device selects a trained neural network model for power generation prediction based on data similarity.

[0024] In the technical solution provided in the present application, in order to reduce the training cost of the neural network model, multiple neural network models are pre-trained with different types of data, and then the subsequent energy storage power station selects the trained neural network model based on the similarity of the information collected by itself. Therefore, the technical solution provided in the present application can reduce the model training cost.

[0025] Furthermore, all samples H collected by the wind and photovoltaic prediction devices are collected in advance. i For sample H i The classification is carried out in the following way;

[0026] S1: For H i ={x, y} for normalization, where x is the predicted information and y is the actual power generation information;

[0027] S2: Randomly select a feature from x as the separation value, and put all samples H i The values ​​of x are divided into two subsets, the ones greater than the eigenvalue are the right subset, and the ones less than the eigenvalue are the left subset;

[0028] S3: Repeat the above S2 for the samples in the right subset and the left subset, and randomly generate a new separation value each time;

[0029] S4: Continue to segment the samples until any of the following conditions are met;

[0030] The number of splits has reached the preset maximum value;

[0031] There is only one sample in the subset;

[0032] All samples in the subset have the same feature value;

[0033] S5: Generate an isolated tree each time the stopping condition is reached, and divide the sample multiple times to obtain multiple isolated trees;

[0034] For each sample H i , traverse all isolated trees and record sample H i The number of edges passed when reaching a leaf node to obtain sample Hi The height h of the sample point i ;

[0035] Calculate H for each sample i The average value of the height E i ;

[0036] Calculate H for each sample i The density S;

[0037]

[0038] Where c(n) represents the average path length between the separation point and the group, E i Represents sample H i The average value of the height, S i Represents sample H i The sparsity of n represents the total number of samples;

[0039] S6: For all samples H i The density S generates an n×n matrix R, where each element in the matrix R is S i and S j The similarity distance D i,j ,,i and j are the indexes of the samples, i≠j;

[0040]

[0041] S7: Use clustering algorithm to divide the matrix R into several groups, based on the samples H in each group. i The source of the data is used to determine the classification basis of wind and solar power prediction devices.

[0042] In the technical solution provided in the present application, all the data that can be collected by the wind and solar power prediction device are first subjected to isolation tree calculation to obtain sparsity, and then a sparsity matrix D is generated based on the sparsity, and classification is performed based on the sparsity matrix D. In the whole process, the characteristics between samples can be effectively found, the data can be accurately classified, and the clustering effect between the data can be clearly highlighted.

[0043] Large neural network models have sufficient model accuracy and stronger learning ability. When sufficient samples are input, the prediction accuracy of large neural networks will continue to increase. However, large neural network models have large computational complexity and slow iteration speed, and are not suitable for home energy storage stations. To this end, this application provides the following technical solutions:

[0044] The wind power prediction module and the photovoltaic prediction module are respectively equipped with pre-trained prediction models;

[0045] The prediction model is trained in a training device;

[0046] The training device comprises:

[0047] Knowledge output unit, with a pre-trained large neural network model built in;

[0048] A training unit, in which the prediction model to be trained is arranged;

[0049] The training data are input into the knowledge output unit and the training unit respectively, wherein the knowledge output unit extracts information from the training data to guide the training of the built-in prediction model of the training unit.

[0050] In the technical solution provided in the present application, the prediction model is not a large neural network model, but a small neural network model. However, in order to achieve the accuracy of the large neural network model, the large neural network model is used to guide the training of the small prediction model. Therefore, in practice, it can ensure that the prediction model has sufficient accuracy on the one hand, but when in use, the model has a small amount of calculation and high prediction efficiency.

[0051] Furthermore, for sample H i It is divided into photovoltaic prediction samples and wind power prediction samples. The photovoltaic prediction samples and wind power prediction samples are used for photovoltaic power generation prediction and wind power generation prediction respectively.

