Dynamic optimization control system of energy storage cluster based on edge computing

Abstract

The present invention discloses a dynamic optimization control system for energy storage clusters based on edge computing, which relates to the technical field of energy storage clusters, including collecting historical data of energy storage clusters and local data of edge nodes; using an ARIMA model to predict time series, using a BP neural network to identify and correct nonlinear errors, and outputting a final power prediction value. The method of the present invention can respond to power fluctuations more quickly, optimize power distribution, and ensure the stable operation of the energy storage system in the power grid through short-term trend prediction of ARIMA and error correction of BP neural network. By training the local BP neural network model on each edge node, distributed computing resources can be fully utilized. Through the synergistic effect of a double-layer optimization model, the final power distribution scheme and transmission path optimization results are output to ensure that under the fluctuation of load demand and power supply, it can operate efficiently, reduce energy consumption, and maximize the utilization of renewable energy.

Description

Technical Field

[0001] The present invention relates to the technical field of energy storage clusters, and in particular to a dynamic optimization control system of an energy storage cluster based on edge computing. Background Art

[0002] With the widespread application of renewable energy and the continuous development of distributed energy technology, energy storage clusters have become a key component of energy management systems. Energy storage systems can store excess electricity when there is an oversupply of electricity and release electricity when electricity demand peaks, thereby achieving balance and regulation of power loads. Traditional energy storage system management and control methods have low efficiency in energy distribution and are difficult to respond to dynamic load changes in real time. With the introduction of edge computing technology, computing resources are moved to the edge of the network, which improves the real-time and reliability of data processing, allowing energy storage systems to more efficiently manage and dispatch the energy of distributed nodes.

[0003] However, traditional power forecasting methods are sensitive to the linear characteristics of time series and have difficulty coping with nonlinear changes in the operation of energy storage clusters, resulting in inaccurate forecasting results. At the same time, most existing energy scheduling algorithms are based on fixed rules and models and lack a real-time feedback mechanism for the dynamic path loss between energy storage nodes and load nodes, resulting in low power allocation and transmission efficiency. Summary of the invention

[0004] In view of the above-mentioned problems existing in the existing energy storage cluster dynamic optimization control system based on edge computing, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is that the traditional power prediction method is more sensitive to the linear characteristics of the time series, and it is difficult to cope with the nonlinear changes in the operation of the energy storage cluster, resulting in inaccurate prediction results. At the same time, most of the existing energy scheduling algorithms are based on fixed rules and models, and lack a real-time feedback mechanism for the dynamic path loss between the energy storage node and the load node, resulting in low power allocation and transmission efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a dynamic optimization control system of an energy storage cluster based on edge computing, which includes:

[0007] Data collection module, which collects historical data of energy storage clusters and local data of edge nodes;

[0008] The power prediction module uses the ARIMA model to predict time series, uses the BP neural network to identify and correct nonlinear errors, outputs the final power prediction value, deploys the model based on edge nodes, and performs global updates;

[0009] The two-layer optimization model module builds a two-layer optimization model to dynamically allocate and optimize the paths of energy storage clusters. It calculates the transmission path and loss based on the lower-layer optimization model, and feeds back the transmission loss to affect the power allocation of the upper-layer optimization model. It iteratively calculates the power allocation results.

[0010] Dynamic unloading strategy module, which defines scheduling strategy based on energy storage cluster and dynamic unloading strategy based on edge nodes;

[0011] The anomaly detection module builds an anomaly detection model based on the LSTM neural network to identify abnormalities in the operation of the energy storage cluster and perform local and cloud storage backup of edge node data.

[0012] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the collection of energy storage cluster historical data and edge node local data includes:

[0013] Based on the historical data of the energy storage cluster, including the real-time operating status data of each node's power output, load, voltage, and current, as well as the environmental variable data of temperature and humidity, data collection and normalization are performed;

[0014] The local data corresponding to the edge nodes of the energy storage cluster are collected and stored.

[0015] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the ARIMA model is used to predict the time series, the BP neural network is used to identify and correct the nonlinear error, and the final power prediction value is output, including:

[0016] Based on the normalized historical power data of the energy storage cluster, the ARIMA model is used to perform time series forecasting, which is expressed as:

[0017] ;

[0018] in represents the power prediction value at time point t, Represents the historical power data at time ti, represents the error at time point tj, represents the autoregressive term coefficient, represents the moving average coefficient, and Represent the orders of autoregression and moving average respectively;

[0019] The maximum likelihood estimation method (MLE) is used to estimate the parameters of the ARIMA model, and the model is trained with historical data. MLE is used to optimize and determine and ;

[0020] Construct a BP neural network model, including an input layer, a hidden layer and an output layer. The input layer inputs the output of the ARIMA model, the hidden layer identifies and corrects the nonlinear error of the output of the ARIMA model, and the output layer outputs the final power prediction value, which is expressed as:

[0021] ;

[0022] in Indicates the real power value, represents the predicted power value, N represents the total number of real data, and E represents the error term;

[0023] Use the training set to train the model, select the cross entropy loss function to calculate the difference between the prediction of the BP neural network model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the weights of the BP neural network model, and stop iterating if the loss of the model no longer decreases significantly during the continuous iteration process, output the model parameters and update the model;

[0024] After completing the power forecast, the sum of the linear prediction of the ARIMA model and the nonlinear error correction of the BP neural network is used as the final power forecast value, which is also used as the basis for the next step of optimal scheduling.

