Energy storage efficiency improvement method and system combined with energy management

Abstract

The present invention discloses a method and system for improving energy storage efficiency in combination with energy management, which relates to the field of smart grid technology. The method comprises the following steps: collecting energy storage unit status data, environmental data and load data through a sensor network, obtaining grid information through the Internet, and processing the data in real time; constructing a prediction model based on a multi-level memory network to predict future load demand and power supply; generating preliminary charge and discharge instructions based on the prediction results and combining the real-time status of the energy storage unit with the grid information, and sending the instructions to the edge node; making preliminary decisions on the preliminary charge and discharge instructions at the edge node, generating local charge and discharge plans, and uploading them to a cloud server; the cloud server summarizes the local charge and discharge plans, performs global optimization, generates final charge and discharge instructions, and sends them to the energy storage unit; the present invention significantly improves the accuracy of the prediction by constructing a prediction model based on the multi-level memory network, and ensures the credibility of the prediction results by embedding a physical constraint mechanism.

Description

Technical Field

[0001] The present invention relates to the field of smart grid technology, and in particular to a method and system for improving energy storage efficiency in combination with energy management. Background Art

[0002] With the transformation of the global energy structure and the rapid development of renewable energy, energy storage technology has received unprecedented attention as a key means to balance electricity supply and demand and improve system flexibility. In recent years, chemical energy storage technologies such as lithium-ion batteries and sodium-sulfur batteries, as well as physical energy storage technologies such as supercapacitors and flywheel energy storage, have made significant progress, providing strong support for the stable operation of the power system.

[0003] Although existing energy management systems have improved the operating efficiency of energy storage systems to a certain extent, they still have many shortcomings. In particular, in complex and changing power grid environments, traditional EMS often finds it difficult to achieve accurate predictions and real-time adjustments. Most existing systems rely on static models or historical data for load forecasting, which cannot accurately reflect the changing trends in actual electricity demand, resulting in low prediction accuracy. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for improving energy storage efficiency combined with energy management to solve the problem that the energy management system is insufficient in prediction accuracy, resulting in low energy utilization efficiency.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for improving energy storage efficiency in combination with energy management, which comprises:

[0008] Collect energy storage unit status data, environmental data, and load data through sensor networks, obtain grid information through the Internet, and process it in real time;

[0009] Build a prediction model based on a multi-level memory network to predict future load demand and power supply;

[0010] Based on the prediction results, combined with the real-time status of the energy storage unit and grid information, preliminary charging and discharging instructions are generated and sent to the edge node;

[0011] Make preliminary decisions on the initial charge and discharge instructions at the edge node, generate a local charge and discharge plan, and upload it to the cloud server;

[0012] The cloud server summarizes the local charging and discharging plans, performs global optimization, generates the final charging and discharging instructions, and sends them to the energy storage unit.

[0013] As a preferred solution of the energy storage efficiency improvement method combined with energy management described in the present invention, wherein: the energy storage unit status data, environmental data and load data are collected through the sensor network, the power grid information is obtained through the Internet, and real-time processing is performed. The specific steps are as follows:

[0014] Real-time monitoring and collection of energy storage unit status data, energy storage unit load demand, and energy storage unit environmental data;

[0015] Obtain real-time power grid information through the Internet interface;

[0016] The energy storage unit status data, load demand and environmental data, and grid information are sorted by the timestamp of the acquisition and integrated into a multi-source data set;

[0017] Filter and normalize multi-source datasets.