[0052] When predicting wind power generation, it is necessary to use weather information for comprehensive prediction, thereby capturing the changes in weather and the impact of weather changes on wind power generation. To this end, the prediction information x is mainly weather information, and the wind direction, temperature, and humidity in the weather information have different effects on power generation. To this end, the present application provides the following technical solutions:

[0053] Furthermore, the normalization method in S1 is as follows: all wind speeds, wind directions, humidity and air pressure are initially normalized, and then corresponding weight coefficients are set for each wind speed, wind direction, humidity and air pressure, and x is obtained by adding them up;

[0054] x=α1A1+α2A2+α3A3+α4A4; wherein α1, α2, α3, α4 are set weight coefficients.

[0055] In the technical solution provided in the present application, when calculating x, the corresponding weight coefficient is pre-set, and the influence of each piece of information is controlled according to the weight coefficient, so that the final obtained x can represent effective information.

[0056] Weather information is crucial when predicting wind power generation, but there is obvious redundancy in weather information, and the impact between information is very complex. It is actually impossible to reasonably give a weight coefficient for each weather information, and then use this weight coefficient to accurately describe the prediction information x. To this end, this application provides the following technical solutions:

[0057] Furthermore, the weight coefficient α of each item of weather information is calculated as follows:

[0058] α u =C r (u,U)P;

[0059] Where P is the conversion factor, C r (u,U) is the correlation ratio between the u-th information and the rest of the information;

[0060]

[0061]

[0062] Among them, u represents the uth information, U represents all the information except u, C represents the power generation, I r (r,U) is the correlation coupling rate, which measures the degree of additional information sharing between u and C when all information except u is known, and its value range is 0 to 1. I(u;C|U) represents the amount of information sharing between u and C when U is known. H(u) represents the entropy of u, H(C) represents the entropy of C, and T r (A, B) represents the reduction of information sharing between u and C when U is known, and I(u; C) represents the mutual information between u and C;

[0063] The entropy is calculated as follows:

[0064]

[0065] Where H(u) represents the entropy of the u-th information in the weather information, i represents the index of the sample, n represents the total number of samples, b = 2, p(u i ) indicates that u takes the value u i probability.

[0066] In the technical solution provided in the present application, when calculating the weight coefficient α, it is converted according to the correlation ratio between each weather information and other weather information. Therefore, when converting the weight coefficient, the intrinsic connection between each piece of information can be accurately measured to increase the accuracy of the weight coefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] The drawings constituting a part of this application are used to provide a further understanding of this application, so that other features, purposes and advantages of this application become more obvious. The illustrative embodiment drawings and their descriptions of this application are used to explain this application and do not constitute an improper limitation on this application.

[0068] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that the components and elements are not necessarily drawn to scale.

[0069] In the attached picture:

[0070] Figure 1 This is the structure of the household substation energy storage station system.

[0071] Figure 2 is the temperature information matrix.

[0072] Figure 3 This is the network structure diagram during prediction model training. DETAILED DESCRIPTION

[0073] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are only for exemplary purposes and are not intended to limit the scope of protection of the present application.

[0074] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other.

[0075] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments.

[0076] Reference Figure 1 The household substation energy storage system is a household energy storage system, which is generally used in areas with abundant wind power and photovoltaic power. After the household installs the system, it can basically achieve self-sufficiency in electricity, and then transmit the excess electricity to the state grid. In order to ensure the quality of electricity, it is necessary to accurately predict the total amount of power generation so that when transmitting or receiving electricity to the state grid, the required amount of electricity can be sent in advance, which is convenient for the grid to dispatch.

[0077] The energy storage station system of the household substation includes: wind power module, photovoltaic module, energy storage module, and grid-connected control module. Among them, the wind power module is used to generate electricity using wind power; the photovoltaic collection module is used to collect historical photovoltaic data of the photovoltaic module; the energy storage module is used to store electric energy; the grid-connected control module is based on the future photovoltaic power generation and the remaining energy storage of the energy storage module, and the grid-connected transmission capacity. Among them, the wind power module, photovoltaic module, energy storage module, and grid-connected control module are the main energy storage station systems, and their specific power generation methods, energy storage methods, and grid-connected scheduling methods are not repeated here. In addition, the grid-connected control module flexibly adjusts the operation strategy to improve energy utilization based on the respective characteristics and usage conditions of solar energy, wind energy, and energy storage systems, while ensuring the life of each module, and gives priority to ensuring a stable supply of electricity, and the surplus electricity can be connected to the Internet for profit. The specific adjustment strategy can be adjusted according to the actual situation, and the key lies in accurately predicting the power generation.