[0025] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the edge node-based deployment model and global update include:

[0026] The local BP neural network model is trained for each edge node, and the training data used is the local power history data;

[0027] After local training is completed, the updated model parameters are sent to the central server;

[0028] The central server performs weighted averaging of the model parameters of all nodes and updates the global model;

[0029] The updated global model is sent back to each edge node for the next round of training.

[0030] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the construction of a two-layer optimization model for dynamic allocation and path optimization of the energy storage cluster includes:

[0031] Based on the calculation of the final power prediction value, a two-layer optimization model is constructed to perform dynamic allocation and path optimization of the energy storage cluster, including an upper optimization model and a lower optimization model;

[0032] Define the module or device that receives and consumes the electric energy released by the energy storage unit as a load node;

[0033] The goal of the upper optimization model is defined as minimizing the total energy consumption and operating cost. The objective function is expressed as:

[0034] ;

[0035] in represents the total power consumption, M represents the total number of energy storage nodes, represents the power consumption cost of the i-th energy storage node, represents the power consumption cost of the i-th load node;

[0036] The goal of the lower optimization model is defined as finding the transmission path with the least loss between the energy storage unit and the load node through the shortest path algorithm. The objective function is expressed as:

[0037] ;

[0038] ;

[0039] Where M and B represent the number of energy storage nodes and load nodes respectively. represents the power transmission loss between energy storage node i and load node j, s represents the total transmission loss, represents the line resistance between nodes i and j, represents the current between nodes i and j;

[0040] The Dijkstra algorithm is used to calculate the shortest path from the energy storage node to the load node.

[0041] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the transmission path and loss are calculated based on the lower-level optimization model, and the transmission loss is fed back to affect the power allocation of the upper-level optimization model, and the power allocation result is iteratively calculated, including:

[0042] Initialize power allocation, and each energy storage node is allocated according to its remaining power and the demand of the load node, which is expressed as:

[0043] ;

[0044] in represents the power allocated to load node j by energy storage and energy saving, represents the total available energy of the energy storage unit at time t, and denote the energy requirements of load nodes j and k at time t, respectively, and B denotes the number of load nodes;

[0045] The calculated initial power allocation of each energy storage unit is substituted into the objective function of the upper optimization model to calculate the total consumption of the initial power allocation, which is expressed as:

[0046] ;

[0047] in represents the total power allocation cost, and They represent the power consumption cost of energy storage node i and the power demand cost of load node j respectively;

[0048] Based on the lower optimization model, the Dijkstra algorithm is used to calculate the transmission path and loss between the energy storage node and the load node, and the transmission loss is fed back to affect the power distribution of the upper optimization model, which is expressed as:

[0049] ;

[0050] in represents the total power cost, including the distribution power cost and transmission loss;

[0051] According to the feedback transmission loss, the power allocation strategy is adjusted using the gradient descent method to minimize the objective function , calculate the new power allocation scheme;

[0052] A convergence threshold is defined based on historical data. During multiple iterations of power allocation, when the change of the objective function is lower than the preset convergence threshold, the iteration is terminated to obtain the final power allocation result.

[0053] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the scheduling strategy defined based on the energy storage cluster includes:

[0054] Balance load demand, renewable energy generation, energy storage system discharge and grid power supply, and make sure that the sum of renewable energy power, grid power and energy storage unit discharge power is equal to the sum of total load demand and energy storage system charging power;

[0055] Define the dispatch strategy and check the power output of renewable energy. If the output of renewable energy exceeds the load demand, it is used to charge the energy storage unit. If the output of renewable energy is not enough to meet the load demand, the insufficient power is supplemented by the energy storage unit. If the energy storage unit is also insufficient to meet the load demand, it is supplemented by the power grid.

[0056] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the dynamic unloading strategy defined based on edge nodes includes:

[0057] Based on the edge nodes that have completed training, define dynamic unloading strategies, monitor the operating data of each edge node in real time, calculate the load rate, and monitor the network bandwidth;

[0058] The offloading feasibility index for edge node tasks is calculated as follows:

[0059] ;

[0060] in represents the offloading feasibility index of the kth task, represents the delay time of task k being processed locally, represents the delay time from task k to edge cloud processing, Indicates the current available network bandwidth. represents the bandwidth required for offloading task k, represents the maximum computing power of edge node n, Indicates the current performance utilization of node n;

[0061] The feasibility index of offloadable tasks is calculated based on historical data, and the sum of the mean and standard deviation is used as the feasibility threshold. If it is greater than or equal to the feasibility threshold, it means that task k can be offloaded;

[0062] Based on the decision of whether to offload task K, determine the offloading decision variable , if task k is offloaded to the edge cloud, it means 1, otherwise it means 0;

[0063] Define the execution condition, if Greater than , it means that the task can be offloaded if the total bandwidth requirement of the offloaded task is less than , it means that the task can be offloaded. If the load rate of the edge node exceeds the maximum load rate, it means that the task can be offloaded;

[0064] Based on minimizing the total energy consumption and delay, a dynamic decision is defined, which is expressed as:

[0065] ;

[0066] Where minG represents the minimization optimization target, Indicates the total number of tasks, and They represent the energy consumption of task k offloading to the edge cloud and the energy consumption of task k processed locally;

[0067] Based on the minimization optimization goal, the offload task is executed in real time, and the network transmission is performed, and the task is handed over to the cloud node for processing;

[0068] Continuously monitor the offloaded tasks to ensure that they are completed within the scheduled time. If network congestion or delay exceeds the limit during the offloading process, the system will try to reallocate the tasks back to the local execution.