[0018] As a preferred solution of the energy storage efficiency improvement method combined with energy management described in the present invention, wherein: the prediction model is constructed based on a multi-level memory network to predict future load demand and power supply conditions. The specific steps are as follows:

[0019] Decompose the processed multi-source data set into short-term, medium-term and long-term data subsets according to the time scale;

[0020] Based on the short-term data subset, a short-term memory subnetwork is constructed using standard LSTM units to capture the short-term fluctuation characteristics of the energy storage unit;

[0021] Based on the medium-term data subset, a medium-term memory sub-network is constructed using bidirectional LSTM units to capture the medium-term fluctuation characteristics of the energy storage unit;

[0022] Based on the long-term data subset, the LSTM unit with attention enhancement mechanism is used to build a long-term memory sub-network to capture the long-term fluctuation characteristics of the energy storage unit;

[0023] By using the attention enhancement mechanism to dynamically weight the output vectors of the short-term, medium-term, and long-term sub-networks, the interactive relationship between different time scales is captured, and the fusion vector of the multi-level memory network is obtained. The expression is as follows:

[0024]

[0025] Among them, α i is the attention weight of the i-th sub-network, i is the index of the number of sub-networks, W a is the weight matrix of the attention enhancement mechanism, h i is the hidden vector of the i-th sub-network, b ais the bias term of the attention enhancement mechanism, N is the number of sub-networks, and H is the fusion vector of the multi-level memory network;

[0026] Build a prediction model based on a multi-layer memory network;

[0027] The fusion vector H is received by the fully connected layer of the prediction model for training, and H is projected into the target output space. The physical constraint embedding mechanism is introduced into the prediction formula, and the expression is:

[0028] y=σ(W o H+b o )×clip(A,0,1);

[0029]

[0030] Among them, y is the predicted value, σ is the Sigmoid activation function, W o is the weight matrix of the fully connected layer, b o is the bias vector of the fully connected layer, A is the physical constraint, and clip(A,0,1) means limiting the physical constraint A to [0,1];

[0031] During the prediction model training process, the gradient descent method is used to optimize the prediction model parameters W o and b o ;

[0032] The prediction performance of the model was evaluated using mean square error and mean absolute error;

[0033] The trained prediction model receives real-time multi-source data sets and makes predictions on load demand and power supply.

[0034] As a preferred solution of the energy storage efficiency improvement method combined with energy management described in the present invention, wherein: according to the prediction results, combined with the real-time status of the energy storage unit and the grid information, preliminary charge and discharge instructions are generated and sent to the edge node. The specific steps are as follows:

[0035] The predicted load demand value and power supply value are obtained through the inverse normalization method;

[0036] Through sensor networks and Internet interfaces, real-time status data of energy storage units and grid information are obtained to assess the charging and discharging needs of energy storage units.

[0037] Combine the predicted load demand value and power supply value to generate preliminary charging and discharging instructions;

[0038] The preliminary discharge instructions are formatted into a standard format and transmitted to the edge nodes of each energy storage unit through a communication protocol.

[0039] As a preferred solution of the energy storage efficiency improvement method combined with energy management described in the present invention, wherein: the edge node makes a preliminary decision on the preliminary charge and discharge instructions, generates a local charge and discharge plan, and uploads it to the cloud server. The specific steps are as follows:

[0040] The edge node receives preliminary charge and discharge instructions through the communication protocol;

[0041] Extract charging and discharging power and time by parsing preliminary charging and discharging instructions;

[0042] The edge node obtains the status data and environmental parameters of the energy storage unit in real time through the local sensor network to obtain a local data set;

[0043] Optimize preliminary charge and discharge instructions based on local data sets to generate local charge and discharge plans;

[0044] The local charge and discharge plan is uploaded to the cloud server through the communication protocol.

[0045] As a preferred solution of the energy storage efficiency improvement method combined with energy management described in the present invention, the cloud server summarizes the local charge and discharge plans, performs global optimization, generates the final charge and discharge instructions, and sends them to the energy storage unit. The specific steps are as follows:

[0046] The cloud server receives the local charge and discharge plans uploaded by all edge nodes and integrates them into a global charge and discharge plan;

[0047] The constraints are the safe range of the energy storage unit's state of charge, the rated power of charging and discharging, and the fluctuation range of the energy storage unit on the power grid.

