Robust distributed photovoltaic power station generation power prediction method and system

Abstract

The invention relates to a robust distributed photovoltaic power station generation power prediction method and system, and the method comprises the steps: obtaining historical data of a distributed photovoltaic power station, and carrying out the preprocessing; selecting weather characteristic data with strong correlation with photovoltaic output power as input data; obtaining a robust distributed photovoltaic prediction model; training a weight parameter of the distributed photovoltaic prediction model, and selecting an optimal hyper-parameter; and obtaining a real-time prediction result. The method has the beneficial effects that the outliers are considered in the designed system model, so that the interference of noise on a prediction result in photovoltaic output power prediction can be effectively reduced; and the distributed model takes the system as a whole, and a plurality of stations are cooperatively trained, so that the prediction precision can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic power generation, and more specifically, to a robust distributed photovoltaic power station power generation prediction method and system. Background Art

[0002] Solar energy is the most abundant clean energy available to mankind and an important component of future energy. Distributed photovoltaic power generation is a new energy form that uses photovoltaic modules to directly convert solar energy into electrical energy. Compared with centralized photovoltaic power generation, distributed photovoltaic power stations have the characteristics of flexible distribution, close to users, and high resource utilization efficiency. They are particularly suitable for deployment in scenarios such as house roofs, commercial buildings, and industrial parks.

[0003] The application of distributed photovoltaic power stations faces a series of technical challenges. Due to the randomness and intermittency of photovoltaic power generation, its power output is easily affected by weather, geographical location and day and night changes, resulting in unstable power generation. Compared with centralized power stations, distributed photovoltaic power stations have more complex system characteristics due to their dispersion and diversity, which puts higher requirements on prediction methods. In addition, the access and management of distributed photovoltaic power generation systems and power grids are also very complex. Especially in high penetration scenarios, how to achieve stable grid scheduling and supply and demand balance has become a key issue that needs to be solved urgently.

[0004] Therefore, studying how to improve the power generation prediction accuracy, noise resistance and grid adaptability of distributed photovoltaic power generation systems is of great significance for improving the overall utilization efficiency of distributed photovoltaics and promoting the sustainable development of new energy. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and to propose a robust distributed photovoltaic power station power generation prediction method and system.

[0006] In a first aspect, a robust distributed photovoltaic power station power generation prediction method is provided, comprising:

[0007] Step 1: Obtain historical data of distributed photovoltaic power stations and perform preprocessing; the historical data includes a variety of weather characteristic data and photovoltaic power generation output power at corresponding times;

[0008] Step 2: Use correlation analysis to check the correlation between various weather characteristic data and photovoltaic output power, and select weather characteristic data with strong correlation with photovoltaic output power as input data;

[0009] Step 3: Obtain a robust distributed photovoltaic prediction model;

[0010] Step 4: obtaining a training set and a validation set, using the training set to train the weight parameters of the distributed photovoltaic prediction model, and using the validation set to select the best hyperparameters;

[0011] Step 5: Obtain real-time prediction results based on the input data and the trained distributed photovoltaic prediction model.

[0012] Preferably, in step 1, the multiple weather characteristic data include horizontal irradiance, diffuse irradiance, temperature and humidity data; and the preprocessing includes: interpolation processing of missing data and data normalization.

[0013] Preferably, in step 3, the mathematical model of the distributed photovoltaic prediction model is expressed as:

[0014]

[0015] Among them, Y l represents the observed value of the output power of the lth power station in the distributed photovoltaic system, H l represents the reservoir output of the neural network of the lth power station in the distributed photovoltaic system, W l is the output layer weight of the neural network of the lth power station in the distributed photovoltaic system, λ l is the optimal hyperparameter of the lth power station system, and L represents the number of power stations in the distributed photovoltaic system.

[0016] Preferably, in step 4, an alternating direction multiplier method is used to design a multi-site coordinated training process of a distributed photovoltaic prediction model.

[0017] Preferably, in the training process of the distributed photovoltaic prediction model in step 4, the mean value information is transmitted between sites through the DAC criterion.