[0069] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, wherein: the abnormality detection model is constructed based on the LSTM neural network to identify the abnormal operation of the energy storage cluster, including:

[0070] An anomaly detection model is built based on the LSTM neural network, including an input layer, an LSTM layer, and an output layer. The input layer inputs the historical operation data of the energy storage cluster, the LSTM layer is used to capture the dynamic changes in the time series, and the output layer outputs the operation status of the energy storage cluster at the next time step.

[0071] Use the cross entropy loss function selected by the training set to calculate the difference between the prediction result of the anomaly detection model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the parameters of the anomaly detection model, and stop iterating and output the model parameters when the model loss no longer decreases significantly during continuous iterations;

[0072] The sum of the mean value and double standard deviation of historical data is used as the abnormal threshold. If the predicted output of the anomaly detection model is greater than the abnormal threshold, it is regarded as an operational abnormality of the energy storage unit, and dynamic load balancing scheduling is performed based on the number of other non-faulty energy storage units.

[0073] As a preferred solution of the energy storage cluster dynamic optimization control system based on edge computing described in the present invention, the local and cloud storage backup of edge node data includes:

[0074] Based on the determination of abnormal operation of the energy storage unit, an abnormal alarm is issued to maintenance personnel through an alarm;

[0075] The alarm time and data are recorded using RSA encryption and transmitted to the cloud database using the TLS security protocol;

[0076] Synchronize data between edge nodes and central servers for local backup and cloud storage.

[0077] The beneficial effects of the present invention are as follows: through the short-term trend prediction of ARIMA and the error correction of the BP neural network, it is possible to respond to power fluctuations more quickly, optimize power distribution, and ensure the stable operation of the energy storage system in the power grid; by training the local BP neural network model on each edge node, it is possible to make full use of distributed computing resources; through the synergistic effect of the two-layer optimization model, the final power distribution plan and transmission path optimization results are output to ensure efficient operation under fluctuations in load demand and power supply, reduce energy consumption and maximize the use of renewable energy; through the dynamic unloading strategy for each edge node, it is ensured that even when problems occur in cloud processing, the normal operation of the task can still be maintained, thereby enhancing the robustness of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0079] Figure 1 This is a schematic diagram of the structure of the dynamic optimization control system of the energy storage cluster based on edge computing.

[0080] Figure 2 This is a flow chart of the dynamic optimization control system of the energy storage cluster based on edge computing. DETAILED DESCRIPTION

[0081] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0082] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0083] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0084] Example 1, reference Figure 1 and Figure 2, which is the first embodiment of the present invention, and provides a dynamic optimization control system for energy storage clusters based on edge computing. The dynamic optimization control system for energy storage clusters based on edge computing includes:

[0085] S1, collects historical data of energy storage cluster and local data of edge nodes;

[0086] Preferably, the historical data of the energy storage cluster and the local data of the edge nodes are collected, including:

[0087] Based on the historical data of the energy storage cluster, including the real-time operating status data of each node's power output, load, voltage, and current, as well as the environmental variable data of temperature and humidity, data collection and normalization are performed;

[0088] The local data corresponding to the edge nodes of the energy storage cluster are collected and stored.

[0089] By collecting historical data of the energy storage cluster (such as power output, load, etc.), combined with real-time operating status (such as voltage, current) and normalized environmental variables (such as temperature, humidity), data support is provided for predicting energy storage demand and load. Data normalization can ensure that data of different units, ranges and magnitudes can be processed uniformly, reducing the impact of noise and outliers on system decisions.

[0090] S2, uses the ARIMA model to predict time series, uses the BP neural network to identify and correct nonlinear errors, outputs the final power prediction value, deploys the model based on edge nodes and performs global updates;

[0091] Preferably, the ARIMA model is used to predict the time series, the BP neural network is used to identify and correct the nonlinear error, and the final power prediction value is output, including:

[0092] Based on the normalized historical power data of the energy storage cluster, the ARIMA model is used to perform time series forecasting, which is expressed as:

[0093] ;

[0094] in represents the power prediction value at time point t, Represents the historical power data at time ti, represents the error at time point tj, represents the autoregressive term coefficient, represents the moving average coefficient, and Represent the order of autoregression and moving average respectively. Based on the autocorrelation function (ACF) and partial autocorrelation function (PACF), the ACF and PACF diagrams are used to determine the autoregression order p and the moving average order q of the ARIMA model. In the ACF diagram, the significant attenuation of the lagged value can help determine the appropriate q value. In the PACF diagram, the truncation of the lagged value can be used to determine the size of p.

[0095] The maximum likelihood estimation method (MLE) is used to estimate the parameters of the ARIMA model, and the model is trained with historical data. MLE is used to optimize and determine and ;

[0096] Construct a BP neural network model, including an input layer, a hidden layer and an output layer. The input layer inputs the output of the ARIMA model, the hidden layer identifies and corrects the nonlinear error of the output of the ARIMA model, and the output layer outputs the final power prediction value, which is expressed as:

[0097] ;

[0098] in Indicates the real power value, represents the predicted power value, N represents the total number of real data, and E represents the error term;

[0099] Use the training set to train the model, select the cross entropy loss function to calculate the difference between the prediction of the BP neural network model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the weights of the BP neural network model, and stop iterating if the loss of the model no longer decreases significantly during the continuous iteration process, output the model parameters and update the model;

[0100] After completing the power forecast, the sum of the linear prediction of the ARIMA model and the nonlinear error correction of the BP neural network is used as the final power forecast value, which is also used as the basis for the next step of optimal scheduling.

