A charging and discharging method of a photovoltaic energy storage system and an energy storage system thereof

Abstract

The present invention relates to the field of photovoltaic energy storage technology, and specifically to a charging and discharging method of a photovoltaic energy storage system and its energy storage system, comprising the following steps: obtaining power parameters of a photovoltaic unit, a load unit, a new energy device unit, and an energy storage unit in the photovoltaic energy storage system; using a neural network to perform deep learning on the mapping relationship of power parameters among the photovoltaic unit, the load unit, the new energy device unit, and the energy storage unit according to the power parameters, and constructing a charging parameter prediction model and a discharging parameter prediction model; twinning the charging parameter prediction model and the discharging parameter prediction model to obtain a charging and discharging coordination model; a battery management unit in the photovoltaic energy storage system performs charging and discharging coordination control on the energy storage unit according to the charging and discharging coordination model. The present invention performs charging and discharging coordination control on the energy storage unit according to the charging and discharging coordination model, so as to achieve autonomous optimization control of charging and discharging of photovoltaic energy storage, without the need for human intervention, and with higher timeliness.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic energy storage, and in particular to a charging and discharging method of a photovoltaic energy storage system and an energy storage system thereof. Background Art

[0002] Photovoltaic energy storage combines solar photovoltaic power generation systems with energy storage technology to store the electricity generated by photovoltaic power generation so that it can be supplied when needed. The electricity generated by the photovoltaic power generation system first meets its own load use, and the excess electricity can be sold to external power grid companies; if the amount of electricity generated by photovoltaic power generation is insufficient for load use, it will be supplemented by power from the external power grid. With the increasing perfection of photovoltaic power generation technology and the continuous reduction of equipment costs, photovoltaic energy storage systems have entered ordinary households. Home energy storage systems must consider maximizing the use of solar energy while not causing excessive fluctuations in public grid voltage.

[0003] When the current mainstream user-side power supply and distribution system introduces photovoltaic energy, user energy storage, electric vehicles and user loads, electric vehicles and charging piles are connected to the system through AC coupling, and the photovoltaic storage system and electric vehicle charging and discharging system are set up and operated separately according to their own energy management. Human intervention is required to coordinate them, resulting in reduced efficiency and failure to achieve intelligent management and optimal benefits of the entire user-side system. Summary of the invention

[0004] The purpose of the present invention is to provide a charging and discharging method of a photovoltaic energy storage system and its energy storage system, so as to solve the technical problem in the prior art that the photovoltaic storage system and the electric vehicle charging and discharging system are separately set up and operated according to their own energy management, and human intervention is required for coordination, resulting in reduced efficiency and failure to achieve intelligent management and optimal benefits of the entire user-side system.

[0005] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0006] A charging and discharging method for a photovoltaic energy storage system comprises the following steps:

[0007] Obtaining power parameters of photovoltaic units, load units, new energy equipment units and energy storage units in the photovoltaic energy storage system;

[0008] Using a neural network to perform deep learning on the mapping relationship of power parameters among the photovoltaic unit, the load unit, the new energy equipment unit and the energy storage unit according to the power parameters, and constructing a charging parameter prediction model and a discharging parameter prediction model;

[0009] The charging parameter prediction model and the discharging parameter prediction model are twinned to obtain a charging and discharging synergistic model;

[0010] The battery management unit in the photovoltaic energy storage system performs coordinated charging and discharging control on the energy storage unit based on the charge and discharge coordinated model.

[0011] As a preferred solution of the present invention, the power parameters of the photovoltaic unit are the power generation power of the photovoltaic unit, the power parameters of the load unit are the power consumption of the load unit, the power parameters of the new energy equipment unit include the charging power and discharging power of the new energy equipment, and the power parameters of the energy storage unit include the charging power and discharging power of the energy storage unit.