[0018] Preferably, in step 4, the distributed photovoltaic prediction model is a state echo network, which is used to explore the nonlinear relationship between photovoltaic power generation and weather characteristic data; and the training set, validation set, and test set are divided according to a certain ratio; the robust state echo network is trained by the training set, and the optimal hyperparameters are found by grid search on the validation set.

[0019] Preferably, in step 4, the state echo network consists of a randomly generated input layer, a storage layer and an output layer to be trained, and the storage layer has recurrent neural network characteristics.

[0020] In a second aspect, a robust distributed photovoltaic power station power generation prediction system is provided, which is used to execute any method described in the first aspect, including:

[0021] The first acquisition module is used to acquire historical data of the distributed photovoltaic power station and perform preprocessing; the historical data includes a variety of weather characteristic data and photovoltaic power generation output power at corresponding moments;

[0022] An analysis module is used to test the correlation between various weather characteristic data and photovoltaic output power by using correlation analysis, and select weather characteristic data with strong correlation with photovoltaic output power as input data;

[0023] The second acquisition module is used to obtain a robust distributed photovoltaic prediction model;

[0024] A third acquisition module is used to acquire a training set and a validation set, use the training set to train the weight parameters of the distributed photovoltaic prediction model, and use the validation set to select the best hyperparameters;

[0025] The fourth acquisition module is used to obtain real-time prediction results according to the input data and the trained distributed photovoltaic prediction model.

[0026] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any method described in the first aspect.

[0027] In a fourth aspect, an electronic device is provided, including:

[0028] Memory, used to store computer programs;

[0029] A processor is used to execute the computer program to implement any method as described in the first aspect.

[0030] The beneficial effects of the present invention are:

[0031] 1. The present invention adopts a machine learning method, which has a small amount of calculation and a simple model structure, and can be run on edge devices of large-scale regional photovoltaic power stations.

[0032] 2. The system model designed by the present invention takes outliers into consideration, which can effectively reduce the interference of noise in photovoltaic output power prediction on the prediction results.

[0033] 3. The distributed model of the present invention takes the system as a whole, and multiple sites are trained collaboratively, which can effectively improve the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A flowchart of a robust distributed photovoltaic power generation power prediction method provided by the present invention;

[0035] Figure 2 A structural diagram of a robust distributed photovoltaic power station prediction model provided by the present invention;

[0036] Figure 3 A prediction system diagram of a single photovoltaic power station provided by the present invention;

[0037] Figure 4 A schematic diagram of the structure of the robust distributed photovoltaic power generation prediction system provided by the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0039] Embodiment 1:

[0040] To solve the problems of the prior art, the present application provides a robust distributed photovoltaic power station power generation prediction method, such as Figure 1 As shown, including:

[0041] Step 1: Obtain historical data of distributed photovoltaic power stations and perform preprocessing; the historical data includes various weather characteristic data and photovoltaic power generation output power at corresponding times.

[0042] In step 1, the various weather characteristic data include environmental variables such as horizontal irradiance, diffuse irradiance, wind speed, temperature and humidity data; the preprocessing includes: interpolation of missing data, cleaning of outliers in the data and data normalization. In addition, since distributed photovoltaic power generation data is time series data, it is also necessary to perform time alignment and interpolation on multiple power station data to meet the needs of distributed power station prediction training.

[0043] Specifically, interpolation processing is performed on the missing data, including: performing secondary interpolation on data with short missing data segments, and discarding data with long missing data segments.

[0044] In addition, the historical data of the photovoltaic power station are normalized in each feature dimension. The normalization formula is as follows:

[0045]

[0046] x represents the true value of a feature data. max , x′ respectively represent the maximum and minimum values ​​of a feature dimension, and x′ represents the normalized data.

[0047] Step 2: Use correlation analysis to check the correlation between various weather characteristic data and photovoltaic output power, and select weather characteristic data with strong correlation with photovoltaic output power as input data.

[0048] For example, weather characteristic data is represented as: Among them, m = 1, 2, 3, 4, corresponding to temperature, humidity, total horizontal irradiance and total diffuse irradiance, respectively, and several features with the strongest correlation are selected.

