Training method and system of radial basis function neural network model, control method of radial basis function neural network model and controller

Abstract

The invention provides a radial basis function neural network model training method and system, a control method thereof and a controller, and the method comprises the steps: determining a first threshold value and a first weight value when a radial basis function neural network is trained through a first training sample, the first threshold value is used for determining that training of the radial basis function neural network model is finished, and the first weight value is a weight value between a hidden layer and an output layer; when the radial basis function neural network runs with a first weight value and a first threshold value, training the radial basis function neural network with a second training sample to update the first weight value and the first threshold value, the first training sample comprising historical operation data corresponding to a previous moment to N + 1 moments before the current moment, and the second training sample comprising historical operation data corresponding to N + 1 moments before the current moment; the second training sample comprises historical operation data corresponding to N moments before the current moment. Wherein the trained radial basis function neural network can accurately express the relation in the energy system, and an accurate prediction result is given.

Description

Technical Field

[0001] This invention relates to the field of control technology, and more specifically, to a training method, system, and control method and controller for a radial basis function neural network model. Background Technology

[0002] A home energy management system consists of three parts: function, energy storage, and energy consumption. Each part involves a number of devices, and in order to achieve optimal energy dispatch, the parameters involved in all three parts need to be considered simultaneously.

[0003] The control strategies of existing energy systems are mostly based on mathematical models. However, for some factors, such as photovoltaic power generation on the energy supply side, which is greatly affected by weather and environmental factors, and on the energy consumption side, which is greatly affected by weather and environmental factors and user habits, existing mathematical models cannot effectively represent these factors and cannot fully utilize the efficiency of energy system use. Summary of the Invention

[0004] The present invention aims to at least solve the problem that existing mathematical models in the prior art or related technologies cannot effectively characterize the above-mentioned factors and cannot fully utilize the efficiency of energy systems.

[0005] Therefore, a first aspect of the present invention is to provide a method for training a radial basis function neural network model.

[0006] A second aspect of the present invention is that it provides a method for controlling an energy system.

[0007] A third aspect of the present invention is that a training apparatus for a radial basis function neural network model is provided.

[0008] A fourth aspect of the present invention is that a control device for an energy system is provided.

[0009] A fifth aspect of the invention is that it provides another training apparatus for a radial basis function neural network model.

[0010] A sixth aspect of the invention is that it provides another control device for an energy system.

[0011] A seventh aspect of the present invention is that a readable storage medium is provided.

[0012] An eighth aspect of the invention is that a controller is provided.

[0013] A ninth aspect of the present invention is that an energy system is provided.

[0014] In view of the above, according to a first aspect of the present invention, the present invention provides a training method for a radial basis function neural network (RBN) model, the RBN model comprising an input layer, a hidden layer, and an output layer connected in sequence, the training method comprising: when training the RBN model with a first training sample, determining a first threshold and a first weight value, the first threshold being a threshold used to determine the end of training of the RBN model, and the first weight value being a weight value between the hidden layer and the output layer; while the RBN model is running with the first weight value and the first threshold, training the RBN model with a second training sample to update the first weight value and the first threshold; wherein the first training sample includes historical running data corresponding to the previous time step to the N+1 time steps before the current time step, and the second training sample includes historical running data corresponding to the N time steps before the current time step.

[0015] This invention proposes a training method for a radial basis function neural network (RBN) model. In this method, the RBN model is trained using historical operating data of the energy system. The RBN model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, enabling the trained RBN model to accurately express the relationships within the energy system and provide accurate prediction results. This maximizes the efficiency of the energy system, that is, fully utilizes the efficiency of the energy system.

[0016] In this technical solution, the second training sample includes historical running data corresponding to the current moment. Therefore, the radial basis function neural network model can be trained in real time, thereby ensuring the accuracy of the radial basis function neural network model.

[0017] In related technical solutions, load forecasting models using BP (back propagation) neural networks and photovoltaic power generation forecasting models are used to control the energy system. However, these models suffer from slow convergence speed and are prone to getting trapped in local minima.

[0018] Among them, the BP (back propagation) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm.

[0019] In the technical solution of this application, the first training sample includes historical operating data corresponding to the previous N+1 times from the current time, and the second training sample includes historical operating data corresponding to the N times from the current time. The second training sample at a later time is the first training sample that discards the historical operating data corresponding to the N+1 times from the current time and adds the historical operating data corresponding to the current time. Obviously, the number of historical operating data contained in each training sample is fixed. Therefore, the radial basis function neural network model is a model that uses a fixed-time-length data window for rolling training. The fixed-time-length data window contains N consecutive historical operating data of the energy system.

[0020] In this process, the use of a radial basis function neural network model and a training method with a fixed-time-length data window rolling training can accelerate the convergence speed and avoid the problem of easily getting trapped in local minima.

[0021] The fixed time length is positively correlated with the value of N.

[0022] In the above technical solution, during the training process, the first weight value and the first threshold are updated to select appropriate weight values ​​and thresholds for the radial basis function neural network model. This allows for improved flexibility and efficiency of the energy system, reduced energy costs, and reduced environmental pollution when the trained radial basis function neural network model is used to control the energy system.

[0023] In addition, the training method for the radial basis function neural network model proposed in this application has the following additional technical features.

[0024] In some technical solutions, the training method may optionally include: obtaining the predicted value of the radial basis function neural network model when the second training sample is used as input; obtaining the actual value corresponding to the second training sample; determining the root mean square of the predicted value and the actual value; and determining that the training of the radial basis function neural network model ends when the root mean square is less than or equal to a first threshold.

[0025] In this technical solution, the objective function of the radial basis function neural network model is selected as the root mean square (RMS). In this process, it can be understood that the RMS is selected as the error metric function of the radial basis function neural network model. After obtaining the predicted value output by the radial basis function neural network model, the RMS of the predicted value and the actual value are calculated. The RMS is then compared with the first threshold to determine whether the radial basis function neural network model meets the training requirements, thereby determining whether to continue training the radial basis function neural network model.