[0012] As a preferred solution of the present invention, the construction of the charging parameter prediction model includes:

[0013] The power generation power of the photovoltaic unit, the power consumption of the load unit, the charging power and the discharging power of the new energy equipment are used as input items of the first neural network, and the charging power of the energy storage unit is used as the output item of the first neural network;

[0014] The first neural network is used to learn the mapping relationship between the input item of the first neural network and the output item of the first neural network, so as to obtain a charging parameter prediction model for predicting the charging power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharge power of the new energy equipment;

[0015] The charging parameter prediction model is:

[0016] Din=BP1(K1,K2,K3,K4);

[0017] In the formula, Din is the charging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment respectively, and BP1 is the first neural network.

[0018] As a preferred solution of the present invention, the construction of the discharge parameter prediction model includes:

[0019] The power generation power of the photovoltaic unit, the power consumption of the load unit, the charging power and the discharging power of the new energy equipment are used as input items of the second neural network, and the discharging power of the energy storage unit is used as the output item of the second neural network;

[0020] The second neural network is used to learn the mapping relationship between the input items of the second neural network and the output items of the second neural network, so as to obtain a discharge parameter prediction model for predicting the discharge power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharge power of the new energy equipment;

[0021] The discharge parameter prediction model is:

[0022] Dout=BP2(K1,K2,K3,K4);

[0023] Where Dout is the discharge power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment respectively, and BP2 is the second neural network.

[0024] As a preferred solution of the present invention, the method for constructing the charge-discharge coordination model includes:

[0025] The mapping relationship between input items and output items in the discharge parameter prediction model and the charging parameter prediction model is reversed and combined into a twin network, which is:

[0026]

[0027] In the formula, Din is the charging power of the energy storage unit, Dout is the discharging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharging power of the new energy equipment respectively, BP1 is the first neural network, and BP2 is the second neural network;

[0028] The loss function of the twin network is corrected, and the twin network is retrained based on the corrected loss function to obtain the charge-discharge coordination model.

[0029] As a preferred solution of the present invention, the method for correcting the loss function of the twin network includes:

[0030] Quantify the prediction performance of the discharge parameter prediction model and the charge parameter prediction model;

[0031] Setting weights of the first neural network and the second neural network accordingly based on the prediction performance;

[0032] Using the weights of the first neural network and the second neural network to correct the loss function of the twin network, to obtain a corrected loss function;

[0033] The loss function of the twin network is:

[0034] Loss = CE[(K1, K2, K3, K4) BP1 ,(K1,K2,K3,K4) BP2 ];

[0035] Where Loss is the loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output term of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4)BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between

[0036] The modified loss function is:

[0037] Loss_amended=CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ];

[0038] Where Loss_amended is the corrected loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output of the second neural network, R1 is the weight of the first neural network, R2 is the weight of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between .

[0039] As a preferred embodiment of the present invention, the quantification of the prediction performance includes:

[0040] The prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified, and the prediction accuracy is:

[0041]

[0042] Where X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, and L t1 is the loss function of the discharge parameter prediction model from the initial energy storage moment to the current energy storage moment at time t1, L t2 is the loss function of the charging parameter prediction model at time t2 from the initial energy storage time to the current energy storage time, and T is the current energy storage time;

[0043] The prediction stability of the prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified, and the prediction stability is:

[0044]

[0045] Where, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charge parameter prediction model, and F t1 is Dout at time t1 from the initial energy storage moment to the current energy storage moment, F t2 is Din at time t2 from the initial energy storage time to the current energy storage time, T is the current energy storage time, and T0 is the initial energy storage time.

[0046] As a preferred solution of the present invention, the weight setting method of the first neural network and the second neural network is:

[0047] The prediction accuracy and prediction stability are decisively combined using a Softmax function to obtain the weight;

[0048] The weights are:

[0049] R1=Softmax(X1,Y1);

[0050] R2=Softmax(X2,Y2);

[0051] Wherein, R1 is the weight of the first neural network, R2 is the weight of the second neural network, X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charging parameter prediction model, Softmax is the Softmax function, Softmax(X1,Y1) is to select the optimal value between X1 and Y1, and Softmax(X2,Y2) is to select the optimal value between X2 and Y2.