[0048] If the optimized global charge and discharge plan exceeds any constraint, it will be invalidated;

[0049] Use Ant Colony Optimization (ACO) to optimize the global charging and discharging plan;

[0050] The objective function is defined based on maximizing the benefits of the energy storage unit and minimizing the grid frequency fluctuation;

[0051] Ants leave pheromones on the path as virtual markers. In the initialization phase, the pheromone values ​​on all paths are set to the same initial value;

[0052] Define heuristic factors based on the objective function;

[0053] Each ant starts from the first node and selects the next path based on the pheromone concentration and heuristic factor on the path until it traverses the entire time series;

[0054] Calculate the probability of an ant choosing a path. The expression is as follows:

[0055]

[0056] Among them, Q j,j+1 is the probability of an ant moving from node j to node j+1, φ j,j+1 is the pheromone concentration of path j→j+1, ψ j,j+1 is the heuristic factor of path j→j+1, γ is the coefficient of controlling the influence of pheromone on path selection, and δ is the weight coefficient of controlling the influence of heuristic factor on path selection;

[0057] Using pheromone updates to record the historical advantages and disadvantages of paths and guide subsequent ants' path selection;

[0058] The ant algorithm gradually optimizes through multiple rounds of iterations, gradually approaching the optimal global charge and discharge plan until the convergence conditions are met;

[0059] The maximum number of iterations is set according to the scale of the energy storage unit and the predicted time. After the iterative optimization is completed, the final charge and discharge instructions are output and sent to the edge node for execution on the energy storage unit.

[0060] In a second aspect, the present invention provides an energy storage efficiency improvement system combined with energy management, comprising a data acquisition module, a prediction module, a preliminary instruction module, a preliminary decision module and a final instruction module.

[0061] The data acquisition module is used to collect energy storage unit status data, environmental data and load data through the sensor network, obtain power grid information through the Internet, and process it in real time;

[0062] The prediction module is used to build a prediction model based on a multi-level memory network to predict future load demand and power supply;

[0063] The preliminary instruction module is used to generate preliminary charge and discharge instructions based on the prediction results, combined with the real-time status of the energy storage unit and the grid information, and send them to the edge node;

[0064] The preliminary decision module is used to make preliminary decisions on preliminary charge and discharge instructions at the edge node, generate a local charge and discharge plan, and upload it to the cloud server;

[0065] The final instruction module is used for the cloud server to summarize the local charge and discharge plans, perform global optimization, generate the final charge and discharge instructions, and send them to the energy storage unit.

[0066] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the method for improving energy storage efficiency combined with energy management as described in the first aspect of the present invention is implemented.

[0067] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the method for improving energy storage efficiency combined with energy management as described in the first aspect of the present invention.

[0068] The beneficial effects of the present invention are as follows: the present invention uses sensor networks and Internet interfaces to collect energy storage unit status, environmental data, load demand, and power grid information in real time, ensuring the comprehensiveness and real-time nature of the data, laying the foundation for subsequent predictions. The prediction model constructed based on a multi-level memory network, combined with short-term, medium-term, and long-term sub-networks and an attention enhancement mechanism, accurately predicts future load demand and power supply, reduces prediction errors, and improves scheduling foresight. Edge nodes locally optimize preliminary charge and discharge instructions to generate plans adapted to local conditions, reduce the computing burden on the cloud, and improve real-time response capabilities. From a global perspective, the cloud server uses an ant algorithm to globally optimize local plans, and generates optimal charge and discharge instructions that maximize revenue and minimize grid fluctuations while satisfying multiple constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0070] Figure 1 This is a flow chart of the method for improving energy storage efficiency combined with energy management in Example 1.

[0071] Figure 2 This is a module diagram of the energy storage efficiency improvement system combined with energy management in Example 1. DETAILED DESCRIPTION

[0072] 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 with reference to the accompanying drawings.

[0073] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. 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.