[0026] Specifically, if the root mean square (RMS) is not higher than the first threshold, the radial basis function neural network (RBN) model is considered to meet the training requirements, and the training of the RBN model ends. Conversely, if the RMS is higher than the first threshold, the RBN model is considered not to meet the training requirements, and the RBN model continues to be trained until it meets the training requirements.

[0027] In some technical solutions, the hidden layer optionally includes at least one hidden node, which obtains the predicted value of the radial basis function neural network model when the second training sample is used as input. Specifically, this includes: obtaining the output value of the hidden node when the second training sample is used as input; determining the product of the output value and the first weight value; and determining the predicted value of the radial basis function neural network model based on the product value.

[0028] In this process, the radial basis function neural network model first performs clustering on the second training sample input to obtain the center vector of the transfer function of each hidden node. After determining the center vector, the weight values ​​from the hidden layer to the output layer are trained. In this process, training the weight values ​​from the hidden layer to the output layer is essentially an optimization problem of a system of linear equations.

[0029] In some technical solutions, the output value of the hidden node can optionally be calculated using the following formula:

[0030]

[0031] Where x is the second training sample, m is the number of nodes in the hidden layer, and u i (x) is the output value of the i-th hidden node, c i σ is the center vector when the action function of the i-th hidden node is chosen as a Gaussian function. i is the normalization constant of the i-th hidden node, and T is the transpose.

[0032] In some technical solutions, optionally, the output layer includes p output units, and the predicted values ​​of the radial basis function neural network model are calculated using the following formula:

[0033]

[0034] Among them, y j (n) is the predicted value output by the j-th output unit, w i,j It is the weight value corresponding to the i-th hidden node to the j-th output unit in the first weight value, z j It is the bias compensation for the j-th output unit, u i (x) is the output value of the i-th hidden node, x is the second training sample, and m is the number of nodes in the hidden layer.

[0035] In this technical solution, the output of the radial basis function neural network model, i.e. the predicted value, is the result obtained by performing bias compensation based on the sum of the products of the output value of the hidden node and the corresponding weight values ​​from the hidden layer to the output layer.

[0036] In some technical solutions, the historical operating data may optionally include at least one of the following: energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile operating mode, heat pump operating mode, household energy storage integrated machine operating mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature.

[0037] According to a second aspect of the present invention, the present invention provides a control method for an energy system, comprising: acquiring historical operating data of the energy system for N time points prior to the current time; inputting the historical operating data into a radial basis function neural network model trained by the training method of the radial basis function neural network model of the first aspect to obtain target prediction data; and controlling the energy system to operate according to the target prediction data.

[0038] In this technical solution, the radial basis function neural network model is a radial basis function neural network model trained using the training method of the radial basis function neural network model in the first aspect. In the training method of the radial basis function neural network model, the radial basis function neural network model is trained using historical operating data of the energy system. The radial basis function neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, so that the trained radial basis function neural network model can accurately express the relationship within the energy system and give accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the utilization efficiency of the energy system.

[0039] By inputting historical operating data of the energy system from N time points prior to the current moment into a trained radial basis function (RBF) neural network model, the model can output target prediction data. By controlling the energy system to operate according to the target prediction data, the trained RBF neural network model can accurately represent the relationships within the energy system and provide accurate prediction results, thereby maximizing the efficiency of the energy system's use, i.e., fully utilizing the energy system's efficiency.

[0040] In the process of controlling the energy system, we can improve the flexibility and efficiency of the energy system, reduce energy costs, and reduce environmental pollution.

[0041] According to a third aspect of the present invention, the present invention provides a training apparatus for a radial basis function neural network (RBN) model, the RBN model including an input layer, a hidden layer and an output layer connected in sequence, the training apparatus comprising: a determining unit, configured to determine a first threshold and a first weight value when training the RBN model with a first training sample, wherein the first threshold is a threshold used to determine the end of training of the RBN model, and the first weight value is a weight value between the hidden layer and the output layer; and an updating unit, configured to train the RBN model with a second training sample while the RBN model is running with the first weight value and the first threshold, to update the first weight value and the first threshold; wherein the first training sample includes historical running data corresponding to the previous time step to the N+1 time steps before the current time step, and the second training sample includes historical running data corresponding to the N time steps before the current time step.

[0042] This invention proposes a training device for a radial basis function neural network (RBN) model. The RBN model is trained using historical operating data of an energy system. The RBN model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand. This allows the trained RBN model to accurately express the relationships within the energy system and provide accurate prediction results, thereby maximizing the efficiency of the energy system and making full use of its efficiency.

[0043] In related technical solutions, load forecasting models using BP (back propagation) neural networks and photovoltaic power generation forecasting models are used to control the energy system. However, these models suffer from slow convergence speed and are prone to getting trapped in local minima.

[0044] In this technical solution, the second training sample includes historical running data corresponding to the current moment. Therefore, the radial basis function neural network model can be trained in real time, thereby ensuring the accuracy of the radial basis function neural network model.

[0045] Among them, the BP (back propagation) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm.

[0046] In the technical solution of this application, the first training sample includes historical operating data corresponding to the previous N+1 times from the current time, and the second training sample includes historical operating data corresponding to the N times from the current time. The second training sample at a later time is the first training sample that discards the historical operating data corresponding to the N+1 times from the current time and adds the historical operating data corresponding to the current time. Obviously, the number of historical operating data contained in each training sample is fixed. Therefore, the radial basis function neural network model is a model that uses a fixed-time-length data window for rolling training. The fixed-time-length data window contains N consecutive historical operating data of the energy system.

