Intelligent operation control system and method based on machine learning to drive new energy storage
Through the intelligent operation control method of new energy storage driven by machine learning, we predict changes in new energy production capacity and dynamically adjust energy storage equipment, solving the problem of smooth operation of new energy energy storage systems in the face of unstable production capacity and dynamic electricity demands. , improving the utilization rate of new energy and system stability.
Patent Information
- Application Number
- CN202411207146.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-30
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-08-30
AI Technical Summary
When facing unstable new energy production capacity and dynamic electricity demand, it is difficult for new energy storage systems to achieve smooth operation, and with the optimization of new energy resource allocation, the requirements for intelligent control are further improved.
Using intelligent operation control methods based on machine learning, we can collect weather data and new energy production area information, build a new energy production capacity estimate model and energy storage control strategy adjustment model, use RNN or CNN models to predict production capacity changes, and use reinforcement learning The model dynamically adjusts the charging and discharging strategies of energy storage equipment, and considers human influencing factors to optimize capacity prediction and control.
It realizes more accurate prediction of changes in new energy production capacity and intelligent scheduling of energy storage systems, effectively balances power demand, improves new energy utilization rate and system stability, and has scalability and flexibility.
Smart Images

Figure CN118713188B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of new energy storage technology, and more specifically, to a control system and method for intelligent operation of new energy storage driven by machine learning. Background Art
[0002] As an important infrastructure for the widespread application of renewable energy, the new energy storage system's main task is to effectively absorb, store and dispatch unstable new energy electricity, thereby providing reliable guarantee for the stable power supply of the power system.
[0003] In fact, since the instability of electricity produced by new energy resources such as solar energy and wind energy comes from the complex and changeable natural conditions such as weather, its electricity production will fluctuate greatly every hour, day, and even year. This poses a very high technical challenge for the new energy storage system to steadily absorb and store new energy electricity while meeting the ever-changing electricity demand.
[0004] The Chinese patent with the existing authorization announcement number CN114389272B discloses a multi-mode coordinated control method for wind, solar and energy storage new energy power stations, including: the wind, solar and energy storage new energy power station control system collects the operating data of the wind power system, photovoltaic system, and energy storage system in real time, and reads the limited parameters of the wind power system, photovoltaic system, and energy storage system; the wind, solar and energy storage new energy power station control system reads the wind, solar and energy storage scene operation mode flag issued by the dispatching center to determine the way in which the wind, solar and energy storage system participates in power control; in the wind and solar mode, select proportional control, margin control, wind power priority or photovoltaic priority mode to complete the power distribution between the wind power system and the photovoltaic system; in the case of energy storage system participating in the control, the command tracking, power smoothing, wind and solar consumption, and peak power supply control modes are integrated. This patent realizes the active collaborative optimization control of the wind, solar and energy storage station group, smoothes the voltage and frequency fluctuations of the power grid caused by the access of new energy, and improves the stability of new energy grid connection.
[0005] On the one hand, the energy storage system needs to adjust the charging and discharging strategies of the energy storage equipment in a timely manner according to the ever-changing production capacity of new energy, to minimize the fluctuation of new energy production capacity and achieve the highest possible utilization rate of new energy. On the other hand, it also needs to allocate and dispatch the stored energy in real time according to the dynamic power load on the user side to ensure the continuous and reliable supply of electricity to end users.
[0006] As the scale of my country's new energy development continues to expand, the configuration of various new energy sources, mainly solar and wind energy, in different regions is also being optimized, which also increases the uncertainty of the overall capacity structure and spatial distribution of new energy. This puts forward higher-level control intelligence requirements for new energy storage systems. Summary of the invention
[0007] In order to solve the above technical difficulties, the present invention provides a method for intelligent operation control of new energy storage driven by machine learning, the method comprising:
[0008] Collect weather data, structural coefficients of new energy production areas, and estimated production capacity and electricity demand of new energy production areas;
[0009] Input weather data, new energy production area structural coefficients and new energy production area estimated production capacity into a pre-built new energy production capacity estimation model to output a new energy production capacity change time series, train the new energy production capacity estimation model based on each set of weather data, new energy production area structural coefficients and new energy production area estimated production capacity, and the corresponding new energy production capacity change time series, so that the new energy production capacity estimation model outputs a new energy production capacity change time series, wherein the new energy production capacity estimation model is an RNN model or a CNN model;
[0010] Inputting the time series of new energy capacity changes and the electric energy demand into the energy storage control strategy adjustment model to output the operation strategy, and based on the time series of new energy capacity changes and the electric energy demand, and the corresponding operation strategy, performing structural definition, structural and training design, training and environmental feedback on the energy storage control strategy adjustment model, so that the energy storage control strategy adjustment model outputs the operation strategy, wherein the energy storage control strategy adjustment model is a reinforcement learning model;
[0011] Control the charging and discharging operations of new energy storage equipment according to the operation strategy.