[0074] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0075] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, provides a method for improving energy storage efficiency combined with energy management, comprising the following steps:

[0076] S1. Collect energy storage unit status data, environmental data, and load data through the sensor network, obtain power grid information through the Internet, and process it in real time.

[0077] Real-time monitoring and collection of energy storage unit status data, energy storage unit load demand, and energy storage unit environmental data;

[0078] Obtain real-time power grid information through the Internet interface;

[0079] The energy storage unit status data, load demand and environmental data, and grid information are sorted by the timestamp of the acquisition and integrated into a multi-source data set;

[0080] Specifically, status data includes the energy storage unit's input voltage, current, and state of charge, collected once per second; load data includes the energy storage unit's output voltage and current data, collected once per minute; environmental parameters include temperature and humidity, collected once per minute; and grid information includes real-time electricity prices and grid frequency, updated once an hour.

[0081] Filter and normalize multi-source data sets;

[0082] Use Kalman filter to filter multi-source data sets, remove noise and smooth data;

[0083] The multi-source datasets are normalized using the minimum-maximum method to ensure that the numerical ranges of different features are consistent.

[0084] S2. Build a prediction model based on a multi-level memory network to predict future load demand and power supply.

[0085] Decompose the processed multi-source data set into short-term, medium-term and long-term data subsets according to the time scale;

[0086] The short-term data subset includes status data collected every second, the medium-term data subset includes load data and environmental data collected every minute, and the long-term data subset includes power grid information collected every hour;

[0087] Based on the short-term data subset, a short-term memory subnetwork is constructed using standard LSTM units to capture the short-term fluctuation characteristics of the energy storage unit;

[0088] Based on the medium-term data subset, a medium-term memory sub-network is constructed using bidirectional LSTM units to capture the medium-term fluctuation characteristics of the energy storage unit;

[0089] Based on the long-term data subset, the LSTM unit with attention enhancement mechanism is used to build a long-term memory sub-network to capture the long-term fluctuation characteristics of the energy storage unit;

[0090] Specifically, the short-term memory subnetwork is constructed using a standard LSTM unit, with an input time step of 1 second and a window length of 60 seconds. The input data is a short-term dataset, and each time step contains the collected values ​​of short-term features (voltage, current, and state of charge of the energy storage unit);

[0091] The medium-term memory subnetwork is constructed using bidirectional LSTM (Bi-LSTM) units with a time step of 1 minute and a window length of 60 minutes. The input data is a medium-term dataset, and each time step contains the collected values ​​of medium-term features (voltage and current output by the energy storage unit, and temperature and humidity of the environment in which the energy storage unit is located).

[0092] The long-term memory subnetwork is constructed using LSTM units with an attention-enhanced mechanism, with a time step of 1 hour and a window length of 24 hours. The input data is the collected values ​​of long-term features (electricity price, grid frequency) at each time step of the long-term dataset.

[0093] By using the attention enhancement mechanism to dynamically weight the output vectors of the short-term, medium-term, and long-term sub-networks, the interactive relationship between different time scales is captured, and the fusion vector of the multi-level memory network is obtained. The expression is as follows:

[0094]

[0095] Among them, α i is the attention weight of the i-th sub-network, i is the index of the number of sub-networks, W a is the weight matrix of the attention enhancement mechanism, h i is the hidden vector of the i-th sub-network, b ais the bias term of the attention enhancement mechanism, N is the number of sub-networks; H is the fusion vector of the multi-level memory network;

[0096] Build a prediction model based on a multi-layer memory network;

[0097] The fusion vector H is received by the fully connected layer of the prediction model for training, and H is projected into the target output space. The physical constraint embedding mechanism is introduced into the prediction formula, and the expression is:

[0098] y=σ(W o H+b o )×clip(A,0,1);

[0099]