[0049] Specifically, the correlation analysis includes:

[0050] Calculate the Pearson correlation coefficient between each weather characteristic and photovoltaic output power. The steps include:

[0051] Calculate the Pearson correlation coefficient r between the input features and the power m , the formula is:

[0052]

[0053] Where n represents the number of sampling points. The i-th sample value of the m-th feature in the data set, y i The photovoltaic output power data of the i-th sampling point, represents the mean of the mth feature, represents the mean value of the output power, select |r m |Large weather features are used as input models of the network.

[0054] Step 3: Obtain a robust distributed photovoltaic prediction model.

[0055] In step 3, in order to deal with the noise problem existing in the photovoltaic data collected by the sensor, this application designs a reasonable mathematical model network. The neural network mathematical model of a single power station should be:

[0056]

[0057] Among them, Y represents the observed value of the output power in the prediction of photovoltaic power stations, H is the reservoir output of the neural network, W is the output layer weight of the neural network, and λ represents the hyperparameter required by the model. Furthermore, the prediction problem of distributed photovoltaic power stations is to optimize the overall prediction results of the neural network of multiple sites in the distributed system. The l1 norm with outliers is used as the loss function, and ridge regularization improves the stability of the model. The mathematical model is expressed as:

[0058]

[0059] Among them, Y l represents the observed value of the output power of the lth power station in the distributed photovoltaic system, Hl represents the reservoir output of the neural network of the lth power station in the distributed photovoltaic system, W l is the output layer weight of the neural network of the lth power station in the distributed photovoltaic system, λ l is the optimal hyperparameter of the lth power station system, and L represents the number of power stations in the distributed photovoltaic system.

[0060] Step 4: Obtain a training set and a validation set, use the training set to train the weight parameters of the distributed photovoltaic prediction model, and use the validation set to select the best hyperparameters.

[0061] In step 4, multiple sites of distributed photovoltaics are considered, and the complex system uses the alternating direction method of multipliers (ADMM) to design the collaborative training process of multiple networks of the distributed photovoltaic prediction model in step 3. In addition, the prediction model of the robust distributed photovoltaic system considers the collaboration of multiple sites, and the collaborative prediction requires the transmission of mean information between sites, and communication between sites is required.

[0062] In step 4, the training process of the model is designed using the ADMM algorithm to transform the system optimization problem. The transformation result is:

[0063]

[0064] Specifically, in step 4, Based on the converted model, we get the augmented Lagrangian form of the problem:

[0065]

[0066] Use ADMM algorithm to transform the problem into l ,D l The three sub-problems of and Z are to optimize the three variables respectively. After enough iterations, the optimal parameters of the neural network are finally obtained. The training process of the neural network model is as follows:

[0067] 1. About the k+1th time The sub-problem can be expressed as:

[0068]

[0069] By solving it, we can get for:

[0070]

[0071] 2. k+1th Z k+1 The sub-problem can be expressed as:

[0072]

[0073] By solving it, we can get for:

[0074]

[0075] in, and Respectively represent and The mean of :

[0076]

[0077] 3. k+1th time The sub-problem can be expressed as:

[0078]

[0079] There is an l1 norm in the problem, which is difficult to solve using the gradient descent method. It is necessary to use a soft threshold function to solve the challenge brought by the l1 norm. for:

[0080]

[0081] 4. For Lagrange multipliers and To update:

[0082]

[0083] By using the above training process to train the robust distributed photovoltaic power generation neural network, a robust distributed ESN prediction model can be obtained. and The DAC criterion is iteratively obtained in distributed sites, and finally a robust network model is obtained.

[0084] In addition, the above iterative method is used to train a robust distributed prediction model, and the grid search method is used to obtain the optimal hyperparameter λ on the validation set.

[0085] Step 5: Obtain real-time prediction results based on the input data and the trained distributed photovoltaic prediction model.

[0086] Embodiment 2:

[0087] Based on Example 1, Example 2 of the present application provides a more specific and robust distributed photovoltaic power station power generation prediction method, including:

[0088] Step 1: Obtain historical data of distributed photovoltaic power stations and perform preprocessing; the historical data includes various weather characteristic data and photovoltaic power generation output power at corresponding times.