[0101] The ARIMA model can capture seasonal fluctuations and cyclical changes in power output, so that the energy storage system can more accurately predict future power demand and load fluctuations. The BP neural network can capture nonlinear trends that the ARIMA model cannot identify during the training process, identify nonlinear errors in the prediction and correct them. Through the combination of ARIMA and BP neural network, the system can accurately predict future power demand and provide a reliable basis for optimized scheduling. Through the short-term trend prediction of ARIMA and the error correction of BP neural network, the system can respond to power fluctuations more quickly, optimize power distribution, and ensure the stable operation of the energy storage system in the power grid. By analyzing the ACF and PACF graphs, the autoregressive (AR) order and moving average order in the ARIMA model can be determined, so as to establish a suitable model. The analysis of the ACF and PACF graphs reduces the trial and error process and makes the model structure more reasonable. The optimal parameters are estimated by maximizing the probability of the observed data appearing under the given model. Through the short-term trend prediction of ARIMA and the error correction of BP neural network, the system can respond to power fluctuations more quickly, optimize power distribution, and ensure the stable operation of the energy storage system in the power grid.

[0102] Furthermore, the model is deployed on edge nodes and updated globally, including:

[0103] The local BP neural network model is trained for each edge node, and the training data used is the local power history data;

[0104] After local training is completed, the updated model parameters (such as weights and biases) are sent to the central server;

[0105] The central server performs weighted averaging of the model parameters of all nodes and updates the global model;

[0106] The updated global model is sent back to each edge node for the next round of training.

[0107] By training the local BP neural network model on each edge node, the system can make full use of distributed computing resources and avoid the overhead of transmitting all data to the central server. By transmitting only the parameters of the model (such as weights and biases) instead of the entire data set, the system can significantly reduce the demand for network bandwidth. After the central server performs weighted averaging on the model parameters of each edge node, the updated global model has a wider range of representativeness and can combine the local information of each edge node, thereby accelerating the convergence of the global model. By aggregating the model parameters of different nodes, the system can obtain data training results from various environments, load conditions, and power characteristics, thereby improving the generalization ability of the global model.

[0108] S3, builds a two-layer optimization model to perform dynamic allocation and path optimization of the energy storage cluster. Based on the lower-layer optimization model, the transmission path and loss are calculated, and the transmission loss is fed back to affect the power allocation of the upper-layer optimization model. The power allocation result is iteratively calculated.

[0109] Preferably, a two-layer optimization model is constructed to perform dynamic allocation and path optimization of energy storage clusters, including:

[0110] Based on the calculation of the final power prediction value, a two-layer optimization model is constructed to perform dynamic allocation and path optimization of the energy storage cluster, including an upper optimization model and a lower optimization model;

[0111] Define the module or device that receives and consumes the electric energy released by the energy storage unit as a load node;

[0112] The goal of the upper optimization model is defined as minimizing the total energy consumption and operating cost. The objective function is expressed as:

[0113] ;

[0114] in represents the total power consumption, M represents the total number of energy storage nodes, represents the power consumption cost of the i-th energy storage node, represents the power consumption cost of the i-th load node;

[0115] Based on the battery power limit of the energy storage node, it is defined that the capacity state of the energy storage node must be within the maximum and minimum energy storage capacity range of the battery, and the power constraint condition is defined, which is expressed as:

[0116] ;

[0117] in Expressed as the load demand power at time t, It is represented by the power generated by renewable energy at time t, Expressed as the energy storage node discharge power at time t, Expressed as the grid power at time t:

[0118] The goal of the lower optimization model is defined as finding the transmission path with the least loss between the energy storage unit and the load node through the shortest path algorithm. The objective function is expressed as:

[0119] ;

[0120] ;

[0121] Where M and B represent the number of energy storage nodes and load nodes respectively. represents the power transmission loss between energy storage node i and load node j, s represents the total transmission loss, represents the line resistance between nodes i and j, represents the current between nodes i and j;

[0122] The Dijkstra algorithm is used to calculate the shortest path from the energy storage node to the load node. The energy storage node and the matching node are set and the transmission power loss calculated for each edge is assigned. Starting from the energy storage node as the origin, all nodes j adjacent to the energy storage node i are traversed, and the shortest path from the energy storage node to the matching node is updated. When all load nodes are visited, the algorithm terminates and outputs the shortest path from the energy storage unit to the load node and the corresponding loss value.

[0123] Through the synergy of the two-layer optimization model, the system will output the final power allocation plan and transmission path optimization results to ensure that the system can operate efficiently under the fluctuations of load demand and power supply, reduce energy consumption and maximize the utilization of renewable energy. The dynamic power allocation of the energy storage system is combined with the optimization of the power transmission path. The global optimal power management is achieved through the two-layer optimization model, which not only improves the energy efficiency of the system, but also reduces the loss in power transmission. By dynamically adjusting the power allocation strategy of the energy storage system and combining it with real-time path optimization, the system can operate efficiently under different load conditions and maximize the utilization of renewable energy. The upper layer optimization: based on the predicted power demand and battery status, optimize the power allocation strategy of each energy storage unit to make the most of renewable energy. The lower layer optimization: using the power allocation results of the upper layer, calculate the shortest path between the energy storage unit and the load node to ensure the minimum loss of energy transmission. The Dijkstra algorithm can find the shortest path from the starting point to the target node in the weighted graph.