[0100] Among them, y is the predicted value, σ is the Sigmoid activation function, which means that the linear transformation result is normalized to the range of (0,1), and W o is the weight matrix of the fully connected layer, which is used to map the hidden state H from the feature space to the output space, b o is the bias vector of the fully connected layer, A is the physical constraint, and clip(A,0,1) means limiting the physical constraint A to [0,1];

[0101] Physical constraint A, when forecasting power supply, is the ratio of the supply power to the rated input power of the energy storage unit; when forecasting load demand, is the ratio of the load demand to the capacity of the energy storage unit;

[0102] During the prediction model training process, the gradient descent method is used, and Adam optimizes the prediction model parameters W. o and b o ;

[0103] According to the loss function, the model parameters W o and b o The partial derivative of is used to obtain the gradient value, and the model parameters are updated according to the dynamic learning rate adjustment rule of Adam.

[0104] The training process continues to iterate until the loss function change value is lower than the set threshold or the maximum number of training rounds is reached;

[0105] The threshold of the loss function change value is set to 10 based on the scale of the prediction model and the volume of the multi-source dataset. -3 , gradually reduced to balance training efficiency and accuracy;

[0106] After the training is completed, the model parameters are fixed and the test data set is used for prediction. The mean square error and mean absolute error between the predicted value and the actual value are calculated.

[0107] If the values ​​of mean square error and mean absolute error are small, it means that the model has high prediction accuracy and robustness;

[0108] If the mean square error is much higher than the mean absolute error, it may indicate that there are outliers in the data and the data processing or model structure needs to be optimized;

[0109] Through comprehensive evaluation of mean square error and mean absolute error, we verify whether the model meets the accuracy and stability requirements of practical applications, and provide a reliable prediction model for subsequent applications.

[0110] The trained prediction model receives the real-time multi-source dataset, sorts the real-time multi-source dataset by timestamp, and inputs it into the trained prediction model;

[0111] The prediction model outputs a fusion vector based on the dynamic weighted fusion of short-term, medium-term and long-term memory sub-networks and the attention enhancement mechanism;

[0112] The fusion vector is mapped to the target output space through the fully connected layer to predict the load demand value and power supply value in the future time period;

[0113] It should be noted that by constructing a multi-level memory network and building a prediction model, we can fully capture short-term fluctuations and long-term trends, significantly improve the prediction accuracy and adaptability, and the prediction output optimization model based on physical constraints realizes the integration of data-driven and domain knowledge by embedding physical rules in the prediction formula.

[0114] S3. Based on the prediction results, combined with the real-time status of the energy storage unit and grid information, preliminary charging and discharging instructions are generated and sent to the edge node.

[0115] The predicted load demand value and power supply value are obtained through the inverse normalization method;

[0116] Through sensor networks and Internet interfaces, real-time status data of energy storage units and grid information are obtained to assess the charging and discharging needs of energy storage units.

[0117] Specifically, for example, if the real-time power of the energy storage unit is between 50% and 80%, which meets the charging and discharging requirements, then charging and discharging operations can be performed. If the real-time power of the energy storage unit is greater than 80%, only discharging is required. If the real-time power of the energy storage unit is less than 50%, charging is required.

[0118] Combine the predicted load demand value and power supply value to generate preliminary charging and discharging instructions;

[0119] Charge and discharge instructions include charge and discharge power and time;

[0120] For example, if the load demand value is predicted to be high in the next few days, it is recommended to charge the energy storage unit in advance to meet the high power demand in the future. If the power supply value is predicted to be high in the next few days, it is recommended to discharge the energy storage unit in advance to meet the power supply demand.

[0121] The preliminary charge and discharge instructions are formatted into a standard format and transmitted to the edge nodes of each energy storage unit through a communication protocol (MQTT protocol).

[0122] S4. Make preliminary decisions on the preliminary charge and discharge instructions at the edge node, generate a local charge and discharge plan, and upload it to the cloud server.