[0047] In this process, the use of a radial basis function neural network model and a training method with a fixed-time-length data window rolling training can accelerate the convergence speed and avoid the problem of easily getting trapped in local minima.

[0048] The fixed time length is positively correlated with the value of N.

[0049] In the above technical solution, during the training process, the first weight value and the first threshold are updated to select appropriate weight values ​​and thresholds for the radial basis function neural network model. This allows for improved flexibility and efficiency of the energy system, reduced energy costs, and reduced environmental pollution when the trained radial basis function neural network model is used to control the energy system.

[0050] In addition, the training device for the radial basis function neural network model proposed in this application has the following additional technical features.

[0051] In some technical solutions, the update unit may optionally also be used to obtain the predicted value of the radial basis function neural network model when the second training sample is used as input; obtain the actual value corresponding to the second training sample; determine the root mean square of the predicted value and the actual value; and determine that the training of the radial basis function neural network model ends when the root mean square is less than or equal to a first threshold.

[0052] In this technical solution, the objective function of the radial basis function neural network model is selected as the root mean square (RMS). In this process, it can be understood that the RMS is selected as the error metric function of the radial basis function neural network model. After obtaining the predicted value output by the radial basis function neural network model, the RMS of the predicted value and the actual value are calculated. The RMS is then compared with the first threshold to determine whether the radial basis function neural network model meets the training requirements, thereby determining whether to continue training the radial basis function neural network model.

[0053] Specifically, if the root mean square (RMS) is not higher than the first threshold, the radial basis function neural network (RBN) model is considered to meet the training requirements, and the training of the RBN model ends. Conversely, if the RMS is higher than the first threshold, the RBN model is considered not to meet the training requirements, and the RBN model continues to be trained until it meets the training requirements.

[0054] In some technical solutions, the hidden layer optionally includes at least one hidden node, and the update unit is specifically used to: obtain the output value of the hidden node when the second training sample is used as input; determine the product value of the output value and the first weight value; and determine the predicted value of the radial basis function neural network model based on the product value.

[0055] In this process, the radial basis function neural network model first performs clustering on the second training sample input to obtain the center vector of the transfer function of each hidden node. After determining the center vector, the weight values ​​from the hidden layer to the output layer are trained. In this process, training the weight values ​​from the hidden layer to the output layer is essentially an optimization problem of a system of linear equations.

[0056] In some technical solutions, the output value of the hidden node can optionally be calculated using the following formula:

[0057]

[0058] Where x is the second training sample, m is the number of nodes in the hidden layer, and u i (x) is the output value of the i-th hidden node, c i σ is the center vector when the action function of the i-th hidden node is chosen as a Gaussian function. i is the normalization constant of the i-th hidden node, and T is the transpose.

[0059] In some technical solutions, optionally, the output layer includes p output units, and the predicted values ​​of the radial basis function neural network model are calculated using the following formula:

[0060]

[0061] Among them, y j (n) is the predicted value output by the j-th output unit, w i,j It is the weight value corresponding to the i-th hidden node to the j-th output unit in the first weight value, z j It is the bias compensation for the j-th output unit, u i (x) is the output value of the i-th hidden node, x is the second training sample, and m is the number of nodes in the hidden layer.

[0062] In some technical solutions, the historical operating data may optionally include at least one of the following: energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile operating mode, heat pump operating mode, household energy storage integrated machine operating mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature.

[0063] According to a fourth aspect of the present invention, the present invention provides a control device for an energy system, comprising: an acquisition unit for acquiring historical operating data of the energy system for N time points prior to the current time; a prediction unit for inputting the historical operating data into a radial basis neural network model trained by the training method of the radial basis neural network model of the first aspect, to obtain target prediction data; and a control unit for controlling the energy system to operate according to the target prediction data.

[0064] In this technical solution, the radial basis function neural network model is a radial basis function neural network model trained using the training method of the radial basis function neural network model in the first aspect. In the training method of the radial basis function neural network model, the radial basis function neural network model is trained using historical operating data of the energy system. The radial basis function neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, so that the trained radial basis function neural network model can accurately express the relationship within the energy system and give accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the utilization efficiency of the energy system.

[0065] By inputting historical operating data of the energy system from N time points prior to the current moment into a trained radial basis function (RBF) neural network model, the model can output target prediction data. When the energy system is controlled to operate according to the target prediction data, the trained RBF neural network model can accurately represent the relationships within the energy system and provide accurate prediction results.

[0066] In the process of controlling the energy system, we can improve the flexibility and efficiency of the energy system, reduce energy costs, and reduce environmental pollution.

[0067] According to a fifth aspect of the present invention, the present invention provides a training apparatus for a radial basis function neural network model, comprising a first processor and a first memory, the first memory storing a program or instructions executable on the first processor, wherein the program or instructions, when executed by the first processor, implement the steps of the training method for a radial basis function neural network model as described above.

[0068] According to a sixth aspect of the present invention, the present invention provides a control device for an energy system, including a second processor and a second memory, the second memory storing a program or instructions executable on the second processor, the program or instructions being executed by the second processor to implement the steps of the energy system control method described above.

[0069] According to a seventh aspect of the present invention, a readable storage medium is provided on which a program or instructions are stored, which, when executed by a processor, implement the steps of any of the methods described above.

[0070] According to an eighth aspect of the present invention, the present invention provides a controller comprising: a training device for any of the radial basis function neural network models described above; and / or a control device for any of the energy systems described above; and / or a readable storage medium as described above.

[0071] According to a ninth aspect of the present invention, an energy system is provided, comprising: a controller as described above.

[0072] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0073] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0074] Figure 1 A flowchart illustrating a training method for a radial basis function neural network model according to an embodiment of the present invention is shown.