[0124] The objective function of the upper optimization model takes into account the power consumption costs of energy storage nodes and load nodes. By optimizing the power consumption of energy storage nodes, the system can reduce the overall operating cost. The power limit conditions of energy storage nodes enable the system to avoid unnecessary power loss when using electrical energy, thereby reducing the operating costs of energy storage units and the electricity costs of load nodes.

[0125] Based on the maximum and minimum energy storage capacity constraints of the energy storage node battery, the system can ensure the flexibility and accuracy of power dispatch under the conditions of dynamic power demand and fluctuations in renewable energy. The dynamic dispatch of energy storage units enables the system to flexibly adjust the charging and discharging behavior of energy storage units according to the comprehensive situation of load demand, renewable energy power output and grid power, so as to better cope with changes in real-time power demand.

[0126] The lower-level optimization model uses the Dijkstra algorithm to calculate the shortest path from the energy storage node to the load node, and selects the power transmission path with the least loss based on the objective function of power transmission loss. By optimizing the power transmission path, the energy storage system can respond quickly when the load node demand changes and select the optimal path for power transmission. By reasonably allocating the power output of the energy storage unit and avoiding over-discharge or over-charging, the usage pattern of the energy storage unit can be optimized and the life of the energy storage equipment can be extended.

[0127] Furthermore, the transmission path and loss are calculated based on the lower optimization model, and the transmission loss is fed back to affect the power allocation of the upper optimization model, and the power allocation result is iteratively calculated, including,

[0128] Initialize power allocation, and each energy storage node is allocated according to its remaining power and the demand of the load node, which is expressed as:

[0129] ;

[0130] in represents the power allocated to load node j by energy storage and energy saving, represents the total available energy of the energy storage unit at time t, and denote the energy requirements of load nodes j and k at time t, respectively, and B denotes the number of load nodes;

[0131] The calculated initial power allocation of each energy storage unit is substituted into the objective function of the upper optimization model to calculate the total consumption of the initial power allocation, which is expressed as:

[0132] ;

[0133] in represents the total power allocation cost, and They represent the power consumption cost of energy storage node i and the power demand cost of load node j respectively;

[0134] Based on the lower optimization model, the Dijkstra algorithm is used to calculate the transmission path and loss between the energy storage node and the load node, and the transmission loss is fed back to affect the power distribution of the upper optimization model, which is expressed as:

[0135] ;

[0136] in represents the total power cost, including the distribution power cost and transmission loss;

[0137] According to the feedback transmission loss, the power allocation strategy is adjusted using the gradient descent method to minimize the objective function , calculate the new power allocation scheme, expressed as:

[0138] ;

[0139] ;

[0140] in Indicates power distribution The gradient of the total cost represents the effect of the change in power allocation on the total cost. Represents the objective function Power distribution The partial derivative of Indicates power transmission loss Power distribution The partial derivative of and They represent the power distribution after and before the update, represents the learning rate;

[0141] A convergence threshold is defined based on historical data (the convergence threshold is a minimum value). In multiple iterations of power allocation, when the change of the objective function is lower than the preset convergence threshold, the iteration is terminated to obtain the final power allocation result.

[0142] By calculating the shortest path and transmission loss between the energy storage node and the load node through the lower-level optimization model and feeding it back to the upper-level optimization model, the system can adjust the power distribution more accurately. The feedback mechanism can ensure that the power distribution plan not only takes into account the allocated power demand, but also the energy loss in the transmission path, thereby improving the accuracy of power distribution and the overall operating efficiency of the system. The goal of the upper-level optimization model is to minimize total energy consumption and operating costs. Through the feedback of initial power allocation and transmission loss, the system can optimize the power allocation. The lower-level optimization model quantifies the transmission loss by calculating the resistance and current in the transmission path, and feeds back to the upper-level model for adjustment, so that the system can choose a more efficient power transmission path, further reducing the overall energy consumption and operating costs. The gradient descent method is used to iteratively adjust the power allocation strategy multiple times, so that the system can make adaptive adjustments under real-time conditions. With the feedback of transmission loss and power consumption, the system can automatically optimize the power allocation plan to ensure that each energy storage node dynamically adjusts its output power according to load demand and transmission conditions, so that the adaptive power allocation mechanism can better cope with fluctuations and emergencies in the energy storage system. Through iterative optimization of power allocation and transmission paths, the system can flexibly respond to changes in different load demands and power conditions of energy storage nodes. In each iteration, the power allocation plan is adjusted to adapt to actual power transmission conditions and load demand changes, thereby improving the robustness of the system.

[0143] S4, defines the scheduling strategy based on the energy storage cluster and the dynamic unloading strategy based on the edge node;

[0144] Preferably, a dispatch strategy is defined based on the energy storage cluster, including:

[0145] Balance load demand, renewable energy generation, energy storage system discharge and grid power supply, and make sure that the sum of renewable energy power, grid power and energy storage unit discharge power is equal to the sum of total load demand and energy storage system charging power;

[0146] Define the dispatch strategy and check the power output of renewable energy. If the output of renewable energy exceeds the load demand, it is used to charge the energy storage unit. If the output of renewable energy is not enough to meet the load demand, the insufficient power is supplemented by the energy storage unit. If the energy storage unit is also insufficient to meet the load demand, it is supplemented by the power grid.