[0012] Furthermore, the weather data includes light intensity, water level of a hydropower station and wind speed; the new energy production area structure coefficient includes 1 and 0; and the method for constructing the new energy production capacity estimation model includes:
[0013] Each set of weather data, new energy production area structure coefficient and new energy production area expected production capacity is converted into the form of the first eigenvector, the elements of which include light intensity, hydropower station water level, wind speed, new energy production area structure coefficient and new energy production area expected production capacity;
[0014] The set of all first eigenvectors is used as the input of a new energy capacity estimation model, the new energy capacity estimation model uses the new energy capacity change time series predicted for each group of first eigenvectors as the output, uses the actual new energy capacity change time series corresponding to each group of first eigenvectors as the prediction target, and uses minimizing the sum of the first prediction accuracies of all predicted new energy capacity change time series as the training target;
[0015] Among them, the calculation formula for the first prediction accuracy is: ,in, is the first prediction accuracy, For the The predicted new energy capacity change time series corresponding to the first eigenvector of the group, For the The actual new energy production capacity change time series corresponding to the first eigenvector of the group; Represents the group number of the first eigenvector; the new energy capacity estimation model is trained until the sum of the first prediction accuracies reaches convergence and the training is stopped; the new energy capacity estimation model is an RNN model or a CNN model.
[0016] Furthermore, the structure definition of the energy storage control strategy adjustment model includes:
[0017] Define the state space:
[0018] State S is the new energy production capacity and power demand; the state vector dimension is set to 2;
[0019] Define the action space:
[0020] Action A is the energy storage charging power and discharging power at the next time point, A={charging A1, discharging A2}; A1 is a floating point number between 0 and 1, representing the charging ratio, and A2 is a floating point number between 0 and 1, representing the discharging ratio.
[0021] Furthermore, the structure and training design of the energy storage control strategy adjustment model include:
[0022] Design Actor network structure:
[0023] The Actor network structure includes 1 first input layer, 2 hidden layers and 1 first output layer; the first input layer has 2 state vectors; the 2 hidden layers include the first hidden layer and the second hidden layer, the first hidden layer includes 32 nodes, the activation function is Relu, the second hidden layer includes 16 nodes, the activation function is Relu; the number of nodes in the first output layer is 2, with 3-dimensional continuous action values and no activation function;
[0024] Design Critic network structure:
[0025] The critic network structure includes 1 second input layer, 2 hidden layers and 1 second output layer; the state vector of the second input layer is 4; the 2 hidden layers include the third hidden layer and the fourth hidden layer, the third hidden layer includes 64 nodes, the activation function is Relu, the fourth hidden layer includes 32 nodes, the activation function is Relu; the second output layer includes 1 node, without activation function;
[0026] Design training configuration:
[0027] Actor network learning rate α_actor=0.0001; Critic network learning rate α_critic=0.001;
[0028] Experience replay memory size D = 10000; batch size B = 64; update network step size τ = 0.001; discount factor =0.99.
[0029] Furthermore, the environmental feedback of the energy storage control strategy adjustment model includes:
[0030] Select a random action A and get feedback from the environment ( , , done), Represents the reward value obtained after selecting an action; reward value The calculation methods include:
[0031] ;
[0032] ;
[0033] ;
[0034] In the formula, Indicates reward for satisfying needs; represents the power fluctuation penalty; is the power system load, representing the power demand; is the amount of electricity released or absorbed, i.e., the action variable; is the state vector at the previous time point; is the state vector at the current time point; To meet the weight coefficient of the required reward item, is the weight coefficient of the power fluctuation penalty term; the total load capacity represents the maximum range of the power system load;
[0035] done is a flag indicating whether it is a terminal state. If the task is completed, done = True; if the task fails, done = True; if it is a non-terminal state and the task is still continuing, done = False; is the next state; the action A and the corresponding environmental feedback after each training are saved to the experience replay memory.