[0078] Specifically, the household substation energy storage station system also includes: a wind power collection module, a photovoltaic collection module, a weather module, and a wind and photovoltaic prediction device; the wind power collection module is used to collect historical wind power data of the wind power module; the photovoltaic collection module is used to collect historical photovoltaic data of the photovoltaic module; the weather module is used to collect local weather information; the wind and photovoltaic prediction device predicts future wind power generation based on historical wind power data and weather information, and predicts future photovoltaic power generation based on historical photovoltaic data and weather information;

[0079] The wind and photovoltaic prediction device comprises a wind power prediction module and a photovoltaic prediction module. The wind power prediction module is used to predict future wind power generation, and the photovoltaic prediction module is used to predict future photovoltaic power generation.

[0080] It should be noted that the weather module in this solution is installed at the location of the household substation energy storage station system, and the data obtained by direct measurement is not public data provided by the meteorological department. The specific method of measuring weather information will not be repeated here. The weather information measured in this way is more representative.

[0081] In this solution, wind power generation and photovoltaic power generation are predicted using two independent modules, but the structure of the neural network model built in the module is the same, only the training process is different. Therefore, only the neural network model and training process of wind power are introduced here.

[0082] Historical wind power data includes historical power output data, so historical wind power data is a power curve that changes over time. The same is true for historical photovoltaic data.

[0083] Weather information includes historical wind speed, historical wind direction, historical temperature, historical humidity and historical air pressure. In addition to the above four points, weather information can also include other information.

[0084] For wind speed and wind direction, in practice, vector conversion can be performed according to the orientation of the wind turbine in the wind power module, and the wind speed and wind direction can be combined into one piece of information. For example, if the wind turbine faces north and south, the wind direction is the wind speed calculated according to the vector to the orientation of the wind turbine. Of course, if some wind turbines have the function of wind direction tracking, the wind direction factor is not considered directly.

[0085] The historical wind speed, historical wind direction, historical temperature, historical humidity and historical air pressure are the data change curves for 24 hours a day within the preset date of each year;

[0086] The preset date here is actually the local season. For example, in a certain area, the north wind blows in winter and the southeast wind blows in spring. Because there are two obvious differences, the winter and spring data are selected separately. In winter, the winter data is used to train the wind power prediction module, and then the wind power prediction module is used to perform wind power prediction. In spring, the spring data is used for prediction, and then the spring model is used for wind power prediction.

[0087] Therefore, in this solution, different models need to be selected for training and prediction in different seasons. How to replace the model in the wind power prediction module is an existing technology. This solution only provides a model training method. The structure of the models in different seasons is the same. When a model structure and training method are known, the models of other seasons can also be easily obtained.

[0088] For each item of weather information, organize it into an information matrix with hours in horizontal direction and dates in vertical direction;

[0089] When making wind power forecasts, the forecast is usually made on a daily basis, so it is necessary to sort out the weather information. After sorting, multiple information matrices can be obtained. Take the information matrix obtained from temperature as an example;

[0090] The horizontal direction of the information matrix is ​​the average temperature of each hour in 24 hours, so that a series of data can be obtained, and then the row data of each day is collected and arranged downward to obtain an information matrix. Figure 2 shown.

[0091] In this solution, weather information is processed in this way, and historical wind power data is also processed in this way, so the final weather information and historical wind power data have the same data format, which is the format of information matrix. And as the date progresses, the information matrix will be gradually improved. Of course, in order to ensure the accuracy of the prediction when the next season just arrives, the data of previous years can be used as filling. When filling, the year of the filled data and the generation of the completion can be determined according to the actual situation.