[0052] As a preferred solution of the present invention, the loss functions of the discharge parameter prediction model and the charge parameter prediction model are the same, which are the cross entropy of the predicted value and the true value.

[0053] As a preferred solution of the present invention, the present invention provides a photovoltaic energy storage system, which is applied to a charging and discharging method of the photovoltaic energy storage system, and the system includes:

[0054] Photovoltaic units, battery management units, energy storage units, load units and new energy equipment units;

[0055] The photovoltaic unit is used to generate power using photovoltaic power generation;

[0056] The energy storage unit is used to store the power generated by the photovoltaic unit and the charging power of the new energy equipment unit;

[0057] Used to provide electrical power to load units and new energy equipment units;

[0058] The battery management unit is used to control the charging power of the energy storage unit to store the power generation power of the photovoltaic unit and the charging power of the new energy equipment unit based on the charge and discharge coordination model, and the discharge power to supply the power consumption of the load unit and the new energy equipment unit.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention uses a neural network to perform deep learning on the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units according to the power parameters, and constructs a charging parameter prediction model and a discharging parameter prediction model; the charging parameter prediction model and the discharging parameter prediction model are twinned to obtain a charging and discharging coordination model; the battery management unit in the photovoltaic energy storage system performs charging and discharging coordination control on the energy storage unit according to the charging and discharging coordination model, so as to achieve autonomous optimization control of charging and discharging of photovoltaic energy storage, without the need for human intervention, and with higher timeliness. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0062] Figure 1 A flow chart of a charging and discharging method for a photovoltaic energy storage system provided by an embodiment of the present invention;

[0063] Figure 2 A block diagram of a photovoltaic energy storage system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] like Figure 1 As shown, the present invention provides a charging and discharging method for a photovoltaic energy storage system.

[0066] Obtaining power parameters of photovoltaic units, load units, new energy equipment units and energy storage units in the photovoltaic energy storage system;

[0067] Using neural networks, we conduct deep learning on the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units (new energy vehicles) and energy storage units based on power parameters, and build charging parameter prediction models and discharging parameter prediction models.

[0068] The charging parameter prediction model and the discharging parameter prediction model are twinned to obtain a charging and discharging synergistic model;

[0069] The battery management unit in the photovoltaic energy storage system performs coordinated charging and discharging control on the energy storage unit based on the charge and discharge coordinated model.

[0070] The present invention uses a neural network to perform deep learning on the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units, and can construct a parameter prediction model that characterizes the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units, namely a charging parameter prediction model and a discharging parameter prediction model, so as to predict the charging power / discharging power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharging power of the new energy equipment.

[0071] Furthermore, in order to achieve the synergy of charging power / discharging power control of the energy storage unit, the present invention combines the charging parameter prediction model and the discharging parameter prediction model using a twin network, and ensures the charging and discharging power of the energy storage unit at the same time in the form of a twin network, with the same power generation power of the photovoltaic unit, power consumption of the load unit, and charging power and discharging power of the new energy equipment. This indicates that the charging and discharging power predicted by the model is synergistic, that is, the charging power that stores the power generation power of the photovoltaic unit and the charging power of the new energy equipment unit, and the discharge power that supplies the power consumption of the load unit and the new energy equipment unit are collaboratively controlled.

[0072] In order to improve the accuracy and safety of the model's control over the charging and discharging power of the energy storage unit, the present invention uses a loss function in the form of a twin network to transform the charging and discharging coordination model, that is, the loss function is corrected using the model's accuracy performance and stability performance, so that the charging parameter prediction model and the discharging parameter prediction model in the charging and discharging coordination model both achieve optimal prediction accuracy and prediction stability, thereby ensuring that the charging and discharging control accuracy and safety of the energy storage unit generated by the predicted charging and discharging power are in the best state.