[0123] The edge node receives preliminary charge and discharge instructions through the communication protocol (MQTT protocol);

[0124] Extract charging and discharging power and time by parsing preliminary charging and discharging instructions;

[0125] The edge node obtains the status data and environmental parameters of the energy storage unit in real time through the local sensor network to obtain a local data set;

[0126] Optimize preliminary charge and discharge instructions based on local data sets to generate local charge and discharge plans;

[0127] Specifically, if the energy storage unit status and environmental conditions reflected in the local data set cannot fully meet the requirements of the preliminary charge and discharge instructions, the preliminary charge and discharge instructions are optimized. For example, if the preliminary charge and discharge instructions are to charge at rated power, and the local data set shows that the energy storage unit is in a high temperature environment, the charging power is reduced, the charging time is extended, and the ventilation equipment power is increased;

[0128] The local charge and discharge plan is uploaded to the cloud server through the communication protocol (MQTT protocol).

[0129] S5. The cloud server summarizes the local charging and discharging plans, performs global optimization, generates the final charging and discharging instructions, and sends them to the energy storage unit.

[0130] The cloud server receives the local charge and discharge plans uploaded by all edge nodes and integrates them into a global charge and discharge plan;

[0131] The constraints are the safe range of the energy storage unit's state of charge, the rated power of charging and discharging, and the fluctuation range of the energy storage unit on the power grid.

[0132] If the optimized global charge and discharge plan exceeds any constraint, it will be invalidated;

[0133] Use Ant Colony Optimization (ACO) to optimize the global charging and discharging plan;

[0134] The state of the energy storage unit at a certain point in time is regarded as a node, and the node attributes include state of charge, charge and discharge power, and electricity price;

[0135] In the process of optimizing the global charge and discharge plan by the ant algorithm, the charge and discharge behavior of the energy storage unit between two time points is used as the path;

[0136] The objective function is defined based on maximizing the benefits of the energy storage unit and minimizing the grid frequency fluctuation;

[0137] Ants leave pheromones along their paths as virtual markers. During the initialization phase, the pheromone values ​​on all paths are set to the same initial value. The pheromone concentration represents the global attractiveness of the path. Paths with high pheromone concentrations correspond to better charging and discharging schemes (e.g., high returns and low volatility), while paths with low pheromone concentrations may correspond to unreasonable charging and discharging schemes (e.g., exceeding the state of charge limit or having low returns).

[0138] Based on the objective function, the heuristic factor is defined. The heuristic factor is the local attraction of the path, reflecting the priority of the path at the current node. The larger the heuristic factor, the greater the local attraction to the ants.

[0139]

[0140] Among them, ψ j,j+1 is the inspiration factor of path j→j+1, P t is the charge and discharge power of the energy storage unit, a positive number indicates charging, a negative number indicates discharging, and ∈ is a small positive number to avoid the denominator being 0;

[0141] Each ant starts from the first node and selects the next path based on the pheromone concentration and heuristic factor on the path until it traverses the entire time series;

[0142] Calculate the probability of an ant choosing a path. The expression is as follows:

[0143]

[0144] Among them, Q j,j+1 is the probability of an ant moving from node j to node j+1, φ j,j+1 is the pheromone concentration of path j→j+1, ψ j,j+1 is the heuristic factor of path j→j+1, γ is the coefficient of controlling the influence of pheromone on path selection, with a value range of [0.5, 2.0], and δ is the weight coefficient of controlling the influence of heuristic factor on path selection, with a value range of [1.0, 3.0];

[0145] Using pheromone updates, we can record the historical advantages and disadvantages of the paths and guide the subsequent ants’ path selection. The expression is as follows:

[0146]

[0147] in, is the pheromone concentration of the updated path j→j+1, θ is the pheromone volatility coefficient, ranging from [0.1, 0.9] to avoid over-reliance on historical paths, φ j,j+1 is the pheromone concentration of path j→j+1, Δφ j,j+1 is the pheromone increment of path j→j+1, and f is the objective function value;

[0148] When the next ant chooses j→j+1, When the next ant chooses path j→j+1, then Δφ ij =0;

[0149] The ant algorithm gradually optimizes through multiple rounds of iterations, gradually approaching the optimal global charge and discharge plan until the convergence conditions are met;

[0150] The maximum number of iterations is set according to the scale of the energy storage unit and the predicted time. After the iterative optimization is completed, the final charge and discharge instructions are output and sent to the edge node for execution on the energy storage unit.

