Photovoltaic power prediction method for optical storage charging station

Abstract

The invention relates to a photovoltaic power prediction method for an optical storage charging station. The method comprises the following steps: constructing a training sample set by using historical data of photovoltaic power of the optical storage charging station; constructing an incremental extreme learning machine model in which a correction term and a correction solution are introduced, and performing training by using the training sample set; and inputting the plurality of illumination parameter values at the current time point into the incremental extreme learning machine trained in the step S2, and predicting the photovoltaic power value of the optical storage charging station at the current time point. The method can overcome the defects that many redundant hidden layer nodes which reduce accuracy and learning efficiency exist in an existing method, hidden layer node parameters are generated randomly, and the stability of an incremental extreme learning machine is influenced, and can meet the requirement for photovoltaic power prediction of the photovoltaic storage charging station to a certain extent; and meanwhile, a new thought and a new way are provided for more accurate photovoltaic power prediction of the photovoltaic storage charging station.

Description

Technical Field

[0001] The present invention belongs to the technical field of power systems, and particularly relates to a method for predicting photovoltaic power of a photovoltaic energy storage charging station. Background Art

[0002] With the wide application of renewable energy and the rapid development of the electric vehicle industry, the photovoltaic energy storage charging station, as a new type of facility integrating photovoltaic power generation, energy storage system and electric vehicle charging functions, has gradually become a research hotspot in the energy field. In a photovoltaic energy storage charging station, accurately predicting the photovoltaic power output is of crucial significance for optimizing system operation, improving energy utilization efficiency, ensuring power supply reliability and reducing operation costs. Through accurate photovoltaic power prediction, the charging station can reasonably arrange the charging plan, make full use of photovoltaic power generation, reduce dependence on the external power grid, and at the same time avoid impacting the power grid due to excessive fluctuations in photovoltaic power.

[0003] Designing a prediction model and learning algorithm is a key issue in the research on photovoltaic power prediction of a photovoltaic energy storage charging station. In the existing technology, the prediction model based on the traditional incremental extreme learning machine has many redundant nodes that reduce the accuracy and efficiency of photovoltaic power prediction in a photovoltaic energy storage charging station. Randomly generating the hidden layer node parameters affects the stability of the incremental extreme learning machine, resulting in a large network training error. Therefore, designing an efficient prediction model is of great significance for photovoltaic power prediction in a photovoltaic energy storage charging station. Summary of the Invention

[0004] The purpose of the present invention is to overcome the deficiencies of the prior art, and provide a method for predicting photovoltaic power of a photovoltaic energy storage charging station. By introducing a correction term and a correction solution into the incremental extreme learning machine model and the training process, an incremental extreme learning machine model based on the correction term and the correction solution is constructed to achieve accurate prediction of the photovoltaic power of the photovoltaic energy storage charging station, and can effectively improve the prediction accuracy and efficiency.

[0005] The technical solution adopted by the present invention is as follows: A method for predicting photovoltaic power of a photovoltaic energy storage charging station includes the following steps:

[0006] S1: Construct a training sample set by using the historical data of the photovoltaic power of the photovoltaic energy storage charging station;

[0007] S2: Construct an incremental extreme learning machine model introducing a correction term and a correction solution, and use the training sample set for training;

[0008] S3: Input the multiple light parameter values at the current time point into the incremental extreme learning machine trained in step S2 to predict the photovoltaic power value of the photovoltaic energy storage charging station at the current time point.

[0009] Specifically, in step S1, the output of the sample is the photovoltaic power value of the photovoltaic energy storage charging station at the selected time point, and the input is the multiple light parameter values at the selected time point.

[0010] The unit of the time point is days, and the value of the illumination parameter at the time point is the average value of the illumination parameters from 0:00 to 24:00 on the same day.

[0011] The values of the illumination parameters include: solar diffuse radiation index, solar direct radiation index, total solar horizontal radiation, fixed tilt radiation, and tracking tilt radiation.