[0147] By giving priority to the use of renewable energy for power generation and energy storage charging, energy waste is effectively avoided. By using the energy storage system to supplement the load when renewable energy is insufficient, dependence on the power grid is further reduced, thereby reducing the demand for fossil energy power generation and improving the overall energy efficiency of the system. This strategy enables the system to flexibly adjust according to changes in renewable energy generation, the remaining energy of the energy storage system, and changes in load demand, ensuring that load demand is always met.

[0148] Furthermore, a dynamic offloading strategy is defined based on edge nodes, including:

[0149] Based on the edge nodes that have completed training, define dynamic unloading strategies, monitor the operating data of each edge node in real time, calculate the load rate, and monitor the network bandwidth;

[0150] The offloading feasibility index for edge node tasks is calculated as follows:

[0151] ;

[0152] in represents the offloading feasibility index of the kth task, represents the delay time of task k being processed locally, represents the delay time from task k to edge cloud processing, Indicates the current available network bandwidth. represents the bandwidth required for offloading task k, represents the maximum computing power of edge node n, Indicates the current performance utilization of node n;

[0153] The feasibility index of offloadable tasks is calculated based on historical data, and the sum of the mean and standard deviation is used as the feasibility threshold. If it is greater than or equal to the feasibility threshold, it means that task k can be offloaded;

[0154] Based on the decision of whether to offload task K, determine the offloading decision variable , if task k is offloaded to the edge cloud, it means 1, otherwise it means 0;

[0155] Define the execution condition, if Greater than , it means that the task can be offloaded if the total bandwidth requirement of the offloaded task is less than , it means that the task can be offloaded. If the load rate of the edge node exceeds the maximum load rate, it means that the task can be offloaded;

[0156] Based on minimizing the total energy consumption and delay, a dynamic decision is defined, which is expressed as:

[0157] ;

[0158] Where minG represents the minimization optimization target, Indicates the total number of tasks, and They represent the energy consumption of task k offloading to the edge cloud and the energy consumption of task k processed locally;

[0159] Based on the minimization optimization goal, the offload task is executed in real time, and the network transmission is performed, and the task is handed over to the cloud node for processing;

[0160] Continuously monitor the offloaded tasks to ensure that they are completed within the scheduled time. If network congestion or delay exceeds the limit during the offloading process, the system will try to reallocate the tasks back to the local execution.

[0161] By real-time monitoring of the operating data and computing load rate of edge nodes, the dynamic offloading strategy can intelligently offload tasks to the edge cloud. The dynamic offloading strategy optimizes task allocation by minimizing total energy consumption to ensure the optimal balance between local processing and edge cloud offloading. By offloading tasks with high computing requirements to the edge cloud, the system can reduce the energy consumption of local nodes while avoiding excessive transmission power consumption. By optimizing energy consumption, the system significantly reduces energy usage during long-term operation. By monitoring network bandwidth and incorporating it into offloading decisions, the system can ensure that tasks are only offloaded when network conditions permit. During the process of offloading tasks, tasks are reallocated back to local processing based on network congestion or task delays. This ensures that even when problems occur in cloud processing, the system can still maintain the normal operation of tasks, enhancing the robustness of the system. Through dynamic offloading strategies for each edge node, the solution can adapt to large-scale distributed computing scenarios, ensuring efficient use of computing resources and on-time completion of tasks.

[0162] S5, builds an anomaly detection model based on LSTM neural network to identify abnormal operation of energy storage clusters and back up edge node data locally and in the cloud;

[0163] Preferably, an anomaly detection model is constructed based on an LSTM neural network to identify abnormal operation of the energy storage cluster, including:

[0164] An anomaly detection model is built based on the LSTM neural network, including an input layer, an LSTM layer, and an output layer. The input layer inputs the historical operation data of the energy storage cluster, the LSTM layer is used to capture the dynamic changes in the time series, and the output layer outputs the operation status of the energy storage cluster at the next time step.

[0165] Use the cross entropy loss function selected by the training set to calculate the difference between the prediction result of the anomaly detection model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the parameters of the anomaly detection model, and stop iterating and output the model parameters when the model loss no longer decreases significantly during continuous iterations;

[0166] The sum of the mean value and double standard deviation of historical data is used as the abnormal threshold. If the predicted output of the abnormal detection model is greater than the abnormal threshold, it is regarded as an abnormal operation of the energy storage unit, and dynamic load balancing scheduling is performed based on the number of other energy storage units that have not failed, which is expressed as:

[0167] ;

[0168] in Indicates the new power allocation value after load balancing, Indicates the power before load balancing, represents the power of the faulty energy storage unit, and Z represents the total number of energy storage units.

[0169] Neural networks are good at processing time series data, and can capture the dynamic changes of historical data in energy storage clusters, identify potential abnormal patterns in energy storage systems, and predict possible failures in advance, thereby improving the prediction accuracy of energy storage system status and reducing sudden failures during system operation. Anomaly detection based on the LSTM model can capture the operating status of the energy storage cluster in real time, especially when abnormal fluctuations occur in data such as power output, load, voltage, and current. The system can respond quickly. By comparing the predicted output with the abnormal threshold, the system can promptly identify the operating abnormalities of the energy storage unit, thereby improving the reliability and safety of the system. By dynamically adjusting the power distribution of the energy storage unit, the energy storage unit that is not faulty can effectively take over the load of the faulty unit, avoiding the waste of power resources, and improving the energy utilization rate of the system, ensuring the efficient management and use of power resources.