[0036] Furthermore, the training method of the energy storage control strategy adjustment model includes:
[0037] The action A and the corresponding environment feedback in the experience replay memory are divided into a training set and a validation set;
[0038] Calculate the current Critic network evaluation value: ;
[0039] In the formula, Indicates the use of the current Critic network;
[0040] Calculate the Critic network target value :
[0041] ;
[0042] In the formula, is the Critic network's response to the next state the quality value resulting from the evaluation;
[0043] Calculating Actor Network Gradients :
[0044] ;
[0045] In the formula, is based on The action produced; It is a Jacobian matrix. Each element of the Jacobian matrix is learned by the Critic network. Actor Network Parameters Partial derivatives in each dimension; Indicates the action's parameters for the Actor network The gradient of
[0046] Calculate the loss value :
[0047] ;
[0048] Continue to train the model using the training set until If it continues to decrease and converge, the training ends.
[0049] Furthermore, it also includes:
[0050] Collect data on air pollutant concentrations;
[0051] The air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series are input into the pre-built human influencing factor prediction model to output the final new energy capacity change time series.
[0052] Furthermore, the method for constructing the human influence factor prediction model includes:
[0053] Pre-collect a group of air pollutant concentration data, new energy production area structural coefficient, new energy production capacity change time series and the final new energy production capacity change time series corresponding to the new energy production capacity change time series, and convert the air pollutant concentration data, new energy production area structural coefficient, new energy production capacity change time series and the final new energy production capacity change time series into a corresponding set of second eigenvectors;
[0054] Each group of second eigenvectors is used as the input of the prediction model of human influence factors. The prediction model of human influence factors uses a group of final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as output, and uses the actual final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the prediction target; and uses minimizing the sum of the prediction errors of all air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the training target; wherein, the calculation formula of the prediction error is: ,in is the prediction error, is the final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, is the actual final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, and h is the second eigenvector group number; the human influence factor prediction model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the human influence factor prediction model is an RNN model.
[0055] Based on machine learning to drive the intelligent operation and control system of new energy storage, the system includes:
[0056] Collection module: used to collect air pollutant concentration data, weather data, new energy production area structure coefficient, new energy production area expected production capacity and electricity demand;
[0057] The first capacity estimation model: used to input weather data, new energy production area structure coefficients and new energy production area estimated capacity into the pre-built new energy capacity estimation model to output the new energy capacity change time series;
[0058] The second capacity estimation model: used to input the air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series into the pre-built human influence factor prediction model, and output the final new energy capacity change time series;
[0059] Operation strategy output module: used to input the final new energy capacity change time series and power demand into the energy storage control strategy adjustment model to output the operation strategy;
[0060] Control module: used to control the charging and discharging operations of new energy storage equipment according to the operation strategy.
[0061] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the intelligent operation control method of new energy storage driven by machine learning is implemented.
[0062] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed, the intelligent operation control method of new energy storage driven by machine learning is implemented.
[0063] The technical effects and advantages of the intelligent operation control system and method of new energy storage driven by machine learning provided by the present invention are as follows:
[0064] The present invention collects detailed weather data and new energy production area structure information, and uses a machine learning model, i.e., a new energy capacity estimation model, to more accurately predict the capacity change trend of new energy (such as photovoltaic, wind energy, and hydropower) under different conditions, and provide a basis for the operation and control of the energy storage system. The present invention recognizes that human activities will also affect the new energy capacity. By collecting environmental data and establishing a human impact prediction model, the degree of human impact can be quantitatively evaluated, and the accuracy of capacity prediction can be further optimized. Based on accurate capacity prediction and power load information, the present invention uses a reinforcement learning energy storage control strategy model to perceive the environmental state in real time, dynamically adjust the charging and discharging actions of the energy storage equipment, effectively meet the power demand and control power fluctuations. Through intelligent predictive control, the present invention can balance the instability of new energy to a certain extent, meet the power demand to the maximum extent, and thus improve the utilization rate of new energy. The algorithm design based on machine learning of the present invention has strong scalability, and the model can be continuously learned and updated to adapt to the changing characteristics of the new energy structure. Compared with traditional methods, the present invention is more intelligent and rigid, can effectively control the system operation risk, and enable the energy storage system to operate stably and reliably for a long time. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 This is a flow chart of the intelligent operation control method of new energy storage driven by machine learning in Example 1;
[0066] Figure 2 This is a flow chart of the intelligent operation control method of new energy storage driven by machine learning in Example 2;
[0067] Figure 3This is a connection diagram of the intelligent operation control system of new energy storage driven by machine learning of the present invention;
[0068] Figure 4 is a schematic diagram of an electronic device of the present invention;
[0069] Figure 5 Schematic diagram of the storage medium of the present invention. DETAILED DESCRIPTION
[0070] 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.