[0075] Figure 2 A schematic diagram of a radial basis function neural network model according to an embodiment of the present invention is shown;

[0076] Figure 3 A flowchart illustrating a control method for an energy system according to an embodiment of the present invention is shown;

[0077] Figure 4 A schematic block diagram corresponding to a control method for an energy system in an embodiment of the present invention is shown;

[0078] Figure 5 A schematic block diagram of a training device for a radial basis function neural network model according to an embodiment of the present invention is shown;

[0079] Figure 6 A schematic block diagram of a control device for an energy system according to an embodiment of the present invention is shown;

[0080] Figure 7 A schematic block diagram of a training device for another radial basis neural network model in an embodiment of the present invention is shown;

[0081] Figure 8 A schematic block diagram of a control device for another energy system according to an embodiment of the present invention is shown. Detailed Implementation

[0082] To better understand the above aspects, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0083] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0084] In one embodiment of this application, such as Figure 1 As shown, a training method for a radial basis function neural network model is provided. The radial basis function neural network model includes an input layer, a hidden layer, and an output layer connected in sequence. The training method includes:

[0085] Step 102: When training the radial basis neural network model with the first training sample, determine the first threshold and the first weight value. The first threshold is the threshold used to determine the end of the training of the radial basis neural network model, and the first weight value is the weight value between the hidden layer and the output layer.

[0086] Step 104: While the radial basis function neural network model is running with the first weight value and the first threshold, the radial basis function neural network model is trained with the second training sample to update the first weight value and the first threshold.

[0087] The first training sample includes historical running data corresponding to the previous N+1 times from the current time, and the second training sample includes historical running data corresponding to the N times from the current time.

[0088] This invention proposes a training method for a radial basis function (RBF) neural network model. In this RBF neural network training method, the RBF neural network model is trained using historical operating data of the energy system. The RBF neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, enabling the trained RBF neural network model to accurately express the relationship within the energy system and provide accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the energy system's utilization efficiency.

[0089] In this embodiment, the second training sample includes historical running data corresponding to the current moment. Therefore, the radial basis function neural network model can be trained in real time, thereby ensuring the accuracy of the radial basis function neural network model.

[0090] In related embodiments, a load forecasting model based on a back propagation (BP) neural network and a photovoltaic power generation forecasting model are used to control the energy system. However, the above models have the problems of slow convergence speed and easy to get trapped in local minima.

[0091] Among them, the BP (back propagation) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm.

[0092] In the embodiments of this application, the first training sample includes historical operating data corresponding to the previous N+1 times from the current time, and the second training sample includes historical operating data corresponding to the N times from the current time. The second training sample at a later time is the first training sample that discards the historical operating data corresponding to the N+1 times from the current time and adds the historical operating data corresponding to the current time. Obviously, the number of historical operating data contained in each training sample is fixed. Therefore, the radial basis function neural network model is a model that uses a data window of fixed time length for rolling training. The data window of fixed time length also contains the historical operating data of the energy system for N consecutive sampling times.

[0093] In this process, the use of a radial basis function neural network model and a training method with a fixed-time-length data window rolling training can accelerate the convergence speed and avoid the problem of easily getting trapped in local minima.

[0094] The fixed time length is positively correlated with the value of N.

[0095] In the above embodiments, during the training process, the first weight value and the first threshold are updated to select appropriate weight values ​​and thresholds for the radial basis function neural network model. This allows for improved flexibility and efficiency of the energy system, reduced energy costs, and reduced environmental pollution when the trained radial basis function neural network model is used to control the energy system.

[0096] In some embodiments, the training method may optionally further include: obtaining the predicted value of the radial basis function neural network model when the second training sample is used as input; obtaining the actual value corresponding to the second training sample; determining the root mean square of the predicted value and the actual value; and determining that the training of the radial basis function neural network model ends when the root mean square is less than or equal to a first threshold.

[0097] In this embodiment, the objective function of the radial basis function neural network model is selected as the root mean square (RMS). In this process, it can be understood that the RMS is selected as the error metric function of the radial basis function neural network model. After obtaining the predicted value output by the radial basis function neural network model, the RMS of the predicted value and the actual value is calculated so that the RMS is compared with the first threshold to determine whether the radial basis function neural network model meets the training requirements, thereby determining whether to continue training the radial basis function neural network model.

[0098] Specifically, if the root mean square (RMS) is not higher than the first threshold, the radial basis function neural network (RBN) model is considered to meet the training requirements, and the training of the RBN model ends. Conversely, if the RMS is higher than the first threshold, the RBN model is considered not to meet the training requirements, and the RBN model continues to be trained until it meets the training requirements.

[0099] In some embodiments, the hidden layer optionally includes at least one hidden node, which obtains the predicted value of the radial basis function neural network model when the second training sample is used as input, specifically including: obtaining the output value of the hidden node when the second training sample is used as input; determining the product of the output value and the first weight value; and determining the predicted value of the radial basis function neural network model based on the product value.

[0100] In this process, the radial basis function neural network model first performs clustering on the second training sample input to obtain the center vector of the transfer function of each hidden node. After determining the center vector, the weight values ​​from the hidden layer to the output layer are trained. In this process, training the weight values ​​from the hidden layer to the output layer is essentially an optimization problem of a system of linear equations.

[0101] In some embodiments, the output value of the hidden node is optionally calculated using the following formula:

[0102]

[0103] Where x is the second training sample, m is the number of nodes in the hidden layer, and u i (x) is the output value of the i-th hidden node, c i σ is the center vector when the action function of the i-th hidden node is chosen as a Gaussian function. i is the normalization constant of the i-th hidden node, and T is the transpose.