[0092] Furthermore, the wind and solar power prediction device selects a trained neural network model for power generation prediction based on data similarity.

[0093] Specifically, there is often not one household substation energy storage station, but rather a group of stations in a certain area. However, due to the aging of the power generation equipment, maintenance habits, and product models of each household substation energy storage station, it is impossible for each substation energy storage station to have the same power generation prediction logic. Therefore, if the same trained neural network model is used for prediction, the accuracy will be low. And if multiple neural network models are used for prediction each time, it is difficult to directly determine which neural network model has a high prediction accuracy.

[0094] Therefore, in this scheme, the trained neural network model will be selected for power generation prediction based on data similarity. The data here refers to weather information and historical wind power data.

[0095] For example, there are 10 energy storage stations in total, and the data similarity obtained from the first 5 energy storage stations is high, which means that the internal connection between the historical wind power data and weather information between these 5 energy storage stations is more similar, so the same neural network model can be used for prediction. However, the data similarity of the last 5 energy storage stations is inconsistent with that of the first 5. Therefore, although the last 5 energy storage stations can use the same neural network model structure as the first 5 energy storage stations, different data are needed for training, so the final weight coefficients of the neural network model are different. Therefore, two neural network models are needed for the first five and last five of these 10 energy storage stations.

[0096] To this end, it is necessary to collect all samples from all household substation energy storage station systems, conduct similarity analysis, and then classify them.

[0097] The specific method is as follows:

[0098] Collect all samples H collected by the wind and solar power prediction device i ;

[0099] For sample H i The samples are divided into photovoltaic prediction samples and wind power prediction samples, which are used for photovoltaic power generation prediction and wind power generation prediction respectively. Therefore, wind power samples and photovoltaic samples cannot be confused during classification.

[0100] For sample H i The classification is carried out in the following way:

[0101] S1: For H i ={x, y} for normalization, where x is the predicted information and y is the actual power generation information.

[0102] Among them, x is the weather information, and y is the historical wind power data. x and y are simplified representations.

[0103] When classifying, because x includes multiple weather elements such as temperature, humidity, etc., x needs to be synthesized first and merged into one piece of information so that x and y can match each other.

[0104] Specifically: The normalization method in S1 is as follows: perform preliminary normalization on all wind speeds, wind directions, humidity and air pressure. The preliminary normalization here is to plan the values ​​so that the values ​​of wind speed, wind direction, humidity and air pressure are within the range of 0 to 1. In order to facilitate the description of wind speed, wind speed and wind direction need to be combined into one value according to the direction of the impeller in the wind power module. This involves the conversion of wind speed according to the vector, and the specific method will not be repeated here.

[0105] Set the corresponding weight coefficient for each wind speed, wind direction, humidity and air pressure, and add them together to get x;

[0106] x=α1A1+α2A2+α3A3+α4A4; wherein α1, α2, α3, α4 are set weight coefficients.

[0107] Because there are only 4 items of weather information in this scheme, there are only 4 items of α. If more parameters are introduced, the corresponding number of α needs to be increased.

[0108] Because there is huge information redundancy in weather information, it is actually difficult to accurately measure the weight of each type of weather information. To this end, this application provides the following solution:

[0109] The weight coefficient α of each item of weather information is calculated as follows:

[0110] α u =C r (u,U)P;

[0111] Where P is the conversion factor, C r (u,U) is the correlation ratio between the u-th information and the rest of the information;

[0112]

[0113] Among them, u represents the uth information, U represents all the information except u, C represents the power generation, I r (r,U) is the correlation coupling rate, which measures the degree of additional information sharing between u and C when all information except u is known, and its value range is 0 to 1. I(u;C|U) represents the amount of information sharing between u and C when U is known. H(u) represents the entropy of u, H(C) represents the entropy of C, and T r(A, B) represents the reduction of information sharing between u and C when U is known, and I(u; C) represents the mutual information between u and C;

[0114] The entropy is calculated as follows:

[0115]

[0116] Where H(u) represents the entropy of the u-th information in the weather information, i represents the index of the sample, n represents the total number of samples, b = 2, p(u i ) indicates that u takes the value u i probability.