[0073] Specifically, the present invention utilizes the accuracy performance of the model to correct the loss function conventionally used in the twin network (the difference between the two network structures in the twin network, the smaller the difference, the higher the charge and discharge synergy predicted by the charge and discharge synergy model, and the better the operation effect of the energy storage system). Adding accuracy performance correction to the conventional loss function can ensure that the charge and discharge synergy model trained in this way, when the model accuracy of the charging parameter prediction model and the discharge parameter prediction model is higher, the corresponding twin network structure has a higher weight, that is, the charge and discharge synergy model focuses on the prediction results of the network structure with high weight, and outputs the results with higher accuracy. Therefore, while the conventional loss function ensures that the charge and discharge synergy predicted by the charge and discharge synergy model is higher and the operation effect of the energy storage system is better, it further ensures that the accuracy of the charge and discharge control of the energy storage unit is higher.

[0074] The present invention utilizes the stability performance of the model to correct the loss function conventionally used in the twin network (the difference between the two network structures in the twin network, the smaller the difference, the higher the charge and discharge synergy predicted by the charge and discharge synergy model, and the better the operation effect of the energy storage system). By adding the stability performance correction to the conventional loss function, it is possible to ensure that the charge and discharge synergy model trained in this way has a higher weight in the case where the model accuracy of the charging parameter prediction model and the discharge parameter prediction model is higher, so that the charge and discharge synergy model focuses on the prediction results of the network structure with high weight, and outputs the results with higher stability. Therefore, while the conventional loss function ensures that the charge and discharge synergy predicted by the charge and discharge synergy model is higher and the operation effect of the energy storage system is better, it further ensures that the stability of the charge and discharge control of the energy storage unit is higher. The stability control corresponds to the safety of the energy storage system, which can be said to further ensure that the safety of the charge and discharge control of the energy storage unit is higher.

[0075] Furthermore, the present invention utilizes the classifier decision performance of the Softmax function to help the model make clear decisions between multiple categories. Therefore, it is used to autonomously decide the optimal correction index in terms of accuracy performance and stability performance / safety performance when correcting the loss function, thereby realizing autonomous optimization decision-making on precise control and safety control of charging and discharging of energy storage units, selecting the most appropriate or optimal control decision at the moment, and achieving autonomous optimization control of charging and discharging of photovoltaic energy storage without the need for human intervention, and with higher timeliness.

[0076] The power parameters of the photovoltaic unit are the power generation power of the photovoltaic unit, the power parameters of the load unit are the power consumption of the load unit, the power parameters of the new energy equipment unit include the charging power and discharging power of the new energy equipment, and the power parameters of the energy storage unit include the charging power and discharging power of the energy storage unit.

[0077] The present invention uses a neural network to perform deep learning on the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units, and can construct a parameter prediction model that characterizes the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units, namely a charging parameter prediction model and a discharging parameter prediction model, so as to predict the charging power / discharging power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption power of the load unit, and the charging power and discharging power of the new energy equipment, as follows:

[0078] The construction of the charging parameter prediction model includes:

[0079] The power generation power of the photovoltaic unit, the power consumption of the load unit, the charging power and the discharging power of the new energy equipment are used as input items of the first neural network, and the charging power of the energy storage unit is used as the output item of the first neural network;

[0080] The first neural network is used to learn the mapping relationship between the input item of the first neural network and the output item of the first neural network, so as to obtain a charging parameter prediction model for predicting the charging power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharge power of the new energy equipment;

[0081] The charging parameter prediction model is:

[0082] Din=BP1(K1,K2,K3,K4);

[0083] In the formula, Din is the charging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment respectively, and BP1 is the first neural network.

[0084] The construction of the discharge parameter prediction model includes:

[0085] The power generation power of the photovoltaic unit, the power consumption of the load unit, the charging power and the discharging power of the new energy equipment are used as input items of the second neural network, and the discharging power of the energy storage unit is used as the output item of the second neural network;

[0086] The second neural network is used to learn the mapping relationship between the input items of the second neural network and the output items of the second neural network, so as to obtain a discharge parameter prediction model for predicting the discharge power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharge power of the new energy equipment;

[0087] The discharge parameter prediction model is:

[0088] Dout=BP2(K1,K2,K3,K4);

[0089] Where Dout is the discharge power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment respectively, and BP2 is the second neural network.