[0012] In the step S2, the incremental extreme learning machine model is as follows:

[0013]

[0014] In the formula, i represents the i-th node in the hidden layer, and L f represents the number of hidden layer nodes determined after training; represents the output matrix of the i-th hidden layer node; is the correction term added for the i-th hidden layer node. Among them, e i-1 is the network training error introduced due to the addition of the (i - 1)-th hidden layer node during training, which represents the difference between the output generated by the learning machine when there are only the 1st to (i - 1)-th hidden layer nodes in the learning machine and the ideal output given by the sample; α i-1 is the correction factor determined for the i-th hidden layer node, which is obtained by iterative calculation of the network training error; is the correction value of the output weight determined for the i-th hidden layer node, which is a linear combination of the randomly given value of the output weight of the hidden layer node and the improved value of the output weight of the hidden layer node. Among them, the improved value of the output weight of the hidden layer node is obtained by iterative calculation of the input weight of the hidden layer node and the threshold of the hidden layer node obtained during training, and the input weight of the hidden layer node and the threshold of the hidden layer node are obtained by iterative calculation of the network training error of the hidden layer node and the input sample during training; the input weight and threshold of the hidden layer node together constitute the correction solution.

[0015] Advantages of the present invention: The present invention can overcome the disadvantages that there are many redundant hidden layer nodes in the existing methods that reduce the accuracy and learning efficiency, and the random generation of hidden layer node parameters affects the stability of the incremental extreme learning machine. To a certain extent, it can meet the needs of photovoltaic power prediction in a photovoltaic energy storage charging station, and at the same time provides new ideas and new ways for more accurate photovoltaic power prediction in a photovoltaic energy storage charging station. Description of the Drawings

[0016] Figure 1 is the algorithm flowchart of the incremental extreme learning machine based on the correction term and the correction solution of the present invention. Detailed Embodiments

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. The following will be specifically described in conjunction with the embodiments.

[0018] As Figure 1 shown, the present invention includes the following steps:

[0019] S1: Construct a training sample set using the historical data of the photovoltaic power of the optical storage charging station; the specific process is as follows:

[0020] In the step S1, the output of the sample is the photovoltaic power value of the optical storage charging station at the selected time point, and the input is multiple illumination parameter values at the selected time point.

[0021] The unit of the time point is days, and the illumination parameter value of the time point is the average value of the illumination parameters from 0:00 to 24:00 on the same day.

[0022] The illumination parameter values include: solar diffuse radiation index, solar direct radiation index, total solar horizontal radiation, fixed tilt radiation, tracking tilt radiation.

[0023] S2: Construct an incremental extreme learning machine model introducing a correction term and a correction solution, and train it using the training sample set; the specific process is as follows:

[0024] In the step S2, the incremental extreme learning machine model is:

[0025]

[0026] In the formula, i represents the i-th node in the hidden layer, and L f represents the number of hidden layer nodes determined after training; represents the output matrix of the i-th hidden layer node; is the correction term added for the i-th hidden layer node, where e i-1 is the network training error introduced due to adding the (i - 1)-th hidden layer node during training, which represents the difference between the output generated by the learning machine when there are only the 1st to (i - 1)-th hidden layer nodes in the learning machine and the ideal output given by the sample; α i-1 is the correction factor determined by the i-th hidden layer node and is obtained by iterative calculation of the network training error; is the correction value of the output weight determined for the i-th hidden layer node, and is a linear combination of the randomly given value of the output weight of the hidden layer node and the improved value of the output weight of the hidden layer node. The improved value of the output weight of the hidden layer node is iteratively calculated from the input weight of the hidden layer node and the threshold of the hidden layer node obtained during training. The input weight of the hidden layer node and the threshold of the hidden layer node are iteratively calculated from the network training error of the hidden layer node and the input sample during training; the input weight and threshold of the hidden layer node together constitute the correction solution.