[0117] In this scheme, by calculating the correlation between each information and the rest of the information, the relationship between each information and the rest of the information can be accurately described, and the weight coefficient can be set according to the intrinsic connection between them. Therefore, when setting the weight information, the redundant relationship between the information can be better considered, thereby increasing the accuracy of the model.

[0118] S2: Randomly select a feature from x as the separation value, and put all samples H i The values ​​of x are divided into two subsets, the ones greater than the eigenvalue are the right subset, and the ones less than the eigenvalue are the left subset.

[0119] The separation value here is actually a randomly selected value. The value of x ranges from 0 to 1, so the separation value can be set to 5 initially.

[0120] S3: Repeat the above S2 for the samples in the right subset and the left subset, and randomly generate a new separation value each time;

[0121] S4: Continue to segment the samples until any of the following conditions are met;

[0122] The number of splits has reached the preset maximum value;

[0123] There is only one sample in the subset;

[0124] All samples in the subset have the same feature value;

[0125] S5: Generate an isolated tree each time the stopping condition is reached, and divide the sample multiple times to obtain multiple isolated trees;

[0126] For each sample H i , traverse all isolated trees and record sample H i The number of edges passed when reaching a leaf node to obtain sample H i The height h of the sample point i ;

[0127] Calculate H for each samplei The average value of the height E i ;

[0128] Calculate H for each sample i The density S;

[0129]

[0130] Where c(n) represents the average path length between the separation point and the group, E i Represents sample H i The average value of the height, S i Represents sample H i The sparsity of n represents the total number of samples;

[0131] S6: For all samples H i The density S generates an n×n matrix R, where each element in the matrix R is S i and S j The similarity distance D i,j , i and j are the indexes of the samples, i≠j;

[0132] The similarity distance is the distance calculated according to the similarity calculation formula, such as the Hamming distance and the Euclidean distance.

[0133]

[0134] S7: Use clustering algorithm to divide the matrix R into several groups, based on the samples H in each group. i The source of the data is used to determine the classification basis of wind and solar power prediction devices.

[0135] The clustering algorithm is an existing technology, and the specific clustering analysis process is not provided. After clustering analysis, all samples H i For example, all samples are divided into 4 categories. The first category of samples includes most of the samples of the first and second energy storage stations. i , then the two energy storage stations are classified into one category, and the same type of data is used to predict the neural network model and then calculate the wind power value.

[0136] The above is the data processing method and data classification method. The following introduces the training process of the neural network model and the model structure.

[0137] Specifically, the wind power prediction module and the photovoltaic prediction module are respectively built with pre-trained prediction models; the structures of the prediction models are the same.

[0138] The prediction model is trained in a training device;

[0139] The training device comprises: a knowledge output unit and a training unit.

[0140] The knowledge output unit has a pre-trained large neural network model built in. The large neural network here is a generative adversarial neural network. Compared with the neural network structure of the prediction model, the basic structure is the same, but there are more hidden layers and weight layers inside, that is, the number of neural network layers is more and the internal structure is more complex.

[0141] The knowledge output unit will be trained with a large amount of data in advance. After the training, the large neural network model has stronger learning ability and can make more accurate predictions. However, the prediction time is long and the calculation amount is large, which is not suitable for large-scale installation and laying.

[0142] A training unit, in which the prediction model to be trained is arranged;

[0143] The training data are input into the knowledge output unit and the training unit respectively, wherein the knowledge output unit extracts information from the training data to guide the training of the built-in prediction model of the training unit.

[0144] Therefore, when making predictions for each small prediction model, you only need to load it into the training unit and then input the corresponding data to execute it.

[0145] Furthermore, both the prediction model and the large neural network model are generative adversarial networks;

[0146] refer to Figure 1 , wherein the prediction model includes a first generator, the large neural network model includes a second generator, and the prediction model and the large neural network model share a discriminator;

[0147] The first generator includes an input layer, a hidden layer, and an output layer.