[0090] In order to realize the coordination of charging power / discharging power control of energy storage unit, the present invention combines charging parameter prediction model and discharging parameter prediction model by using twin network, and ensures the charging and discharging power of energy storage unit at the same time in the form of twin network, with the same power generation power of photovoltaic unit, power consumption of load unit, charging power and discharging power of new energy equipment, indicating that the charging and discharging power predicted by the model is coordinated, that is, the charging power storing the power generation power of photovoltaic unit and the charging power of new energy equipment unit, and the discharge power supplying the power consumption of load unit and new energy equipment unit are coordinated controlled, as follows:

[0091] The construction method of the charge-discharge synergy model includes:

[0092] The mapping relationship between input items and output items in the discharge parameter prediction model and the charging parameter prediction model is reversed and combined into a twin network. The twin network is:

[0093]

[0094] In the formula, Din is the charging power of the energy storage unit, Dout is the discharging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharging power of the new energy equipment respectively, BP1 is the first neural network, and BP2 is the second neural network;

[0095] The loss function of the twin network is corrected, and the twin network is retrained based on the corrected loss function to obtain the charge-discharge coordination model.

[0096] The present invention uses a loss function in the form of a twin network to transform the charge-discharge coordination model, that is, the loss function is corrected by using the model accuracy performance and stability performance, so that the charging parameter prediction model and the discharging parameter prediction model in the charge-discharge coordination model both achieve the best prediction accuracy and prediction stability, thereby ensuring that the charge-discharge control accuracy and safety of the energy storage unit generated by the predicted charge-discharge power are in the best state, specifically as follows:

[0097] The correction methods of the loss function of the twin network include:

[0098] Quantify the prediction performance of the discharge parameter prediction model and the charge parameter prediction model;

[0099] correspondingly setting weights of the first neural network and the second neural network based on the prediction performance;

[0100] The loss function of the twin network is corrected using the weights of the first neural network and the second neural network to obtain a corrected loss function;

[0101] The loss function of the twin network is:

[0102] Loss = CE[(K1, K2, K3, K4) BP1 ,(K1,K2,K3,K4) BP2 ];

[0103] Where Loss is the loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output term of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between

[0104] The corrected loss function is:

[0105] Loss_amended=CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ];

[0106] Where Loss_amended is the corrected loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output of the second neural network, R1 is the weight of the first neural network, R2 is the weight of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between .

[0107] Quantification of predictive performance includes:

[0108] The prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified as follows:

[0109]

[0110] Where X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, and L t1 is the loss function of the discharge parameter prediction model from the initial energy storage moment to the current energy storage moment at time t1, L t2 is the loss function of the charging parameter prediction model at time t2 from the initial energy storage time to the current energy storage time, and T is the current energy storage time;

[0111] The present invention quantifies the degree of discreteness of the model loss function at each moment from the initial energy storage moment to the current energy storage moment through discreteness. The higher the degree of discreteness, the larger the model loss function. The loss function corresponds to the model prediction accuracy, so the worse the model prediction accuracy, the worse the charging and discharging control accuracy based on this result, and the worse the operation effect of the energy storage system.

[0112] The present invention utilizes the accuracy performance of the model to correct the loss function conventionally used in the twin network (the difference between the two network structures in the twin network, the smaller the difference, the higher the charge and discharge synergy predicted by the charge and discharge synergy model, and the better the operation effect of the energy storage system). By adding accuracy performance correction to the conventional loss function, it is possible to ensure that the charge and discharge synergy model trained in this way, when the model accuracy of the charging parameter prediction model and the discharge parameter prediction model is higher, the corresponding twin network structure has a higher weight, that is, the charge and discharge synergy model focuses on the prediction results of the network structure with a high weight, and outputs the results with a higher accuracy. Therefore, while the conventional loss function ensures that the charge and discharge synergy predicted by the charge and discharge synergy model is higher and the operation effect of the energy storage system is better, it further ensures that the accuracy of the charge and discharge control of the energy storage unit is higher.