[0089] Step 2: Use correlation analysis to check the correlation between various weather characteristic data and photovoltaic output power, and select weather characteristic data with strong correlation with photovoltaic output power as input data.

[0090] For example, temperature, humidity, total horizontal irradiance, total diffuse irradiance and half-hour output power of each power station, which are highly correlated with photovoltaic output power, are selected as network inputs.

[0091] Step 3: Obtain a robust distributed photovoltaic prediction model.

[0092] Step 4: Obtain a training set and a validation set, use the training set to train the weight parameters of the distributed photovoltaic prediction model, and use the validation set to select the best hyperparameters.

[0093] In step 4, the ADMM algorithm is used to transform the problem into multiple sub-problems, and the sub-problem variables are optimized respectively. After enough iterations, the optimal parameters of the neural network are finally obtained. and The transmission relies on the DAC criterion to iterate and obtain the mean value to improve the communication stability of the system. Specifically, the communication iteration method between each station and its connected stations is:

[0094]

[0095] Among them, d i represents the degree of the i power station on the power station communication graph, and d represents the degree of the node in the power station communication graph, which represents the maximum degree in the network. The power stations communicate with each other multiple times and transmit according to the elements of the matrix P. and Update to get the mean.

[0096] The specific network prediction structure is as follows Figure 2 As shown, the robust distributed photovoltaic power station prediction model communicates during the training process, and each power station uses its own data to train the network. During the training process, the power stations communicate and transmit the parameters of the training process. The mean required for the specific training process in Example 1 can be obtained without sending the data to an additional central site. Only a small amount of parameter information needs to be transmitted between power stations to achieve a synergistic effect.

[0097] In addition, the distributed photovoltaic prediction model is a state echo network, which uses the state echo network to mine the nonlinear relationship between photovoltaic power generation and weather characteristic data; and divides the training set, validation set, and test set according to a certain ratio (such as 6:2:2). The state echo network consists of a randomly generated input layer, a storage layer, and an output layer to be trained. The storage layer has the characteristics of a recurrent neural network, as follows:

[0098] r(t)=tanh(x(t)W in +r(t-1)W res ),

[0099] h(t)=[r(t;x(t)],

[0100] y(t)=h(t)W.

[0101] Where x(t) is the input vector at time t, r(t-1) is the state vector at time t-1, and h(t) is the reservoir output vector at time t.

[0102] The normalized mean absolute error (e NMAE ) is evaluated, and the expression is:

[0103]

[0104] Among them, P e represents the rated power of the photovoltaic power station, n represents the number of samples, represents the true value of the ith sample of PV output, Represents the i-th predicted value of PV output.

[0105] The state echo network is trained through the training process of step 4, and the optimal hyperparameter λ is determined on the validation set, thereby obtaining the state echo network model under the optimal hyperparameter condition.

[0106] Step 5: Obtain real-time prediction results based on the input data and the trained distributed photovoltaic prediction model.

[0107] For example, the prediction results are shown in Table 1, and the data source is the distributed photovoltaic power station in the Yulara area of ​​Australia. The control group is R-ESN, S-ESN, E-ESN, and OR-ESN, which respectively represent the ridge regularized ESN model, the sparse ESN model, the elastic network E-ESN model, and the outlier robust ESN model. The outlier robust ESN model of this patent is DOR-ESN.

[0108] Table 1 Prediction results of distributed photovoltaic power stations

[0109] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be described in detail in this application.

[0111] Embodiment 3:

[0112] Based on Examples 1 and 2, Example 3 of the present application provides a robust distributed photovoltaic power station power generation prediction system, such as Figure 3 As shown, including:

[0113] The first acquisition module is used to acquire historical data of the distributed photovoltaic power station and perform preprocessing; the historical data includes a variety of weather characteristic data and photovoltaic power generation output power at corresponding moments;

[0114] An analysis module is used to test the correlation between various weather characteristic data and photovoltaic output power by using correlation analysis, and select weather characteristic data with strong correlation with photovoltaic output power as input data;

[0115] The second acquisition module is used to obtain a robust distributed photovoltaic prediction model;

[0116] A third acquisition module is used to acquire a training set and a validation set, use the training set to train the weight parameters of the distributed photovoltaic prediction model, and use the validation set to select the best hyperparameters;

[0117] The fourth acquisition module is used to obtain real-time prediction results according to the input data and the trained distributed photovoltaic prediction model.