[0071] Embodiment 1:
[0072] See also Figure 1 As shown, the intelligent operation control method of new energy storage driven by machine learning in this embodiment includes:
[0073] Collect weather data, structural coefficients of new energy production areas, and estimated production capacity and electricity demand of new energy production areas;
[0074] Input weather data, new energy production area structural coefficients and new energy production area estimated production capacity into a pre-built new energy production capacity estimation model to output a new energy production capacity change time series, train the new energy production capacity estimation model based on each set of weather data, new energy production area structural coefficients and new energy production area estimated production capacity, and the corresponding new energy production capacity change time series, so that the new energy production capacity estimation model outputs a new energy production capacity change time series, wherein the new energy production capacity estimation model is an RNN model or a CNN model;
[0075] Inputting the time series of new energy capacity changes and the electric energy demand into the energy storage control strategy adjustment model to output the operation strategy, and based on the time series of new energy capacity changes and the electric energy demand, and the corresponding operation strategy, performing structural definition, structural and training design, training and environmental feedback on the energy storage control strategy adjustment model, so that the energy storage control strategy adjustment model outputs the operation strategy, wherein the energy storage control strategy adjustment model is a reinforcement learning model;
[0076] Control the charging and discharging operations of new energy storage equipment according to the operation strategy.
[0077] The weather data include light intensity, water level of the hydropower station and wind speed; the light intensity and wind speed are measured by the local meteorological observation station, and the water level of the hydropower station can be obtained from the official data published by the hydropower station; under the same conditions, the higher the light intensity, the greater the solar energy production capacity will be, for example, the solar energy production capacity on a sunny day will be higher than that on a cloudy day; when the water level of the hydropower station is low, the effective water storage capacity of the reservoir is small, and the flood peak regulation and water supply tasks undertaken by the reservoir will be greatly reduced, thereby affecting the production capacity of the hydropower station, while when the water level of the hydropower station is high, the reservoir stores more water, which can better meet the power supply demand, thereby improving the production capacity of the hydropower station, and the water level of the hydropower station also affects the production efficiency of the hydropower station. When the water level of the hydropower station is low, the water flow rate will slow down, thereby affecting the rotation speed and kinetic energy of the water, resulting in a decrease in the power generation efficiency of the hydropower station, and when the water level of the hydropower station is high, the water flow rate will increase, thereby driving the water wheel rotation speed to increase, so that the power generation efficiency of the hydropower station is improved; under the same conditions, the greater the wind speed, the greater the wind energy production capacity will be.
[0078] The new energy production area structure coefficient includes 1 and 0. The present invention mainly considers new energy varieties whose production capacity will be affected by the weather and produce volatility. New energy varieties with small volatility and basic stability are not considered, such as nuclear energy and geothermal energy. However, since the structure of new energy production areas will change with the change of geographical location, for example, Gobi and desert are very suitable as the production areas of solar energy and wind energy, but not suitable as the production areas of hydropower stations, the new energy production area structure coefficient is (1, 1, 0) at this time. When the new energy production area structure coefficient is multiplied by the corresponding weather data, the water level of the hydropower station will be multiplied by 0 and excluded. The new energy production area structure coefficient can help the new energy capacity estimation model eliminate certain interference factors and reduce the occurrence of overfitting.
[0079] The estimated production capacity of the new energy production area is obtained from the official website of the company to which the new energy production area belongs, the official website of the local government and the official website of the Energy Bureau. Because most of the new energy production capacities are not stable enough so far, downstream industries need to carry out production and construction in advance based on the new energy production capacity. At the same time, the new energy production capacity equipment has been under construction or replacement. Therefore, generally speaking, the new energy production area will issue a predetermined production capacity plan to prepare for the construction in the region. However, at the same time, because the predetermined plan considers the prediction results under the same conditions, the detailed impact of the weather is not considered, so it is necessary to add weather data as a consideration factor; the reason why the present invention does not use real-time production capacity data is that when the power grid starts to transmit power, the real-time production capacity data as statistical data is accurate, but it is no longer timely for the new energy storage system. It is meaningless to balance the volatility of the electricity that has been transmitted, and there is no room for regulation.