[0104] In some embodiments, optionally, the output layer includes p output units, and the predictions of the radial basis function neural network model are calculated using the following formula:

[0105]

[0106] Among them, yj (n) is the predicted value output by the j-th output unit, w i,j It is the weight value corresponding to the i-th hidden node to the j-th output unit in the first weight value, z j It is the bias compensation for the j-th output unit, u i (x) is the output value of the i-th hidden node, x is the second training sample, and m is the number of nodes in the hidden layer.

[0107] In this embodiment, the output of the radial basis function neural network model, i.e. the predicted value, is the result obtained by performing bias compensation based on the sum of the products of the output value of the hidden node and the corresponding weight values ​​from the hidden layer to the output layer.

[0108] In some embodiments, the historical operating data may optionally include at least one of the following: energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile operating mode, heat pump operating mode, household energy storage integrated machine operating mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature.

[0109] In one embodiment, the energy system is exemplified by a photovoltaic, energy storage, charging, and heating home energy management and dispatching system.

[0110] In establishing the radial basis function neural network model, the main control parameters include the energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile operating mode, heat pump operating mode, integrated household energy storage unit operating mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature. These parameters exhibit time-varying, nonlinear, and correlated characteristics. Based on the analysis of the photovoltaic-energy storage-charging-heating home energy management and scheduling system, a radial basis function neural network model is used to construct a home energy management and scheduling strategy model.

[0111] Among them, such as Figure 2 As shown, with 9 output units, the output of the radial basis function neural network model is y. i (n), i = 1, 2, 3...9. At this time, the input to the radial basis function neural network model is the historical data of these 9 control parameters, light intensity, ambient temperature, and ambient humidity.

[0112] For example, we select one week's data for modeling. The data for one week is statistically analyzed by hour. There are 7 days in a week and 24 hours in a day, so there are 168 data points in a group. The photovoltaic, energy storage, charging and heating home energy management and scheduling model can be expressed by the following formula.

[0113] y i(n) = f(I1(n - 1), I2(n - 1), I3(n - 1), y1(n - 1), y2(n - 1), y3(n - 1), y3(n - 1), y4(n - 1), y5(n - 1), y6(n - 1), I1(n - 2), I2(n - 2), I3(n - 2), y1(n - 2), y2(n - 2), y3(n - 2), y3(n - 2), y4(n - 2), y5(n - 2), y6(n - 2), …… I1(n - 168), I2(n - 168), I3(n - 168), y1(n - 168), y2(n - 168), y3(n - 168), y3(n - 168), y4(n - 168), y5(n - 168), y6(n - 168))

[0114] where y iy1(n) is the output of the i-th output unit; y2(n) is the current value of the energy storage battery charging power; y3(n) is the current value of the grid power purchase and sale; y4(n) is the current value of the charging pile working mode; y5(n) is the current value of the heat pump working mode; y6(n) is the current value of the household energy storage integrated machine working mode; y7(n) is the current value of the charging pile dynamic charging power; y8(n) is the current value of the heat pump water temperature; y9(n) is the current value of the heat pump air temperature; y1(n-1) is the value of the light intensity one time before the current sampling time; y2(n-1) is the value of the ambient temperature one time before the current sampling time. y1(n-1) is the value at time 1; y2(n-1) is the value at time 1 before the current sampling time of the ambient humidity; y3(n-1) is the value at time 1 before the current sampling time of the energy storage battery charging power; y4(n-1) is the value at time 1 before the current sampling time of the energy storage battery discharging power; y5(n-1) is the value at time 1 before the current sampling time of the energy storage battery discharging power; y6(n-1) is the value at time 1 before the current sampling time of the energy grid purchase and sale power; y7(n-1) is the value at time 1 before the current sampling time of the charging pile operating mode; y8(n-1) is the value at time 1 before the current sampling time of the heat pump operating mode; y9(n-1) is the value at time 1 before the current sampling time of the integrated home energy storage unit operating mode; y1(n-1) is the value at time 1 before the current sampling time of the charging pile dynamic charging power. The values ​​of the current sampling time and the time before sampling time are: y8(n-1) is the heat pump water temperature and the time before sampling time; y9(n-1) is the heat pump air temperature and the time before sampling time; ... I1(n-168) is the light intensity and the time before sampling time; I2(n-168) is the ambient temperature and the time before sampling time; I3(n-168) is the ambient humidity and the time before sampling time; y1(n-168) is the energy storage battery charging power and the time before sampling time; y2(n-168) is the energy storage battery discharging power and the time before sampling time; y3 (n-168) represents the power purchased and sold by the power grid 168 moments before the current sampling time; y4(n-168) represents the charging pile operating mode 168 moments before the current sampling time; y5(n-168) represents the heat pump operating mode 168 moments before the current sampling time; y6(n-168) represents the household storage integrated machine operating mode 168 moments before the current sampling time; y7(n-168) represents the charging pile dynamic charging power 168 moments before the current sampling time; y8(n-168) represents the heat pump water temperature 168 moments before the current sampling time; y9(n-168) represents the heat pump air temperature 168 moments before the current sampling time.

[0115] Specifically, assume that y1(n) is the current value of the energy storage battery charging power; y2(n) is the current value of the energy storage battery discharging power; y3(n) is the current value of the power purchased and sold by the grid; y4(n) is the current value of the charging pile working mode; y5(n) is the current value of the heat pump working mode; y6(n) is the current value of the household energy storage integrated machine working mode; y7(n) is the current value of the dynamic charging power of the charging pile; y8(n) is the current value of the heat pump water temperature; y9(n) is the current value of the heat pump air temperature; and the system output variable group Y(n) = (y1(n), y2(n), y3(n), y4(n), y5(n), y6(n), y7(n), y8(n), y9(n)).

[0116] The energy management system scheduling strategy model uses data from the previous 168 time points (t=n-1, t=n-2, t=n-3, t=n-4, t=n-5, t=n-6, ..., t=n-168) to predict the control data for the current time point (t=n). The next prediction discards the data from time point t=n-168, adds the calculated data from time point t=n as the new data for time point t=n-1, and shifts the data from the remaining time points sequentially to form a new sample for calculating the data for the next time point.