[0027] Further, the training process of the incremental extreme learning machine based on the correction term and the correction solution includes the following steps, which are executed once for each input training sample:

[0028] Define the input vector of the current sample as X = [x1, x2, x3, x4, x5], and the output as y, where x1 represents the solar diffuse radiation index at the current time point, x2 represents the solar direct radiation index at the current time point, x3 represents the total solar horizontal radiation at the current time point, x4 represents the fixed inclination radiation at the current time point, x5 represents the tracking inclination radiation at the current time point, and y represents the photovoltaic power value of the photovoltaic charging station at the current time point.

[0029] Step 1. In the initialization stage of the extreme learning machine network, let the initial value of the number of hidden layer nodes L be 0, and the maximum value be L max , and the network training error e L = e0 = y, and the error expectation value is ε;

[0030] Step 2. Add a node to the hidden layer, let L increment by 1, and at the same time assign L to the newly added hidden layer node as its node number, and randomly generate the output weight β of the newly added hidden layer node L and related parameters v L , z L , and satisfy 0 < v L < z L < 1, v L + z L = 1;

[0031] Step 3. Calculate the output feedback matrix of the newly added hidden layer node according to Equation (2):

[0032] Η L = e L-1 (β L ) -1 (2)

[0033] In the formula, e L-1 is determined during the training process of the previous hidden node added, and it represents the network training error when the number of hidden layer nodes is L - 1;

[0034] Step 4: Calculate the input weights of the newly added hidden layer nodes according to Equation (3):

[0035]

[0036] In the formula, is the Moore-Penrose generalized inverse of x;

[0037] Step 5: Calculate the thresholds of the newly added hidden layer nodes according to Equation (4):

[0038] b L = rmse(Η L -a L ·x) (4)

[0039] In the formula, rmse is the root mean square error function;

[0040] Step 6: Calculate the output matrix of the newly added hidden layer nodes according to Equation (5):

[0041]

[0042] In the formula, u() can adopt a general activation function in the neural network, such as cos(), tanh();

[0043] Step 7: Calculate the correction factor of the newly added hidden layer nodes according to Equation (6):

[0044]

[0045] Step 8: Calculate the improvement value of the output weights of the newly added hidden layer nodes according to Equation (7):

[0046]

[0047] Step 9: Calculate the correction value of the output weights of the newly added hidden layer nodes according to Equation (8):

[0048]

[0049] Step 10: Calculate the network training error value after adding the L-th newly added hidden layer node according to Equation (9):

[0050]

[0051] In the formula, is the correction term added for the i-th hidden layer node;

[0052] Step 11: Judge whether L≥L is satisfied max or ||e L|| ≤ ε. If yes, complete the training, end this process. Otherwise, return to step 2.

[0053] S3: Input the multiple light parameter values at the current time point into the incremental extreme learning machine trained in step S2 to predict the photovoltaic power value of the optical storage charging station at the current time point.

[0054] The effectiveness of the present invention can be further illustrated by the following simulation experiment. The experiment collects the photovoltaic power data of the optical storage charging station for 4 days, records the photovoltaic power data of the optical storage charging station within this time period every 15 minutes, and a total of 384 time point data are recorded.

[0055] The performance of the photovoltaic power prediction model of the optical storage charging station is measured by the root mean square error RMSE (root mean square error) and the model validity MV (model validity) to measure the generalization ability and accuracy of the photovoltaic power prediction model of the optical storage charging station.

[0056] The root mean square error RMSE is expressed as

[0057]

[0058] The model validity MV is expressed as

[0059]

[0060] where t i is the output of the incremental extreme learning machine model based on the correction term and the correction solution; is the true value; is the average value of the true values; N is the number of samples. Among them, the root mean square error RMSE reflects the fluctuation of the model output curve on the actual curve, and the model validity MV reflects the discreteness of the deviation between the model output and the measured value relative to the measured data. The model validity MV of a well-performing model is 1.

[0061] The experiment selects several classic photovoltaic power prediction models in past existing works for comparison: the support vector machine model (SVM), the radial basis neural network model, and the BP neural network model, and uses the parameter tuning method consistent with the relevant literature. The comparison results of each prediction model are shown in Table 1.