[0080] The new energy capacity change time series is a vector set of new energy capacities at time points. Weather data and the estimated capacity of new energy production areas have time attributes when they are input, and the specific time span is determined by the input end. For example, when inputting, the weather data is hourly weather data for 24 hours, and the estimated capacity of the new energy production area is the estimated capacity for one day. Then, there will be 24 corresponding estimated data of new energy capacities at the time of output, and the composed vector is the new energy capacity change time series.
[0081] The electric energy demand is obtained from the power grid dispatching center, which affects the reward value for meeting the demand and thus affects the operation strategy.
[0082] The method for constructing the new energy capacity estimation model includes:
[0083] Each set of weather data, new energy production area structure coefficient and new energy production area expected production capacity is converted into the form of the first eigenvector, the elements of which include light intensity, hydropower station water level, wind speed, new energy production area structure coefficient and new energy production area expected production capacity;
[0084] The set of all first eigenvectors is used as the input of a new energy capacity estimation model, the new energy capacity estimation model uses the new energy capacity change time series predicted for each group of first eigenvectors as the output, uses the actual new energy capacity change time series corresponding to each group of first eigenvectors as the prediction target, and uses minimizing the sum of the first prediction accuracies of all predicted new energy capacity change time series as the training target;
[0085] Among them, the calculation formula for the first prediction accuracy is: ,in, is the first prediction accuracy, For the The predicted new energy capacity change time series corresponding to the first eigenvector of the group, For the The actual new energy production capacity change time series corresponding to the first eigenvector of the group; Represents the group number of the first eigenvector; the new energy capacity estimation model is trained until the sum of the first prediction accuracies reaches convergence and the training is stopped; the new energy capacity estimation model is any one of an RNN model or a CNN model.
[0086] The structural definition of the energy storage control strategy adjustment model includes:
[0087] Define the state space:
[0088] State S is the new energy production capacity and electricity demand; the state vector dimension is set to 2.
[0089] Define the action space:
[0090] Action A is the energy storage charging power and discharging power at the next time point, A={charging A1, discharging A2}; A1 is a floating point number between 0 and 1, representing the charging ratio, and A2 is a floating point number between 0 and 1, representing the discharging ratio.
[0091] The structure and training design of the energy storage control strategy adjustment model include:
[0092] Design Actor network structure:
[0093] The Actor network structure includes 1 first input layer, 2 hidden layers and 1 first output layer; the state vector of the first input layer is 2; the 2 hidden layers include the first hidden layer and the second hidden layer, the first hidden layer includes 32 nodes, the activation function is Relu, the second hidden layer includes 16 nodes, the activation function is Relu; the number of nodes in the first output layer is 2, with 3-dimensional continuous action values and no activation function.
[0094] Design Critic network structure:
[0095] The critic network structure includes 1 second input layer, 2 hidden layers and 1 second output layer; the state vector of the second input layer is 4; the 2 hidden layers include the third hidden layer and the fourth hidden layer, the third hidden layer includes 64 nodes, the activation function is Relu, the fourth hidden layer includes 32 nodes, the activation function is Relu; the second output layer includes 1 node and no activation function.
[0096] Design training configuration:
[0097] Actor network learning rate α_actor = 0.0001;
[0098] Critic network learning rate α_critic=0.001;
[0099] Experience replay memory size D = 10000;
[0100] Batch size B = 64;
[0101] Update network step size τ = 0.001;
[0102] Discount Factor =0.99.
[0103] The environmental feedback of the energy storage control strategy adjustment model includes:
[0104] Select a random action A and get feedback from the environment ( , , done), Represents the reward value obtained after selecting an action; reward value The calculation methods include:
[0105] ;
[0106] ;
[0107] ;
[0108] In the formula, Indicates reward for satisfying needs; represents the power fluctuation penalty; is the power system load, representing the power demand; is the amount of electricity released or absorbed, i.e., the action variable; is the state vector at the previous time point; is the state vector at the current time point; To meet the weight coefficient of the required reward item, is the weight coefficient of the power fluctuation penalty term; the total load capacity represents the maximum range of the power system load;
[0109] done is a flag indicating whether it is a terminal state. If the task is completed, done = True; if the task fails, done = True; if it is a non-terminal state and the task is still continuing, done = False; is the next state; the action A and the corresponding environmental feedback after each training are saved to the experience replay memory.