[0113] The prediction stability of the prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified. The prediction stability is:

[0114]

[0115] Where, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charge parameter prediction model, and F t1 is Dout at time t1 from the initial energy storage moment to the current energy storage moment, F t2 is Din at time t2 from the initial energy storage time to the current energy storage time, T is the current energy storage time, and T0 is the initial energy storage time.

[0116] The present invention quantifies the discreteness of the charge and discharge power predicted by the model at each moment from the initial energy storage moment to the current energy storage moment through discreteness. The higher the discreteness, the worse the stability of the result predicted by the model, resulting in worse stability of the charge and discharge control based on the result, and lower safety of the energy storage system.

[0117] The present invention utilizes the stability performance of the model to correct the loss function conventionally used in the twin network (the difference between the two network structures in the twin network, the smaller the difference, the higher the charge and discharge synergy predicted by the charge and discharge synergy model, and the better the operation effect of the energy storage system). By adding the stability performance correction to the conventional loss function, it is possible to ensure that the charge and discharge synergy model trained in this way has a higher weight in the case where the model accuracy of the charging parameter prediction model and the discharge parameter prediction model is higher, so that the charge and discharge synergy model focuses on the prediction results of the network structure with high weight, and outputs the results with higher stability. Therefore, while the conventional loss function ensures that the charge and discharge synergy predicted by the charge and discharge synergy model is higher and the operation effect of the energy storage system is better, it further ensures that the stability of the charge and discharge control of the energy storage unit is higher. The stability control corresponds to the safety of the energy storage system, which can be said to further ensure that the safety of the charge and discharge control of the energy storage unit is higher.

[0118] Furthermore, the present invention utilizes the classifier decision performance of the Softmax function to help the model make clear decisions between multiple categories. Therefore, it is used to autonomously decide the optimal correction index in the accuracy performance and stability performance / safety performance when correcting the loss function, realize the autonomous optimization decision of the precise control and safety control of the energy storage unit charging and discharging, select the most appropriate or optimal control decision at the moment, and achieve the autonomous optimization control of the charging and discharging of photovoltaic energy storage without human intervention, and with higher timeliness, as follows:

[0119] The weight setting method for the first neural network and the second neural network is:

[0120] The Softmax function is used to make a decisive combination of prediction accuracy and prediction stability to obtain the weight;

[0121] The weights are:

[0122] R1=Softmax(X1,Y1);

[0123] R2=Softmax(X2,Y2);

[0124] Wherein, R1 is the weight of the first neural network, R2 is the weight of the second neural network, X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charging parameter prediction model, Softmax is the Softmax function, Softmax(X1,Y1) is to select the optimal value between X1 and Y1, and Softmax(X2,Y2) is to select the optimal value between X2 and Y2.

[0125] The loss functions of the discharge parameter prediction model and the charging parameter prediction model are the same, which are the cross entropy between the predicted value and the true value.

[0126] The loss functions of the discharge parameter prediction model and the charging parameter prediction model are the same.

[0127] like Figure 2 As shown, the present invention provides a photovoltaic energy storage system, which is applied to a charging and discharging method of the photovoltaic energy storage system, and the system includes:

[0128] Photovoltaic units, battery management units, energy storage units, load units and new energy equipment units;

[0129] The photovoltaic unit is used to generate power using photovoltaic power generation;

[0130] The energy storage unit is used to store the power generated by the photovoltaic unit and the charging power of the new energy equipment unit;

[0131] Used to provide electrical power to load units and new energy equipment units;

[0132] The battery management unit is used to control the charging power of the energy storage unit to store the generated power of the photovoltaic unit and the charging power of the new energy equipment unit based on the charge and discharge coordination model, as well as the discharge power to supply the power consumption of the load unit and the new energy equipment unit.