[0117] The specific modeling process is as follows: First, the network is simulated using data from the first 168 sampling times. After training the network, the network obtains weights and thresholds, predicts the value at the 169th time, and then the value at the 169th time is added in, while the value at the 1st time is discarded. The network is then retrained to obtain new weights and thresholds. The calculation continues to obtain the value at the 170th time, and so on, thereby achieving rolling optimization and updating of parameters.

[0118] In one embodiment, such as Figure 3 As shown, the present invention provides a control method for an energy system, comprising:

[0119] Step 302: Obtain the historical operating data of the energy system for the N moments prior to the current moment;

[0120] Step 304: Input the historical running records into the radial basis function neural network model to obtain target prediction data;

[0121] Step 306: Control the energy system to operate according to the target predicted data.

[0122] Among them, the radial basis function neural network model is a model trained using the radial basis function neural network model training method.

[0123] In this embodiment, the radial basis function neural network model is a radial basis function neural network model trained using a radial basis function neural network model training method. In this method, the radial basis function neural network model is trained using historical operating data of the energy system. The radial basis function neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, so that the trained radial basis function neural network model can accurately express the relationship within the energy system and give accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the utilization efficiency of the energy system.

[0124] By inputting historical operating data of the energy system from N time points prior to the current moment into a trained radial basis function (RBF) neural network model, the model can output target prediction data. When the energy system is controlled to operate according to the target prediction data, the trained RBF neural network model can accurately represent the relationships within the energy system and provide accurate prediction results.

[0125] In the process of controlling the energy system, we can improve the flexibility and efficiency of the energy system, reduce energy costs, and reduce environmental pollution.

[0126] In one embodiment, such as Figure 4 As shown, the control method of the energy system is divided into the following parts: data collection module 402, data preprocessing module 404, RBF network model 406, rolling optimization module 408, execution and monitoring module 410, home energy system 412, and feedback and learning module 414.

[0127] The data collection module 402 collects household energy usage data, including: electricity purchased and sold from the grid, photovoltaic power generation, temperature, humidity, time, and electricity consumption of key loads. Key loads may include charging piles, heat pumps, and energy storage battery power. It also collects external data such as market electricity prices, weather forecasts, and holiday information.

[0128] The data preprocessing module 404 cleans and preprocesses the collected data, including missing value handling, outlier detection, and data standardization. It converts the time series data into a format suitable for input to the RBF network model 406.

[0129] The RBF network model 406 is used to predict energy demand, energy supply and cost in the future. The RBF network model 406 consists of an input layer, a hidden layer and an output layer. The hidden layer uses radial basis functions as activation functions and adjusts the network parameters through the training dataset to optimize the prediction performance.

[0130] The rolling optimization module 408 uses optimization algorithms, such as gradient descent or genetic algorithms, to continuously optimize the energy dispatch strategy over a predicted future time period. The goal is to maximize energy efficiency, minimize costs, and reduce carbon emissions, dynamically adjusting the energy use strategy while taking into account factors such as real-time market electricity prices and weather changes.

[0131] The execution and monitoring module 410, based on the optimized strategy, performs energy scheduling through intelligent devices such as smart meters, solar panel controllers, and intelligent temperature control systems, monitors energy usage in real time, and collects feedback data for continuous learning and optimization of the model.

[0132] The feedback and learning module 414 collects actual energy usage and cost data during execution. This data is compared with predicted data to evaluate the effectiveness of the strategy. This feedback information is used to adjust the RBF network model 406 and the optimization algorithm to improve the accuracy of future predictions and scheduling.

[0133] In this embodiment, the rolling optimization household energy dispatching strategy based on the RBF network model can achieve dynamic and efficient management of household energy use, while taking into account factors such as cost, environmental impact and market changes.

[0134] In one embodiment, such as Figure 5 As shown, the present invention provides a training device 500 for a radial basis function neural network (RBN) model. The RBN model includes an input layer, a hidden layer, and an output layer connected in sequence. The training device includes: a determining unit 502, used to determine a first threshold and a first weight value when training the RBN model with a first training sample. The first threshold is a threshold used to determine the end of the RBN model training, and the first weight value is the weight value between the hidden layer and the output layer; and an updating unit 504, used to train the RBN model with a second training sample while the RBN model is running with the first weight value and the first threshold, so as to update the first weight value and the first threshold. The first training sample includes historical running data corresponding to the previous time step to the N+1 time steps before the current time step, and the second training sample includes historical running data corresponding to the N time steps before the current time step.

[0135] This invention proposes a training device 500 for a radial basis function neural network model. The radial basis function neural network model is trained using historical operating data of the energy system. The radial basis function neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand. This allows the trained radial basis function neural network model to accurately express the relationship within the energy system and provide accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the utilization efficiency of the energy system.

[0136] In this embodiment, the second training sample includes historical running data corresponding to the current moment. Therefore, the radial basis function neural network model can be trained in real time, thereby ensuring the accuracy of the radial basis function neural network model.

[0137] In related embodiments, a load forecasting model based on a back propagation (BP) neural network and a photovoltaic power generation forecasting model are used to control the energy system. However, the above models have the problems of slow convergence speed and easy to get trapped in local minima.

[0138] Among them, the BP (back propagation) neural network is a multi-layer feedforward neural network trained according to the error back propagation algorithm.

[0139] In the embodiments of this application, the first training sample includes historical operating data corresponding to the previous N+1 times from the current time, and the second training sample includes historical operating data corresponding to the N times from the current time. The second training sample at a later time is the first training sample that discards the historical operating data corresponding to the N+1 times from the current time and adds the historical operating data corresponding to the current time. Obviously, the number of historical operating data contained in each training sample is fixed. Therefore, the radial basis function neural network model is a model that uses a data window of fixed time length for rolling training. The data window of fixed time length also contains the historical operating data of the energy system for N consecutive sampling times.