[0110] The training method of the energy storage control strategy adjustment model includes:
[0111] The action A and the corresponding environment feedback in the experience replay memory are divided into a training set and a validation set;
[0112] Calculate the current Critic network evaluation value: ;
[0113] In the formula, Indicates the use of the current Critic network;
[0114] Calculate the Critic network target value :
[0115] ;
[0116] In the formula, is the Critic network's response to the next state the quality value resulting from the evaluation;
[0117] Calculating Actor Network Gradients :
[0118] ;
[0119] In the formula, is based on The action produced; It is a Jacobian matrix. Each element of the Jacobian matrix is learned by the Critic network. Actor Network Parameters Partial derivatives in each dimension; Indicates the action's parameters for the Actor network The gradient of
[0120] Calculate the loss value :
[0121] ;
[0122] Continue to train the model using the training set until If it continues to decrease and converge, the training ends.
[0123] The new energy storage device is controlled to perform charging and discharging operations according to the operation strategy. For example, the operation strategy is {(1:00, discharge 0.2), (3:00, charge 0.1)}, then the new energy storage device discharges on time at 1:00 until the power reserve is reduced by 20% of the total power, and charges on time at 3:00 until the power reserve is increased by 10% of the total power.
[0124] Embodiment 2:
[0125] like Figure 2 As shown, this embodiment further improves the design on the basis of embodiment 1. The difference is that in the actual application process, the production capacity of the new energy production area is affected by human factors in addition to natural factors such as weather. If there are large-scale construction projects near the new energy production area, photovoltaic lighting may be interfered with, thereby increasing load uncertainty; if there are chemical companies around the new energy production area, air pollutants have a significant impact on the output of photovoltaic panels, and in view of the proportion of solar energy in new energy, this may cause production capacity fluctuations and affect the stability of the operation control of the new energy storage system; and the start time of large-scale construction projects and chemical companies is generally during the day, which will occupy a large amount of electricity and affect the new energy production area, which affects the stable operation control of the new energy storage system; therefore, the intelligent operation control method of new energy storage driven by machine learning provided in this embodiment also includes:
[0126] Collect data on air pollutant concentrations;
[0127] The air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series are input into the pre-built human influencing factor prediction model to output the final new energy capacity change time series.
[0128] The air pollutant concentration data can be obtained from the monitoring report of the detection station, such as the PM2.5 concentration report. The purpose of inputting the air pollutant concentration data and the new energy production area structure coefficient together is to multiply the air pollutant concentration data with the new energy production area structure coefficient. When the value corresponding to solar energy in the new energy production area structure coefficient is 0, the air pollutant concentration data will be multiplied by zero and thus filtered. At this time, the final new energy production capacity change time series is the new energy production capacity change time series, which has no effect on subsequent calculations; it helps the prediction model of human influencing factors to eliminate certain interference factors and reduce the occurrence of overfitting.
[0129] The method for constructing the human influence factor prediction model includes:
[0130] Pre-collect a group of air pollutant concentration data, new energy production area structural coefficient, new energy production capacity change time series and the final new energy capacity change time series corresponding to the new energy capacity change time series, and convert the air pollutant concentration data, new energy production area structural coefficient, new energy capacity change time series and the final new energy capacity change time series into a corresponding set of second eigenvectors.
[0131] Each group of second eigenvectors is used as the input of the prediction model of human influence factors. The prediction model of human influence factors uses a group of final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as output, and uses the actual final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the prediction target; and uses minimizing the sum of the prediction errors of all air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the training target; wherein, the calculation formula of the prediction error is: ,in is the prediction error, is the final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, is the actual final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, and h is the second eigenvector group number; the human influence factor prediction model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the human influence factor prediction model is a CNN model.
[0132] Embodiment 3:
[0133] like Figure 3 As shown, based on machine learning, the intelligent operation and control system of new energy storage is driven by the system, which includes:
[0134] Collection module: used to collect air pollutant concentration data, weather data, new energy production area structure coefficient, new energy production area expected production capacity and electricity demand;
[0135] The first capacity estimation model: used to input weather data, new energy production area structure coefficients and new energy production area estimated capacity into the pre-built new energy capacity estimation model to output the new energy capacity change time series;
[0136] The second capacity estimation model: used to input the air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series into the pre-built human influence factor prediction model, and output the final new energy capacity change time series;
[0137] Operation strategy output module: used to input the final new energy capacity change time series and power demand into the energy storage control strategy adjustment model to output the operation strategy;
[0138] Control module: used to control the charging and discharging operations of new energy storage equipment according to the operation strategy.