[0170] Furthermore, edge node data is backed up locally and in the cloud, including:

[0171] Based on the determination of abnormal operation of the energy storage unit, an abnormal alarm is issued to maintenance personnel through an alarm;

[0172] The alarm time and data are recorded using RSA encryption and transmitted to the cloud database using the TLS security protocol;

[0173] Synchronize data between edge nodes and central servers for local backup and cloud storage.

[0174] By using RSA encryption technology to encrypt alarm time and data, the system can ensure that sensitive information will not be stolen or tampered with during data transmission. By using TLS (Transport Layer Security Protocol), it can ensure that data will not be attacked by middlemen or network eavesdropping during transmission. The abnormal alarm mechanism helps maintenance personnel take timely actions to avoid the expansion of faults or system interruptions, thereby improving the overall operational stability and fault handling efficiency of the system. Through the dual data storage method of local backup and cloud storage, the system ensures that all abnormal data and alarm information are safely stored.

[0175] Example 2

[0176] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0177] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0178] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0179] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit with a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit with a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0180] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. The energy storage cluster dynamic optimization control system based on edge computing is characterized by: include, Data collection module, which collects historical data of energy storage clusters and local data of edge nodes; The power prediction module uses the ARIMA model to predict time series, uses the BP neural network to identify and correct nonlinear errors, outputs the final power prediction value, deploys the model based on edge nodes, and performs global updates; The two-layer optimization model module builds a two-layer optimization model to dynamically allocate and optimize the paths of energy storage clusters. It calculates the transmission path and loss based on the lower-layer optimization model, and feeds back the transmission loss to affect the power allocation of the upper-layer optimization model. It iteratively calculates the power allocation results. Dynamic unloading strategy module, which defines scheduling strategy based on energy storage cluster and dynamic unloading strategy based on edge nodes; The anomaly detection module builds an anomaly detection model based on the LSTM neural network to identify abnormal operation of the energy storage cluster and perform local and cloud storage backup of edge node data; The energy storage cluster historical data and edge node local data are collected. include, Based on the historical data of the energy storage cluster, including the real-time operating status data of each node's power output, load, voltage, and current, as well as the environmental variable data of temperature and humidity, data collection and normalization are performed; Collect and store local data corresponding to the edge nodes of the energy storage cluster; The ARIMA model is used to predict the time series, the BP neural network is used to identify and correct the nonlinear error, and the final power prediction value is output, including: Based on the normalized historical power data of the energy storage cluster, the ARIMA model is used to perform time series forecasting, which is expressed as: ； in represents the power prediction value at time point t, Represents the historical power data at time ti, represents the error at time point tj, represents the autoregressive term coefficient, represents the moving average coefficient, and Represent the orders of autoregression and moving average respectively; The maximum likelihood estimation method (MLE) is used to estimate the parameters of the ARIMA model, and the model is trained with historical data. MLE is used to optimize and determine and ; Construct a BP neural network model, including an input layer, a hidden layer and an output layer. The input layer inputs the output of the ARIMA model, the hidden layer identifies and corrects the nonlinear error of the output of the ARIMA model, and the output layer outputs the final power prediction value, which is expressed as: ； in Indicates the real power value, represents the predicted power value, N represents the total number of real data, and E represents the error term; Use the training set to train the model, select the cross entropy loss function to calculate the difference between the prediction of the BP neural network model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the weights of the BP neural network model, and stop iterating if the loss of the model no longer decreases significantly during the continuous iteration process, output the model parameters and update the model; After completing the power forecast, the sum of the linear prediction of the ARIMA model and the nonlinear error correction of the BP neural network is used as the final power forecast value, which is used as the basis for the next step of optimized scheduling; The edge node-based model deployment and global update includes: The local BP neural network model is trained for each edge node, and the training data used is the local power history data; After local training is completed, the updated model parameters are sent to the central server; The central server performs weighted averaging of the model parameters of all nodes and updates the global model; The updated global model is sent back to each edge node for the next round of training.

2. The energy storage cluster dynamic optimization control system based on edge computing according to claim 1, characterized in that: The two-layer optimization model is constructed to dynamically allocate and optimize the paths of energy storage clusters. include, Based on the calculation of the final power prediction value, a two-layer optimization model is constructed to perform dynamic allocation and path optimization of the energy storage cluster, including an upper optimization model and a lower optimization model; Define the module or device that receives and consumes the electric energy released by the energy storage unit as a load node; The goal of the upper optimization model is defined as minimizing the total energy consumption and operating cost. The objective function is expressed as: ； in represents the total power consumption, M represents the total number of energy storage nodes, represents the power consumption cost of the i-th energy storage node, represents the power consumption cost of the i-th load node; The goal of the lower optimization model is defined as finding the transmission path with the least loss between the energy storage unit and the load node through the shortest path algorithm. The objective function is expressed as: ； ； Where M and B represent the number of energy storage nodes and load nodes respectively. represents the power transmission loss between energy storage node i and load node j, s represents the total transmission loss, represents the line resistance between nodes i and j, represents the current between nodes i and j; The Dijkstra algorithm is used to calculate the shortest path from the energy storage node to the load node.