[0140] In this process, the use of a radial basis function neural network model and a training method with a fixed-time-length data window rolling training can accelerate the convergence speed and avoid the problem of easily getting trapped in local minima.

[0141] The fixed time length is positively correlated with the value of N.

[0142] In the above embodiments, during the training process, the first weight value and the first threshold are updated to select appropriate weight values ​​and thresholds for the radial basis function neural network model. This allows for improved flexibility and efficiency of the energy system, reduced energy costs, and reduced environmental pollution when the trained radial basis function neural network model is used to control the energy system.

[0143] In some embodiments, optionally, the update unit 504 is further configured to obtain the predicted value of the radial basis function neural network model when the second training sample is used as input; obtain the actual value corresponding to the second training sample; determine the root mean square of the predicted value and the actual value; and determine that the training of the radial basis function neural network model ends when the root mean square is less than or equal to a first threshold.

[0144] In this embodiment, the objective function of the radial basis function neural network model is selected as the root mean square (RMS). In this process, it can be understood that the RMS is selected as the error metric function of the radial basis function neural network model. After obtaining the predicted value output by the radial basis function neural network model, the RMS of the predicted value and the actual value is calculated so that the RMS is compared with the first threshold to determine whether the radial basis function neural network model meets the training requirements, thereby determining whether to continue training the radial basis function neural network model.

[0145] Specifically, if the root mean square (RMS) is not higher than the first threshold, the radial basis function neural network (RBN) model is considered to meet the training requirements, and the training of the RBN model ends. Conversely, if the RMS is higher than the first threshold, the RBN model is considered not to meet the training requirements, and the RBN model continues to be trained until it meets the training requirements.

[0146] In some embodiments, the hidden layer may optionally include at least one hidden node, and the update unit 504 is specifically configured to: obtain the output value of the hidden node when the second training sample is used as input; determine the product of the output value and the first weight value; and determine the predicted value of the radial basis function neural network model based on the product value.

[0147] In this process, the radial basis function neural network model first performs clustering on the second training sample input to obtain the center vector of the transfer function of each hidden node. After determining the center vector, the weight values ​​from the hidden layer to the output layer are trained. In this process, training the weight values ​​from the hidden layer to the output layer is essentially an optimization problem of a system of linear equations.

[0148] In some embodiments, the output value of the hidden node is optionally calculated using the following formula:

[0149]

[0150] Where x is the second training sample, m is the number of nodes in the hidden layer, and u i (x) is the output value of the i-th hidden node, c i σ is the center vector when the action function of the i-th hidden node is chosen as a Gaussian function. i is the normalization constant of the i-th hidden node, and T is the transpose.

[0151] In some embodiments, optionally, the output layer includes p output units, and the predictions of the radial basis function neural network model are calculated using the following formula:

[0152]

[0153] Among them, y j(n) is the predicted value output by the j-th output unit, w i,j It is the weight value corresponding to the i-th hidden node to the j-th output unit in the first weight value, z j It is the bias compensation for the j-th output unit, u i (x) is the output value of the i-th hidden node, x is the second training sample, and m is the number of nodes in the hidden layer.

[0154] In some embodiments, the historical operating data may optionally include at least one of the following: energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile operating mode, heat pump operating mode, household energy storage integrated machine operating mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature.

[0155] In one embodiment, such as Figure 6 As shown, the present invention provides a control device 600 for an energy system, comprising: an acquisition unit 602 for acquiring historical operating data of the energy system for N time periods prior to the current time; a prediction unit 604 for inputting the historical operating data into a radial basis function neural network model to obtain target prediction data; and a control unit 606 for controlling the energy system to operate according to the target prediction data.

[0156] Among them, the radial basis function neural network model is a model trained using the radial basis function neural network model training method.

[0157] In this embodiment, the radial basis function neural network model is a radial basis function neural network model trained using the training method described above. In this training method, the radial basis function neural network model is trained using historical operating data of the energy system. The radial basis function neural network model, with its nonlinear approximation capability, can effectively capture the complex relationship between energy supply and demand, so that the trained radial basis function neural network model can accurately express the relationship within the energy system and give accurate prediction results, thereby maximizing the utilization efficiency of the energy system, that is, making full use of the utilization efficiency of the energy system.

[0158] By inputting historical operating data of the energy system from N time points prior to the current moment into a trained radial basis function (RBF) neural network model, the model can output target prediction data. When the energy system is controlled to operate according to the target prediction data, the trained RBF neural network model can accurately represent the relationships within the energy system and provide accurate prediction results.

[0159] In the process of controlling the energy system, we can improve the flexibility and efficiency of the energy system, reduce energy costs, and reduce environmental pollution.

[0160] In one embodiment, such as Figure 7 As shown, the present invention provides a training apparatus 700 for a radial basis function neural network model, including a first processor 702 and a first memory 704. The first memory 704 stores programs or instructions that can be run on the first processor 702. When the programs or instructions are executed by the first processor 702, they implement the steps of any of the above-described radial basis function neural network model training methods.

[0161] In one embodiment, such as Figure 8 As shown, the present invention provides a control device 800 for an energy system, including a second processor 802 and a second memory 804. The second memory 804 stores programs or instructions that can be run on the second processor 802. When the programs or instructions are executed by the second processor 802, they implement the steps of the energy system control method described above.

[0162] The first memory 704 and the second memory 804 can be used to store software programs and various data. The first memory 704 and the second memory 804 mainly include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area can store the operating system, application programs or instructions required for at least one function (such as sound playback function, image playback function, etc.). Furthermore, the first memory 704 and the second memory 804 can include volatile memory or non-volatile memory, or both. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory in the embodiments of this application includes, but is not limited to, these and any other suitable types of memory.

