Multi-path MPPT control method of photovoltaic inverter

Abstract

The invention discloses a multi-path MPPT (maximum power point tracking) control method for a photovoltaic inverter. The method comprises the following steps: A, acquiring input current, input voltage, output current and output voltage data of each boost converter; b, calculating the working state of each boost converter, and inputting the working state data of the boost converter into the random forest model to obtain the total fitting prediction output of the boost converter; c, the photovoltaic inverter performs MPPT control according to the total fitting prediction output of the boost converter; and D, performing MPPT control on each boost converter according to the adjusted working parameters of the photovoltaic inverter. According to the invention, the defects in the prior art can be overcome, and the multi-path MPPT control level of the photovoltaic inverter is improved.

Description

Technical Field

[0001] The invention relates to the technical field of photovoltaic power generation grid-connected control, in particular to a multi-path MPPT control method for a photovoltaic inverter. Background Art

[0002] The existing multi-channel MPPT control loop includes a boost converter and an inverter. The output of the photovoltaic cell is boosted by the boost converter and then connected to the grid through the inverter. Since the boost converter and the inverter have their own independent control logic and there is a direct series topological relationship between the two, the existing multi-channel MPPT control algorithm cannot take into account the real-time performance and uniformity between the boost converter and the inverter regulation. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a multi-channel MPPT control method for a photovoltaic inverter, which can solve the deficiencies of the prior art and improve the multi-channel MPPT control level of the photovoltaic inverter.

[0004] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows.

[0005] A multi-channel MPPT control method for a photovoltaic inverter comprises the following steps: A. Obtain the input current, input voltage, output current and output voltage data of each boost converter; B. Calculate the working status of each boost converter, input the working status data of the boost converter into the random forest model, and obtain the total fitting prediction output of the boost converter; C. The photovoltaic inverter performs MPPT control based on the total fitted predicted output of the boost converter; D. Perform MPPT control on each boost converter according to the adjusted working parameters of the photovoltaic inverter.

[0006] Preferably, in step A, the working state of the photovoltaic panel connected to each boost converter is fitted.

[0007] Preferably, the output voltage and output current of each photovoltaic panel connected to the same boost converter are obtained, and normal distribution fitting is performed on the output voltage and output current respectively, and the deviation data is eliminated according to the 3σ principle, and then the average output voltage and average output current are calculated as the working state of the photovoltaic panel after fitting.

[0008] Preferably, in step B, the working state of the boost converter includes the output voltage and output current of the boost converter, and the random forest model includes decision trees corresponding to the boost converters one by one.

[0009] Preferably, the historical output voltage and historical output current of the boost converter are used as the training data set of the random forest model, the output voltage and output current at the same time are composed into a two-dimensional array, a sliding window with a step size of Δt is set to traverse the training data set, and the first-order double dynamic reference data of the training data set is obtained. The extreme points of the fitting curve of the first-order double dynamic reference data are used as the best segmentation features, and the decision tree is recursively split using the training data set to obtain the original decision tree.

[0010] Preferably, the original decision tree is divided into two levels, the output of the first-level decision tree corresponds to the prediction output range of the boost converter, the second-level decision tree performs secondary prediction within the prediction output range provided by the first-level decision tree, and then the prediction results are averaged to obtain the total fitting prediction output of the boost converter.

[0011] Preferably, test data is input into each decision tree for testing, and the decision trees are arranged in descending order according to the stability of the output results of the decision trees. The last half of the decision trees in the sequence are selected to form the first-level decision trees, and the remaining decision trees form the second-level decision trees. After the first-level decision trees output their respective prediction results, the prediction results are mapped to the coordinate system to form prediction result nodes, and all the edge prediction result nodes in the coordinate system are used to enclose a closed area as the prediction output range.