[0133] The present invention uses a neural network to perform deep learning on the mapping relationship of power parameters among photovoltaic units, load units, new energy equipment units and energy storage units according to power parameters, and constructs a charging parameter prediction model and a discharging parameter prediction model; the charging parameter prediction model and the discharging parameter prediction model are twinned to obtain a charging and discharging coordination model; the battery management unit in the photovoltaic energy storage system performs charging and discharging coordination control on the energy storage unit according to the charging and discharging coordination model, so as to achieve autonomous optimization control of charging and discharging of photovoltaic energy storage, without the need for human intervention, and with higher timeliness.

[0134] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A charging and discharging method for a photovoltaic energy storage system, characterized in that: The following steps are involved: Obtaining power parameters of photovoltaic units, load units, new energy equipment units and energy storage units in the photovoltaic energy storage system; Using a neural network to perform deep learning on the mapping relationship of power parameters among the photovoltaic unit, the load unit, the new energy equipment unit and the energy storage unit according to the power parameters, and constructing a charging parameter prediction model and a discharging parameter prediction model; The charging parameter prediction model and the discharging parameter prediction model are twinned to obtain a charging and discharging synergistic model; The battery management unit in the photovoltaic energy storage system performs charge and discharge coordinated control on the energy storage unit based on the charge and discharge coordinated model; The power parameter of the photovoltaic unit is the power generation power of the photovoltaic unit, the power parameter of the load unit is the power consumption of the load unit, the power parameter of the new energy device unit includes the charging power and the discharging power of the new energy device unit, the power parameter of the energy storage unit includes the charging power and the discharging power of the energy storage unit, the new energy device unit includes a new energy vehicle, and the photovoltaic unit uses photovoltaic power generation to provide power consumption to the load unit and the new energy device unit through the energy storage unit; The construction of the charging parameter prediction model includes: The power generation power of the photovoltaic unit, the power consumption power of the load unit, the charging power and the discharging power of the new energy equipment unit are used as input items of the first neural network, and the charging power of the energy storage unit is used as the output item of the first neural network; The first neural network is used to learn the mapping relationship between the input item of the first neural network and the output item of the first neural network, so as to obtain a charging parameter prediction model for predicting the charging power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, and the charging power and discharge power of the new energy equipment unit; The charging parameter prediction model is: Din=BP1(K1,K2,K3,K4); In the formula, Din is the charging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment unit respectively, and BP1 is the first neural network; The construction of the discharge parameter prediction model includes: The power generation power of the photovoltaic unit, the power consumption power of the load unit, the charging power and the discharging power of the new energy equipment unit are used as input items of the second neural network, and the discharging power of the energy storage unit is used as the output item of the second neural network; The second neural network is used to learn the mapping relationship between the input items of the second neural network and the output items of the second neural network, so as to obtain a discharge parameter prediction model for predicting the discharge power of the energy storage unit according to the power generation power of the photovoltaic unit, the power consumption of the load unit, the charging power and the discharge power of the new energy equipment unit; The discharge parameter prediction model is: Dout=BP2(K1,K2,K3,K4); Where Dout is the discharge power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharge power of the new energy equipment unit respectively, and BP2 is the second neural network; The method for constructing the charge-discharge coordination model includes: The mapping relationship between input items and output items in the discharge parameter prediction model and the charging parameter prediction model is reversed and combined into a twin network, which is: In the formula, Din is the charging power of the energy storage unit, Dout is the discharging power of the energy storage unit, K1 is the power generation power of the photovoltaic unit, K2 is the power consumption of the load unit, K3 and K4 are the charging power and discharging power of the new energy equipment unit respectively, BP1 is the first neural network, and BP2 is the second neural network; The loss function of the twin network is corrected, and the twin network is retrained based on the corrected loss function to obtain the charge-discharge coordination model.