[0163] In one embodiment, the present invention provides a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of any of the methods described above.

[0164] In one embodiment, the present invention provides a controller comprising: a training device for any of the radial basis function neural network models described above; and / or a control device for any of the energy systems described above; and / or a readable storage medium as described above.

[0165] In one embodiment, the present invention provides an energy system including a controller as described above.

[0166] The terms "first" and "second" in the specification and claims of this application may explicitly or implicitly include one or more of the features. In the textual description of this invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0167] In the textual description of this invention, it is understood that, unless explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0168] In the claims, description, and accompanying drawings of this invention, the terms "one embodiment," "some embodiments," "specific embodiment," etc., refer to a specific feature, structure, material, or characteristic described in connection with that embodiment or example, which is included in at least one embodiment or example of the invention. In the claims, description, and accompanying drawings of this invention, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0169] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A training method for a radial basis function neural network model, characterized in that, The radial basis function neural network model comprises an input layer, a hidden layer, and an output layer connected in sequence, and the training method includes: When training the radial basis function neural network model with the first training sample, a first threshold and a first weight value are determined. The first threshold is a threshold used to determine the end of training of the radial basis function neural network model, and the first weight value is the weight value between the hidden layer and the output layer. When the radial basis function neural network model is running with the first weight value and the first threshold, the radial basis function neural network model is trained with the second training sample to update the first weight value and the first threshold; The first training sample includes historical running data corresponding to the previous time to the N+1 times before the current time, and the second training sample includes historical running data corresponding to the N times before the current time.

2. The training method for the radial basis function neural network model according to claim 1, characterized in that, The training method also includes: Obtain the predicted value of the radial basis neural network model when the second training sample is used as input; Obtain the actual value corresponding to the second training sample; Determine the root mean square of the predicted value and the actual value; If the root mean square is less than or equal to the first threshold, the training of the radial basis function neural network model is determined to be complete.

3. The training method for the radial basis function neural network model according to claim 2, characterized in that, The hidden layer includes at least one hidden node, and obtaining the predicted value of the radial basis function neural network model when the second training sample is used as input specifically includes: Obtain the output value of the hidden node when the second training sample is used as input; Determine the product of the output value and the first weight value; The predicted value of the radial basis function neural network model is determined based on the product value.

4. The training method for the radial basis function neural network model according to claim 3, characterized in that, The output value of the hidden node is calculated using the following formula: Where x is the second training sample, m is the number of nodes in the hidden layer, and u i (x) is the output value of the i-th hidden node, c i σ is the center vector when the action function of the i-th hidden node is selected as a Gaussian function. i is the normalization constant of the i-th hidden node, and T is the transpose.

5. The training method for the radial basis function neural network model according to claim 3, characterized in that, The output layer comprises p output units, and the predicted values ​​of the radial basis function neural network model are calculated using the following formula: Among them, y j (n) is the predicted value output by the j-th output unit, w i,j It is the weight value corresponding to the i-th hidden node to the j-th output unit in the first weight value, z j It is the bias compensation for the j-th output unit, u i (x) is the output value of the i-th hidden node, x is the second training sample, and m is the number of nodes in the hidden layer.

6. The training method for the radial basis function neural network model according to claim 3, characterized in that, The historical operational data includes at least one of the following: Energy storage battery charging power, energy storage battery discharging power, grid power purchase and sale, charging pile working mode, heat pump working mode, household energy storage integrated machine working mode, charging pile dynamic charging power, heat pump water temperature, and heat pump air temperature.

7. A control method for an energy system, characterized in that, include: Obtain the historical operating data of the energy system for the N moments prior to the current moment; The historical running data is input into the radial basis neural network model trained by the training method of the radial basis neural network model as described in any one of claims 1 to 6 to obtain target prediction data; Control the energy system to operate according to the target predicted data.

8. A training device for a radial basis function neural network model, characterized in that, The radial basis function neural network model includes an input layer, a hidden layer, and an output layer connected in sequence, and the training device includes: A determining unit is configured to determine a first threshold and a first weight value when training the radial basis function neural network model with a first training sample. The first threshold is a threshold used to determine the end of training of the radial basis function neural network model, and the first weight value is a weight value between the hidden layer and the output layer. An update unit is configured to train the radial basis neural network model with a second training sample when the radial basis neural network model is running with the first weight value and the first threshold, so as to update the first weight value and the first threshold. The first training sample includes historical running data corresponding to the previous time to the N+1 times before the current time, and the second training sample includes historical running data corresponding to the N times before the current time.

9. A control device for an energy system, characterized in that, include: The acquisition unit is used to acquire the historical operating data of the energy system for the N times preceding the current time. The prediction unit is used to input the historical running data into the radial basis neural network model trained by the training method of the radial basis neural network model as described in any one of claims 1 to 6, so as to obtain the target prediction data; A control unit for controlling the energy system to operate according to the target predicted data.

10. A training device for a radial basis function neural network model, characterized in that, It includes a first processor and a first memory, the first memory storing a program or instructions that can run on the first processor, the program or instructions being executed by the first processor to implement the steps of the method as described in any one of claims 1 to 6.

11. A control device for an energy system, characterized in that, It includes a second processor and a second memory, the second memory storing a program or instructions that can run on the second processor, the program or instructions being executed by the second processor to implement the steps of the method as described in claim 7.

12. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 7.

13. A controller, characterized in that, include: Training apparatus for radial basis function neural network model as described in claim 8 or 10; and / or The control device for the energy system as described in claim 9 or 11; and / or The readable storage medium as described in claim 12.

14. An energy system, characterized in that, include: The controller as described in claim 13.

Images