[0012] Preferably, in step D, after the working parameters of the boost converter are adjusted, the corresponding decision book is updated; during the update, the first-level decision tree and the second-level decision tree are partially updated synchronously, and the unupdated first-level decision tree and the second-level decision tree are used to maintain the total fitting prediction output of the boost converter.

[0013] The beneficial effect brought by adopting the above technical solution is that the present invention introduces the random forest model and optimizes and innovates the existing random forest algorithm for the MPPT regulation of the photovoltaic system, making it more suitable for the technical field to which the present invention belongs. The present invention can effectively improve the control accuracy and response speed of the multi-channel MPPT of the photovoltaic system. DETAILED DESCRIPTION

[0014] Random forest is an integrated learning algorithm that has been widely used in recent years due to its ease of use and excellent performance. However, in the multi-channel MPPT control of photovoltaic inverters, due to the large number of system working parameters, the use of random forest algorithm consumes a lot of computing power. If more hardware costs are not invested, there will be obvious control delays.

[0015] In order to explore the potential and application scope of the random forest algorithm, we improved the random forest algorithm and successfully introduced it into the multi-channel MPPT control of the photovoltaic system, achieving good results. The specific technical solution is as follows: A. Obtain the input current, input voltage, output current and output voltage data of each boost converter; fit the working state of the photovoltaic panel connected to each boost converter, specifically including obtaining the output voltage and output current of each photovoltaic panel connected to the same boost converter, fitting the output voltage and output current with normal distribution respectively, eliminating the deviation data according to the 3σ principle, and then calculating the average output voltage and average output current as the working state of the photovoltaic panel after fitting. By screening and fitting the photovoltaic panels, the interference of faulty panels on the control of the boost converter can be reduced.

[0016] B. Calculate the working state of each boost converter. The working state of the boost converter includes the output voltage and output current of the boost converter. The random forest model includes a decision tree corresponding to each boost converter. Input the working state data of the boost converter into the random forest model to obtain the total fitting prediction output of the boost converter. By using the random forest model to predict the total output of the boost converter (that is, the output state of the inverter), the photovoltaic inverter can make timely parameter adjustments according to the changes in the input state, thereby improving the response speed of the entire system.

[0017] The construction process of the random forest model is to use the historical output voltage and historical output current of the boost converter as the training data set of the random forest model, form the output voltage and output current at the same time into a two-dimensional array, set the step size to Δt (the value of Δt is determined comprehensively according to the hardware system computing power and control accuracy. The higher the hardware system computing power and the higher the control accuracy requirement, the smaller the value of Δt) to traverse the training data set, obtain the first-order double dynamic reference data of the training data set, take the extreme point of the first-order double dynamic reference data fitting curve as the best segmentation feature, and use the training data set to recursively split the decision tree to obtain the original decision tree. The original decision tree is divided into two levels. The output of the first-level decision tree corresponds to the predicted output range of the boost converter. The second-level decision tree performs secondary prediction within the predicted output range provided by the first-level decision tree, and then the prediction results are averaged to obtain the total fitting prediction output of the boost converter. By optimizing the node topology of the decision tree and adopting the series structure of the output of the two-level decision tree in the entire random forest, the problem of overfitting in the traditional random forest in the process of multi-category data processing is solved.

[0018] Use test data to input each decision tree for testing, arrange the decision trees in descending order according to the stability of the output results of the decision trees, select the last half of the decision trees in the sequence to form the first-level decision trees, and the remaining decision trees form the second-level decision trees; when the first-level decision trees output their respective prediction results, the prediction results are mapped to the coordinate system to form prediction result nodes, and all the edge prediction result nodes in the coordinate system are used to enclose a closed area as the prediction output range.