[0148] Among them, the input layer is used to receive prediction information.

[0149] Hidden layer: It includes a series of convolutional layers to extract information.

[0150] Output layer: A single neuron fully connected layer that uses the Sigmoid activation function to output a probability value. This value represents the probability that the input sample is real data.

[0151] The second generator includes a large input layer and a large hidden layer, and the number of neurons in the large hidden layer is greater than the number of neurons in the hidden layer.

[0152] The same data is input into the first generator and the second generator respectively, and the intermediate features finally generated by the hidden layer of the second generator are input into the first generator to guide the final output of the hidden layer.

[0153] The hidden layer and large hidden layer in this scheme are both StyleGAN networks. However, because the hidden layer has a small number of neurons and a simple network structure, the network can extract the changes in information from the matrix information. After converting the predicted information into an information matrix, the prediction accuracy of the network is higher.

[0154] In order to ensure the accuracy of the trained model, this application provides the following loss function L:

[0155] L=λ G L G +λ d L d ;

[0156] Among them, L G is the adversarial training loss, L d is the loss of distillation training, λ G and λ d are the corresponding first loss weight and second loss weight;

[0157]

[0158] Among them, D S (x) represents the discriminator’s discrimination result on the real power generation information, D S (G S (y)) represents the discriminator's discriminant result on the power generation information generated by the first generator;

[0159] L d =λ p L p +λ l L l ;

[0160] Among them, L p is the loss of detail difference between the predicted power generation information and the actual power generation information, L l The overall loss of predicted power generation information and actual power generation information;

[0161] λ p and λ l are the corresponding third loss weight and fourth loss weight;

[0162] L p =||G s (x)-G t (x)||1;

[0163] L l =LP(G s (x),G t (x));

[0164] Among them, x is the input prediction information, y is the generated power generation information, G s represents the first generator, G t Denotes the second generator D s represents the discriminator; LP represents the perceptual similarity function.

[0165] In this solution, when training the prediction model, the prediction model is trained together with the large neural network model, and the output of the large neural network model is compared with the output of the prediction model, so that in the process of training the prediction model, the law of parameter changes in the large neural network model can be continuously learned, so as to accurately correct the structural parameters in its own neural network model, thereby improving the training rate of the prediction model and increasing the model accuracy. In this solution, the prediction model only includes the first generator. The discriminator is a neural network structure introduced during training.

[0166] After the first generator is trained, it will be extracted and installed into the required wind power prediction module for the prediction of wind power generation.

[0167] The above description is only some preferred embodiments of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above invention concept. For example, the above features are replaced with (but not limited to) the technical features with similar functions disclosed in the embodiments of the present application to form a technical solution.

Claims

1. A household substation energy storage station system, comprising: Wind power module, used to generate electricity using wind power; Photovoltaic modules for generating electricity from solar energy; Energy storage module, used to store electrical energy; A wind power collection module is used to collect historical wind power data of the wind power module; A photoelectric collection module, used to collect historical photoelectric data of the photoelectric module; It is characterized in that Weather module, used to collect local weather information; Wind and photovoltaic power prediction device, which predicts future wind power generation based on historical wind power data and weather information, and predicts future photovoltaic power generation based on historical photovoltaic data and weather information; The grid-connected control module determines the grid-connected transmission capacity based on the future photovoltaic power generation and the remaining energy storage of the energy storage module.

2. The household substation energy storage station system according to claim 1, characterized in that: The wind and photovoltaic prediction device comprises a wind power prediction module and a photovoltaic prediction module. The wind power prediction module is used to predict future wind power generation, and the photovoltaic prediction module is used to predict future photovoltaic power generation.

3. The household substation energy storage station system according to claim 1, characterized in that: Historical wind power data includes historical power output data; Weather information includes historical wind speed, historical wind direction, historical temperature, historical humidity and historical air pressure; The historical wind speed, historical wind direction, historical temperature, historical humidity and historical air pressure are the data change curves for 24 hours a day within the preset date of each year; For each item of weather information, it is organized into an information matrix with hours in the horizontal direction and dates in the vertical direction.