2. A charging and discharging method for a photovoltaic energy storage system according to claim 1, characterized in that: The method for correcting the loss function of the twin network includes: Quantify the prediction performance of the discharge parameter prediction model and the charge parameter prediction model; Setting weights of the first neural network and the second neural network accordingly based on the prediction performance; Using the weights of the first neural network and the second neural network to correct the loss function of the twin network, to obtain a corrected loss function; The loss function of the twin network is: Loss=CE[(K1,K2,K3,K4) BP1 ,(K1,K2,K3,K4) BP2 ]; Where Loss is the loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output term of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between The modified loss function is: Loss_amended=CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ]; Where Loss_amended is the corrected loss function value, CE is the cross entropy operator, (K1, K2, K3, K4) BP1 is the output item of the first neural network, (K1, K2, K3, K4) BP2 is the output of the second neural network, R1 is the weight of the first neural network, R2 is the weight of the second neural network, CE[R1*(K1,K2,K3,K4) BP1 ,R2*(K1,K2,K3,K4) BP2 ] is R1*(K1,K2,K3,K4) BP1 and R2*(K1,K2,K3,K4) BP2 The cross entropy between .

3. A charging and discharging method for a photovoltaic energy storage system according to claim 2, characterized in that: The quantification of the predictive performance includes: The prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified, and the prediction accuracy is: Where X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, and L t1 is the loss function of the discharge parameter prediction model from the initial energy storage moment to the current energy storage moment at time t1, L t2 is the loss function of the charging parameter prediction model at time t2 from the initial energy storage time to the current energy storage time, and T is the current energy storage time; The prediction stability of the prediction accuracy of the discharge parameter prediction model and the charging parameter prediction model is quantified, and the prediction stability is: Where, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charge parameter prediction model, and F t1 is Dout at time t1 from the initial energy storage moment to the current energy storage moment, F t2 is Din at time t2 from the initial energy storage time to the current energy storage time, T is the current energy storage time, and T0 is the initial energy storage time.

4. A charging and discharging method for a photovoltaic energy storage system according to claim 3, characterized in that: The weight setting method of the first neural network and the second neural network is: The prediction accuracy and prediction stability are decisively combined using a Softmax function to obtain the weight; The weights are: R1=Softmax(X1,Y1); R2=Softmax(X2,Y2); Wherein, R1 is the weight of the first neural network, R2 is the weight of the second neural network, X1 is the prediction accuracy of the discharge parameter prediction model, X2 is the prediction accuracy of the charging parameter prediction model, Y1 is the prediction stability of the discharge parameter prediction model, Y2 is the prediction stability of the charging parameter prediction model, Softmax is the Softmax function, Softmax(X1,Y1) is to select the optimal value between X1 and Y1, and Softmax(X2,Y2) is to select the optimal value between X2 and Y2.

5. A charging and discharging method for a photovoltaic energy storage system according to claim 4, characterized in that: The loss functions of the discharge parameter prediction model and the charge parameter prediction model are the same, which are the cross entropy between the predicted value and the true value.

6. A photovoltaic energy storage system, characterized in that: A charging and discharging method for a photovoltaic energy storage system according to any one of claims 1 to 5, the system comprising: Photovoltaic units, battery management units, energy storage units, load units and new energy equipment units; The photovoltaic unit is used to generate power using photovoltaic power generation; The energy storage unit is used to store the power generated by the photovoltaic unit and the charging power of the new energy equipment unit; The energy storage unit is also used to provide electrical power to the load unit and the new energy equipment unit; The battery management unit is used to control the charging power of the energy storage unit to store the power generation power of the photovoltaic unit and the charging power of the new energy equipment unit, and the discharging power to supply the power consumption of the load unit and the new energy equipment unit based on the charge-discharge coordination model; The power parameter of the photovoltaic unit is the power generation power of the photovoltaic unit, the power parameter of the load unit is the power consumption of the load unit, the power parameters of the new energy equipment unit include the charging power and discharging power of the new energy equipment unit, the power parameters of the energy storage unit include the charging power and discharging power of the energy storage unit, the new energy equipment unit includes a new energy vehicle, and the photovoltaic unit uses photovoltaic power generation to provide power consumption to the load unit and the new energy equipment unit through the energy storage unit.

Images