3. The energy storage cluster dynamic optimization control system based on edge computing as claimed in claim 2, characterized in that: The transmission path and loss are calculated based on the lower optimization model, and the transmission loss is fed back to affect the power allocation of the upper optimization model, and the power allocation result is iteratively calculated, including: Initialize power allocation, and each energy storage node is allocated according to its remaining power and the demand of the load node, which is expressed as: ； in represents the power allocated to load node j by energy storage and energy saving, represents the total available energy of the energy storage unit at time t, and denote the energy requirements of load nodes j and k at time t, respectively, and B denotes the number of load nodes; The calculated initial power allocation of each energy storage unit is substituted into the objective function of the upper optimization model to calculate the total consumption of the initial power allocation, which is expressed as: ； in represents the total power allocation cost, and They represent the power consumption cost of energy storage node i and the power demand cost of load node j respectively; Based on the lower optimization model, the Dijkstra algorithm is used to calculate the transmission path and loss between the energy storage node and the load node, and the transmission loss is fed back to affect the power distribution of the upper optimization model, which is expressed as: ； in represents the total power cost, including the distribution power cost and transmission loss; According to the feedback transmission loss, the power allocation strategy is adjusted using the gradient descent method to minimize the objective function , calculate the new power allocation scheme; A convergence threshold is defined based on historical data. During multiple iterations of power allocation, when the change of the objective function is lower than the preset convergence threshold, the iteration is terminated to obtain the final power allocation result.

4. The energy storage cluster dynamic optimization control system based on edge computing as claimed in claim 3, characterized in that: The scheduling strategy defined based on the energy storage cluster includes: Balance load demand, renewable energy generation, energy storage system discharge and grid power supply, and make sure that the sum of renewable energy power, grid power and energy storage unit discharge power is equal to the sum of total load demand and energy storage system charging power; Define the dispatch strategy and check the power output of renewable energy. If the output of renewable energy exceeds the load demand, it is used to charge the energy storage unit. If the output of renewable energy is not enough to meet the load demand, the insufficient power is supplemented by the energy storage unit. If the energy storage unit is also insufficient to meet the load demand, it is supplemented by the power grid.

5. The energy storage cluster dynamic optimization control system based on edge computing as claimed in claim 4, characterized in that: The dynamic unloading strategy based on edge node definition includes: Based on the edge nodes that have completed training, define dynamic unloading strategies, monitor the operating data of each edge node in real time, calculate the load rate, and monitor the network bandwidth; The offloading feasibility index for edge node tasks is calculated as follows: ； in represents the offloading feasibility index of the kth task, represents the delay time of task k being processed locally, represents the delay time from task k to edge cloud processing, Indicates the current available network bandwidth. represents the bandwidth required for offloading task k, represents the maximum computing power of edge node n, Indicates the current performance utilization of node n; The feasibility index of offloadable tasks is calculated based on historical data, and the sum of the mean and standard deviation is used as the feasibility threshold. If it is greater than or equal to the feasibility threshold, it means that task k can be offloaded; Based on the decision of whether to offload task K, determine the offloading decision variable , if task k is offloaded to the edge cloud, it means 1, otherwise it means 0; Define the execution condition, if Greater than , it means that the task can be offloaded if the total bandwidth requirement of the offloaded task is less than , it means that the task can be offloaded. If the load rate of the edge node exceeds the maximum load rate, it means that the task can be offloaded; Based on minimizing the total energy consumption and delay, a dynamic decision is defined, which is expressed as: ； Where minG represents the minimization optimization target, Represents the total number of tasks, and They represent the energy consumption of offloading task k to the edge cloud and the energy consumption of processing task k locally; Based on the minimization optimization goal, the offload task is executed in real time, and the network transmission is performed, and the task is handed over to the cloud node for processing; Continuously monitor the offloaded tasks to ensure that they are completed within the scheduled time. If network congestion or delay exceeds the limit during the offloading process, the system will try to reallocate the tasks back to the local execution.

6. The energy storage cluster dynamic optimization control system based on edge computing as claimed in claim 5, characterized in that: The anomaly detection model is constructed based on the LSTM neural network to identify the abnormal operation of the energy storage cluster, including: An anomaly detection model is built based on the LSTM neural network, including an input layer, an LSTM layer, and an output layer. The input layer inputs the historical operation data of the energy storage cluster, the LSTM layer is used to capture the dynamic changes in the time series, and the output layer outputs the operation status of the energy storage cluster at the next time step. Use the cross entropy loss function selected by the training set to calculate the difference between the prediction result of the anomaly detection model and the actual label, use the Adam optimizer to perform gradient descent optimization, update the parameters of the anomaly detection model, and stop iterating and output the model parameters when the model loss no longer decreases significantly during continuous iterations; The sum of the mean value and double standard deviation of historical data is used as the abnormal threshold. If the predicted output of the anomaly detection model is greater than the abnormal threshold, it is regarded as an operational abnormality of the energy storage unit, and dynamic load balancing scheduling is performed based on the number of other non-faulty energy storage units.

7. The energy storage cluster dynamic optimization control system based on edge computing according to claim 6, characterized in that: The edge node data is backed up locally and in the cloud. include, Based on the determination of abnormal operation of the energy storage unit, an abnormal alarm is issued to maintenance personnel through an alarm; The alarm time and data are recorded using RSA encryption and transmitted to the cloud database using the TLS security protocol; Synchronize data between edge nodes and central servers for local backup and cloud storage.

Images