[0139] Embodiment 4:
[0140] See also Figure 4 As shown, according to another aspect of the present application, an electronic device 500 is also provided. The electronic device 500 may include one or more processors and one or more memories. The memory stores a computer readable code, and when the computer readable code is executed by one or more processors, the intelligent operation control method of new energy storage driven by machine learning as described above can be executed.
[0141] The system according to the embodiment of the present application can also be used by Figure 4 The electronic device architecture shown in FIG. Figure 4 As shown, the electronic device 500 may include a bus 501, one or more CPUs 502, ROM 503, RAM 504, a communication port 505 connected to a network, an input / output 506, a hard disk 507, etc. The storage device in the electronic device 500, such as ROM 503 or hard disk 507, may store the intelligent operation control method of new energy storage driven by machine learning provided in this application. Furthermore, the electronic device 500 may also include a user interface 508. Of course, Figure 4 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 4 One or more components of an electronic device are shown.
[0142] Embodiment 5:
[0143] See also Figure 5 As shown, it is a computer-readable storage medium according to an embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium. When the computer-readable instructions are executed by the processor, the intelligent operation control method based on machine learning driven new energy storage according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc.
[0144] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
[0145] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for intelligent operation and control of new energy storage driven by machine learning, characterized in that: Methods include: Collect weather data, structural coefficients of new energy production areas, and estimated production capacity and electricity demand of new energy production areas; Input weather data, new energy production area structural coefficients and new energy production area estimated production capacity into a pre-built new energy production capacity estimation model to output a new energy production capacity change time series, train the new energy production capacity estimation model based on each set of weather data, new energy production area structural coefficients and new energy production area estimated production capacity, and the corresponding new energy production capacity change time series, so that the new energy production capacity estimation model outputs a new energy production capacity change time series, wherein the new energy production capacity estimation model is an RNN model or a CNN model; Inputting the time series of new energy capacity changes and the electric energy demand into the energy storage control strategy adjustment model to output the operation strategy, and based on the time series of new energy capacity changes and the electric energy demand, and the corresponding operation strategy, performing structural definition, structural and training design, training and environmental feedback on the energy storage control strategy adjustment model, so that the energy storage control strategy adjustment model outputs the operation strategy, wherein the energy storage control strategy adjustment model is a reinforcement learning model; Control the new energy storage equipment to perform charging and discharging operations according to the operation strategy; The structural definition of the energy storage control strategy adjustment model includes: Define the state space: State S is the new energy production capacity and power demand; the state vector dimension is set to 2; Define the action space: Action A is the energy storage charging power and discharging power at the next time point, A={charging A1, discharging A2}; A1 is a floating point number between 0 and 1, representing the charging ratio, and A2 is a floating point number between 0 and 1, representing the discharging ratio; The structure and training design of the energy storage control strategy adjustment model include: Design Actor network structure: The Actor network structure includes 1 first input layer, 2 hidden layers and 1 first output layer; the first input layer has 2 state vectors; the 2 hidden layers include the first hidden layer and the second hidden layer, the first hidden layer includes 32 nodes, the activation function is Relu, the second hidden layer includes 16 nodes, the activation function is Relu; the number of nodes in the first output layer is 2, with 3-dimensional continuous action values and no activation function; Design Critic network structure: The critic network structure includes 1 second input layer, 2 hidden layers and 1 second output layer; the state vector of the second input layer is 4; the 2 hidden layers include the third hidden layer and the fourth hidden layer, the third hidden layer includes 64 nodes, the activation function is Relu, the fourth hidden layer includes 32 nodes, the activation function is Relu; the second output layer includes 1 node, without activation function; Design training configuration: Actor network learning rate α_actor=0.0001; Critic network learning rate α_critic=0.001; Experience replay memory size D = 10000; batch size B = 64; update network step size τ = 0.001; discount factor =0.99; The environmental feedback of the energy storage control strategy adjustment model includes: Select a random action A and get feedback from the environment ( , , done), Represents the reward value obtained after selecting an action; reward value The calculation methods include: ; ; ; In the formula, Indicates reward for satisfying needs; represents the power fluctuation penalty; is the power system load, representing the power demand; is the amount of electricity released or absorbed, i.e., the action variable; is the state vector at the previous time point; is the state vector at the current time point; To meet the weight coefficient of the required reward item, is the weight coefficient of the power fluctuation penalty term; the total load capacity represents the maximum range of the power system load; done is a flag indicating whether it is a terminal state. If the task is completed, done = True; if the task fails, done = True; if it is a non-terminal state and the task is still continuing, done = False; is the next state; the action A and the corresponding environmental feedback after each training are saved to the experience replay memory.