[0151] This embodiment also provides an energy storage efficiency improvement system combined with energy management, including: a data acquisition module, a prediction module, a preliminary instruction module, a preliminary decision module and a final instruction module; the data acquisition module is used to collect energy storage unit status data, environmental data and load data through a sensor network, obtain power grid information through the Internet, and process it in real time; the prediction module is used to build a prediction model based on a multi-level memory network to predict future load demand and power supply; the preliminary instruction module is used to generate preliminary charge and discharge instructions based on the prediction results, combined with the real-time status of the energy storage unit and power grid information, and send them to the edge node; the preliminary decision module is used to make preliminary decisions on the preliminary charge and discharge instructions at the edge node, generate local charge and discharge plans, and upload them to the cloud server; the final instruction module is used for the cloud server to summarize the local charge and discharge plans, perform global optimization, generate final charge and discharge instructions, and send them to the energy storage unit.

[0152] This embodiment also provides a computer device, which is suitable for the case of combining the energy storage efficiency improvement method with energy management, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the energy storage efficiency improvement method combined with energy management proposed in the above embodiment.

[0153] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0154] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for improving energy storage efficiency combined with energy management as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0155] In summary, the present invention uses: sensor networks and Internet interfaces to collect energy storage unit status, environmental data, load demand and power grid information in real time, ensuring the comprehensiveness and real-time nature of the data, laying the foundation for subsequent predictions. The prediction model constructed based on a multi-level memory network, combined with short-term, medium-term and long-term sub-networks and attention enhancement mechanisms, accurately predicts future load demand and power supply, reduces prediction errors, and improves scheduling foresight. Edge nodes locally optimize preliminary charge and discharge instructions to generate plans adapted to local conditions, reduce the cloud computing burden, and improve real-time response capabilities. From a global perspective, the cloud server uses the ant algorithm to globally optimize local plans, and generates optimal charge and discharge instructions that maximize benefits and minimize grid fluctuations while satisfying multiple constraints.

[0156] Example 2, referring to Table 1, is the second example of the present invention. To further verify the technical solution of the present invention, experimental simulation data of the energy storage efficiency improvement method combined with energy management are provided.

[0157] To verify the superiority of a multi-layer memory network-based method for predicting energy storage unit load demand and power supply, as well as the effectiveness of the related charge and discharge instruction generation and optimization process, a set of comparative experiments was designed. The experiments included a performance comparison and analysis between the multi-layer memory network prediction system invented in this paper (hereinafter referred to as the "present invention") and an existing prediction system based on a single-layer LSTM model (hereinafter referred to as the "prior art").

[0158] The energy storage unit model is a 20kWh lithium-ion battery pack equipped with temperature and humidity sensors, current and voltage sensors, and edge nodes and cloud servers that support the MQTT communication protocol. Grid information is obtained through an API interface that connects to real-time electricity prices and grid frequency data for a specific region.

[0159] Through the sensor network, the status data (input voltage, current, state of charge), load data (output voltage, current) and environmental data (temperature, humidity) of the energy storage unit are collected; through the Internet interface, the grid data (real-time electricity price, grid frequency) is obtained, and data collection lasts for 72 hours.

[0160] Status data is collected once per second, load data and environmental data are collected once per minute, and grid data is updated once per hour.

[0161] The data were integrated into multi-source datasets by timestamp and processed by Kalman filtering for denoising, minimum-maximum normalization, and Z-score standardization.