4. The household substation energy storage station system according to claim 1, characterized in that: The wind and solar power prediction device selects a trained neural network model for power generation prediction based on data similarity.

5. The household substation energy storage station system according to claim 3, characterized in that: Collect all samples H collected by the wind and solar power prediction device in advance i For sample H i The classification is carried out in the following way; S1: For H i ={x, y} for normalization, where x is the predicted information and y is the actual power generation information; S2: Randomly select a feature from x as the separation value, and put all samples H i The values ​​of x are divided into two subsets, the ones greater than the eigenvalue are the right subset, and the ones less than the eigenvalue are the left subset; S3: Repeat the above S2 for the samples in the right subset and the left subset, and randomly generate a new separation value each time; S4: Continue to segment the samples until any of the following conditions are met; The number of splits has reached the preset maximum value; There is only one sample in the subset; All samples in the subset have the same feature value; S5: Generate an isolated tree each time the stopping condition is reached, and divide the sample multiple times to obtain multiple isolated trees; For each sample H i , traverse all isolated trees and record sample H i The number of edges passed when reaching a leaf node to obtain sample H i The height h of the sample point i ; Calculate H for each sample i The average value of the height E i ; Calculate H for each sample i The density S i ; Where c(n) represents the average path length between the separation point and the group, E i Represents sample H i The average value of the height, S i Represents sample H i The sparsity of n represents the total number of samples; S6: For all samples H i The density S generates an n×n matrix R, where each element in the matrix R is S i and S j The similarity distance D i,j , i and j are the indexes of the samples, i≠j; S7: Use clustering algorithm to divide the matrix R into several groups, based on the samples H in each group. i The source of the data is used to determine the classification basis of wind and solar power prediction devices.

6. The household substation energy storage station system according to claim 2, characterized in that: For sample H i It is divided into photovoltaic prediction samples and wind power prediction samples. The photovoltaic prediction samples and wind power prediction samples are used for photovoltaic power generation prediction and wind power generation prediction respectively.

7. The household substation energy storage station system according to claim 5, characterized in that: The normalization method in S1 is as follows: all wind speeds, wind directions, humidity and air pressure are initially normalized, and then corresponding weight coefficients are set for each wind speed, wind direction, humidity and air pressure, and the weight coefficients are added to obtain x; x=α1A1+α2A2+α3A3+α4A4; wherein α1, α2, α3, α4 are set weight coefficients.

8. The household substation energy storage station system according to claim 7, characterized in that: The weight coefficient α for each item of weather information is calculated as follows: α u =C r (u,U)P; Where P is the conversion factor, C r (u,U) is the correlation ratio between the u-th information and the rest of the information; Among them, u represents the uth information, U represents all the information except u, C represents the power generation, I r (r,U) is the correlation coupling rate, which measures the degree of additional information sharing between u and C when all information except u is known, and its value range is 0 to 1. I(u;C|U) represents the amount of information sharing between u and C when U is known. H(u) represents the entropy of u, H(C) represents the entropy of C, and T r (A, B) represents the reduction of information sharing between u and C when U is known, and I(u; C) represents the mutual information between u and C; The entropy is calculated as follows: Where H(u) represents the entropy of the u-th information in the weather information, i represents the index of the sample, n represents the total number of samples, b = 2, p(u i ) indicates that u takes the value u i probability.

9. The household substation energy storage station system according to claim 1, characterized in that: The wind power prediction module and the photovoltaic prediction module are respectively equipped with pre-trained prediction models; The prediction model is trained in a training device; The training device comprises: Knowledge output unit, with a pre-trained large neural network model built in; A training unit, in which the prediction model to be trained is arranged; The training data are input into the knowledge output unit and the training unit respectively, wherein the knowledge output unit extracts information from the training data to guide the training of the built-in prediction model of the training unit.

Images