[0019] C. The photovoltaic inverter performs MPPT control based on the total fitted predicted output of the boost converter; D. Perform MPPT control on each boost converter according to the adjusted working parameters of the photovoltaic inverter. After the working parameters of the boost converter are adjusted, the corresponding decision book is updated; during the update, the first-level decision tree and the second-level decision tree are partially updated synchronously, and the first-level decision tree and the second-level decision tree that have not been updated are used to maintain the total fitting prediction output of the boost converter. Whenever the working parameters of the boost converter are adjusted, in order to ensure the accuracy of the total fitting prediction output of the boost converter, the random forest model needs to be updated synchronously. The traditional random forest model needs to be updated offline, which cannot adapt to the continuous output working state of the photovoltaic inverter system. In the random forest model we constructed, since the two-level decision trees are relatively independent, this feature can be used to realize the online update of the random forest model.

[0020] The random forest model we built is not only suitable for boost circuit regulation of photovoltaic systems, but we also plan to extend it to inverter regulation, thereby reducing the software development cost and complexity of the entire photovoltaic grid-connected control system.

[0021] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A multi-channel MPPT control method for a photovoltaic inverter, characterized in that The following steps are involved: A. Obtain the input current, input voltage, output current and output voltage data of each boost converter; B. Calculate the working status of each boost converter, input the working status data of the boost converter into the random forest model, and obtain the total fitting prediction output of the boost converter; C. The photovoltaic inverter performs MPPT control based on the total fitted predicted output of the boost converter; D. Perform MPPT control on each boost converter according to the adjusted working parameters of the photovoltaic inverter.

2. The multi-channel MPPT control method of a photovoltaic inverter according to claim 1, characterized in that: In step A, the working state of the photovoltaic panel connected to each boost converter is fitted.

3. The multi-channel MPPT control method of a photovoltaic inverter according to claim 2, characterized in that: The output voltage and output current of each photovoltaic panel connected to the same boost converter are obtained, and normal distribution fitting is performed on the output voltage and output current respectively. The deviation data is eliminated according to the 3σ principle, and then the average output voltage and average output current are calculated as the working state of the photovoltaic panel after fitting.

4. The multi-channel MPPT control method of a photovoltaic inverter according to claim 3, characterized in that: In step B, the working state of the boost converter includes the output voltage and output current of the boost converter, and the random forest model includes decision trees corresponding to the boost converters one by one.

5. The multi-channel MPPT control method of a photovoltaic inverter according to claim 4, characterized in that: The historical output voltage and historical output current of the boost converter are used as the training data set of the random forest model. The output voltage and output current at the same time are composed into a two-dimensional array. A sliding window with a step size of Δt is set to traverse the training data set to obtain the first-order dual dynamic reference data of the training data set. The extreme points of the fitting curve of the first-order dual dynamic reference data are used as the best segmentation features. The decision tree is recursively split using the training data set to obtain the original decision tree.

6. The multi-channel MPPT control method of a photovoltaic inverter according to claim 5, characterized in that: The original decision tree is divided into two levels. The output of the first-level decision tree corresponds to the prediction output range of the boost converter. The second-level decision tree makes a secondary prediction within the prediction output range provided by the first-level decision tree. The prediction results are then averaged to obtain the total fitting prediction output of the boost converter.

7. The multi-channel MPPT control method of a photovoltaic inverter according to claim 6, characterized in that: Use test data to input each decision tree for testing, arrange the decision trees in descending order according to the stability of the output results of the decision trees, select the last half of the decision trees in the sequence to form the first-level decision trees, and the remaining decision trees form the second-level decision trees; when the first-level decision trees output their respective prediction results, the prediction results are mapped to the coordinate system to form prediction result nodes, and all the edge prediction result nodes in the coordinate system are used to enclose a closed area as the prediction output range.

8. The multi-channel MPPT control method of a photovoltaic inverter according to claim 7, characterized in that: In step D, after the working parameters of the boost converter are adjusted, the corresponding decision book is updated; during the update, the first-level decision tree and the second-level decision tree are partially updated synchronously, and the first-level decision tree and the second-level decision tree that have not been updated are used to maintain the total fitting prediction output of the boost converter.