[0118] Specifically, the system provided in this embodiment is a system corresponding to the method provided in Embodiments 1 and 2. Therefore, the parts in this embodiment that are the same or similar to Embodiments 1 and 2 can be referenced to each other and will not be repeated in this application.

Claims

1. A robust distributed photovoltaic power station power generation prediction method, characterized in that: include: Step 1: Obtain historical data of distributed photovoltaic power stations and perform preprocessing; the historical data includes a variety of weather characteristic data and photovoltaic power generation output power at corresponding times; Step 2: Use correlation analysis to check the correlation between various weather characteristic data and photovoltaic output power, and select weather characteristic data with strong correlation with photovoltaic output power as input data; Step 3: Obtain a robust distributed photovoltaic prediction model; Step 4: obtaining a training set and a validation set, using the training set to train the weight parameters of the distributed photovoltaic prediction model, and using the validation set to select the best hyperparameters; Step 5: Obtain real-time prediction results based on the input data and the trained distributed photovoltaic prediction model.

2. The robust distributed photovoltaic power station power generation prediction method according to claim 1 is characterized in that: In step 1, the various weather characteristic data include horizontal irradiance, diffuse irradiance, temperature and humidity data; the preprocessing includes: interpolation processing of missing data and data normalization.

3. The robust distributed photovoltaic power station power generation prediction method according to claim 2 is characterized in that: In step 3, the mathematical model of the distributed photovoltaic prediction model is expressed as: Among them, Y l represents the observed value of the output power of the lth power station in the distributed photovoltaic system, H l represents the reservoir output of the neural network of the lth power station in the distributed photovoltaic system, W l is the output layer weight of the neural network of the lth power station in the distributed photovoltaic system, λ l is the optimal hyperparameter of the lth power station system, and L represents the number of power stations in the distributed photovoltaic system.

4. The robust distributed photovoltaic power station power generation prediction method according to claim 3 is characterized in that: In step 4, the alternating direction multiplier method is used to design a multi-site coordinated training process for the distributed photovoltaic prediction model.

5. The robust distributed photovoltaic power station power generation prediction method according to claim 4 is characterized in that: In the training process of the distributed photovoltaic prediction model in step 4, the mean value information is transmitted between sites through the DAC criterion.

6. The robust distributed photovoltaic power station power generation prediction method according to claim 5, characterized in that: In step 4, the distributed photovoltaic prediction model is a state echo network, which is used to explore the nonlinear relationship between photovoltaic power generation and weather characteristic data; and the training set, validation set, and test set are divided according to a certain ratio; the robust state echo network is trained through the training set, and the optimal hyperparameters are found by grid search on the validation set.

7. The robust distributed photovoltaic power station power generation prediction method according to claim 6, characterized in that: In step 4, the state echo network is composed of a randomly generated input layer, a storage layer and an output layer to be trained, and the storage layer has recurrent neural network characteristics.

8. A robust distributed photovoltaic power station power generation prediction system, characterized in that: Used to perform the method according to any one of claims 1 to 7, comprising: The first acquisition module is used to acquire historical data of the distributed photovoltaic power station and perform preprocessing; the historical data includes a variety of weather characteristic data and photovoltaic power generation output power at corresponding moments; An analysis module is used to test the correlation between various weather characteristic data and photovoltaic output power by using correlation analysis, and select weather characteristic data with strong correlation with photovoltaic output power as input data; The second acquisition module is used to obtain a robust distributed photovoltaic prediction model; A third acquisition module is used to acquire a training set and a validation set, use the training set to train the weight parameters of the distributed photovoltaic prediction model, and use the validation set to select the best hyperparameters; The fourth acquisition module is used to obtain real-time prediction results according to the input data and the trained distributed photovoltaic prediction model.

9. A computer storage medium, characterized in that The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 7.

10. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 7.

Images