2. The intelligent operation control method of new energy storage based on machine learning according to claim 1 is characterized in that: The weather data includes light intensity, water level of hydropower station and wind speed; The new energy production area structure coefficient includes 1 and 0; the construction method of the new energy production capacity estimation model includes: Each set of weather data, new energy production area structure coefficient and new energy production area expected production capacity is converted into the form of the first eigenvector, the elements of which include light intensity, hydropower station water level, wind speed, new energy production area structure coefficient and new energy production area expected production capacity; The set of all first eigenvectors is used as the input of a new energy capacity estimation model; the new energy capacity estimation model takes the new energy capacity change time series predicted for each group of first eigenvectors as output, takes the actual new energy capacity change time series corresponding to each group of first eigenvectors as a prediction target, and takes minimizing the sum of the first prediction accuracies of all predicted new energy capacity change time series as a training target; the new energy capacity estimation model is trained until the sum of the first prediction accuracies reaches convergence, and the training is stopped; the new energy capacity estimation model is an RNN model or a CNN model.
3. The intelligent operation control method of new energy storage driven by machine learning according to claim 1 is characterized in that: The training method of the energy storage control strategy adjustment model includes: The action A and the corresponding environment feedback in the experience replay memory are divided into a training set and a validation set; Calculate the current Critic network evaluation value: ; In the formula, Indicates the use of the current Critic network; Calculate the Critic network target value : ; In the formula, is the Critic network's response to the next state the quality value resulting from the evaluation; Calculating Actor Network Gradients : ; In the formula, is based on The action produced; It is a Jacobian matrix. Each element of the Jacobian matrix is learned by the Critic network. Actor Network Parameters Partial derivatives in each dimension; Indicates the action's parameters for the Actor network The gradient of Calculate the loss value : ; Continue to train the model using the training set until If it continues to decrease and converge, the training ends.
4. The intelligent operation control method of new energy storage driven by machine learning according to claim 1 is characterized in that: Also includes: Collect data on air pollutant concentrations; The air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series are input into the pre-built human influencing factor prediction model to output the final new energy capacity change time series.
5. The intelligent operation control method of new energy storage driven by machine learning according to claim 4 is characterized in that: The method for constructing the human influence factor prediction model includes: Pre-collect a group of air pollutant concentration data, new energy production area structural coefficient, new energy production capacity change time series and the final new energy production capacity change time series corresponding to the new energy production capacity change time series, and convert the air pollutant concentration data, new energy production area structural coefficient, new energy production capacity change time series and the final new energy production capacity change time series into a corresponding set of second eigenvectors; Each group of second eigenvectors is used as the input of the prediction model of human influence factors. The prediction model of human influence factors uses a group of final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as output, and uses the actual final new energy capacity change time series corresponding to each group of air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the prediction target; and uses minimizing the sum of the prediction errors of all air pollutant concentration data, new energy production area structural coefficient and new energy capacity change time series as the training target; wherein, the calculation formula of the prediction error is: ,in is the prediction error, is the final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, is the actual final new energy capacity change time series corresponding to the h-th group of air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series, and h is the second eigenvector group number; the human influence factor prediction model is trained until the sum of the prediction errors reaches convergence and the training is stopped; the human influence factor prediction model is an RNN model.
6. Based on machine learning to drive the intelligent operation and control system of new energy storage, it is characterized by: The method for intelligent operation control of new energy storage driven by machine learning according to any one of claims 1 to 5 is implemented, and the system includes: Collection module: used to collect air pollutant concentration data, weather data, new energy production area structure coefficient, new energy production area expected production capacity and electricity demand; The first capacity estimation model: used to input weather data, new energy production area structure coefficients and new energy production area estimated capacity into the pre-built new energy capacity estimation model to output the new energy capacity change time series; The second capacity estimation model: used to input the air pollutant concentration data, the new energy production area structure coefficient and the new energy capacity change time series into the pre-built human influence factor prediction model, and output the final new energy capacity change time series; Operation strategy output module: used to input the final new energy capacity change time series and power demand into the energy storage control strategy adjustment model to output the operation strategy; Control module: used to control the charging and discharging operations of new energy storage equipment according to the operation strategy.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the intelligent operation control method of new energy storage driven by machine learning as described in any one of claims 1 to 5.
8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, it implements the intelligent operation control method of new energy storage driven by machine learning as described in any one of claims 1 to 5.