[0062]

[0063] Table 1: Performance comparison of different prediction models

[0064] As can be seen from Table 1, compared with the BP neural network model, the radial basis neural network model, and the support vector machine model, the root mean square error of the photovoltaic power prediction of the energy storage charging station using the incremental extreme learning machine model based on the correction term and the correction solution has decreased, and the effectiveness of its photovoltaic power prediction model for the energy storage charging station has been relatively improved, indicating that it is effective to use the incremental extreme learning machine model based on the correction term and the correction solution to simulate and predict the photovoltaic power of the energy storage charging station.

[0065] In the present invention, by adding a correction term to the existing incremental extreme learning machine model and feeding back the network training error and the correction factor to the output of the hidden layer, the prediction result can be made closer to the output sample, and the number of redundant hidden layer nodes of the incremental extreme learning machine can be reduced, thereby accelerating the network convergence speed of the incremental extreme learning machine.

[0066] By introducing a correction factor and a correction solution in the training process of the existing incremental extreme learning machine model, that is, in the training process, by randomly generating output weights and combining the network training error, the correction factor, and the input samples, more optimal hidden layer node parameters can be calculated, including the input weights, thresholds, output weights, and network training error, which can optimize the network structure, improve the stability of the network training process, and thus effectively reduce the network training error.

[0067] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the same elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. A photovoltaic power prediction method for a photovoltaic energy storage charging station, characterized in that, Including the following steps: S1: Construct a training sample set using the historical data of the photovoltaic power of the photovoltaic energy storage charging station; S2: Construct an incremental extreme learning machine model introducing a correction term and a correction solution, and train it using the training sample set; S3: Input the multiple illumination parameter values at the current time point into the incremental extreme learning machine trained in step S2 to predict the photovoltaic power value of the photovoltaic energy storage charging station at the current time point.

2. The photovoltaic power prediction method for a photovoltaic energy storage charging station according to claim 1, wherein: In step S1, the output of the sample is the photovoltaic power value of the photovoltaic energy storage charging station at the selected time point, and the input is the multiple illumination parameter values at the selected time point.

3. The photovoltaic power prediction method for a photovoltaic energy storage charging station according to claim 2, wherein: The unit of the time point is day, and the illumination parameter value of the time point is the average value of the illumination parameters from 0:00 to 24:00 on the day.

4. A photovoltaic power prediction method for a photovoltaic energy storage charging station according to claim 3, characterized in that: The illumination parameter values include: solar diffuse radiation index, solar direct radiation index, total solar horizontal radiation, fixed tilt radiation, tracking tilt radiation.

5. A photovoltaic power prediction method for a photovoltaic energy storage charging station according to claim 1, characterized in that, In step S2, the incremental extreme learning machine model is: where \(i\) represents the \(i\)-th node in the hidden layer, and \(L\) f represents the number of hidden layer nodes determined after training; represents the output matrix of the \(i\)-th hidden layer node; is the correction term added to the \(i\)-th hidden layer node, where \(e\) i-1 is the network training error introduced due to adding the \((i - 1)\)-th hidden layer node during training, which represents the difference between the output generated by the learning machine when there are only the 1st to \((i - 1)\)-th hidden layer nodes in the learning machine and the ideal output given by the sample; \(\alpha\) i-1 is the \(i\)-th hidden The correction factor determined by the layer nodes is obtained by iteratively calculating the network training error; It is the correction value of the output weight determined for the i-th hidden layer node, and it is a linear combination of the randomly given value of the output weight of the hidden layer node and the improved value of the output weight of the hidden layer node. The improved value of the output weight of the hidden layer node is obtained by iteratively calculating the input weight of the hidden layer node and the threshold of the hidden layer node obtained during training. The input weight of the hidden layer node and the threshold of the hidden layer node are determined by the network training error and input samples of the hidden layer node during training Obtained through iterative calculation; the input weights and thresholds of the hidden layer nodes together constitute the corrected solution. The input weights and thresholds of the hidden layer nodes together constitute the corrected solution.

Images