[0162] Due to the large amount of data collected, the multi-source data sets for the first five time steps after processing are shown here, as shown in Table 1 below:

[0163] Table 1 Processed multi-source data table

[0164]

[0165] The state data is divided into short-term data subsets, the load data and environmental data are divided into medium-term data subsets, and the power grid data is divided into long-term data subsets.

[0166] The short-term, medium-term and long-term data subsets are used to construct the short-term memory sub-network (standard LSTM), medium-term memory sub-network (bidirectional LSTM) and long-term memory sub-network (attention-enhanced LSTM) respectively.

[0167] Obtain the hidden vectors output by the short-term memory subnetwork, the medium-term memory subnetwork, and the long-term memory subnetwork, and construct a multi-level memory network through a dynamic weighted fusion mechanism to output the fusion vector;

[0168] y=σ(W o H+b o )×clip(C,0,1);

[0169]

[0170] The fusion vectors of the data for the first five time steps are shown in Table 2 below:

[0171] Table 2 Fusion vector table of the first five time steps

[0172] Time step Fusion vector H 1 [0.4003,0.5003,0.4503] 2 [0.4050,0.5050,0.4550] 3 [0.3950,0.4950,0.4450] 4 [0.4100,0.5100,0.4600] 5 [0.4050,0.5050,0.4550]

[0173] Based on the prediction results, the predicted load demand value and power supply value are obtained through the inverse normalization method.

[0174] The load demand and power supply for the next 24 hours are predicted by the present invention and the prior art respectively, and the prediction errors (MSE and MAE) are recorded, as shown in Table 3 below:

[0175] Table 3 Experimental data comparison table

[0176]

[0177] The load demand forecast error of the present invention is 0.15, while the error of the existing technology is 0.42. Compared with the existing technology, the error of the present invention is reduced by approximately 64.3%, significantly improving the accuracy of load demand forecasting. This result shows that the multi-level memory network can more effectively capture the short-term fluctuations and medium- and long-term trends of energy storage units by dynamically weighting and fusing features at different time scales.

[0178] The power supply prediction error of the present invention is 0.12, while the prediction error of the existing technology is 0.38, a relative reduction of approximately 68.4%. The long-term memory subnetwork enhanced by the attention mechanism in the multi-layer memory network is significantly better than the single-layer LSTM model in capturing long-term trends in power grid information (such as electricity prices and grid frequency).

[0179] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. 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. A method for improving energy storage efficiency combined with energy management, characterized by: include, Collect energy storage unit status data, environmental data, and load data through the sensor network, obtain grid information through the Internet, and process it in real time; A prediction model is built based on a multi-level memory network to predict future load demand and power supply. The specific steps are as follows: Decompose the processed multi-source data set into short-term, medium-term and long-term data subsets according to the time scale; For the short-term data subset, a standard LSTM unit is used to construct a short-term memory subnetwork to capture the short-term fluctuation characteristics of the energy storage unit; For the medium-term data subset, a bidirectional LSTM unit is used to construct a medium-term memory sub-network to capture the medium-term fluctuation characteristics of the energy storage unit; For the long-term data subset, the LSTM unit with attention enhancement mechanism is used to construct a long-term memory sub-network to capture the long-term fluctuation characteristics of the energy storage unit; By using the attention enhancement mechanism to dynamically weight the output vectors of the short-term, medium-term, and long-term sub-networks, the interactive relationship between different time scales is captured, and the fusion vector of the multi-level memory network is obtained. The expression is as follows: Among them, α i is the attention weight of the i-th sub-network, i is the index of the number of sub-networks, W a is the weight matrix of the attention enhancement mechanism, h i is the hidden vector of the i-th sub-network, b a is the bias term of the attention enhancement mechanism, N is the number of sub-networks, and H is the fusion vector of the multi-level memory network; Build a prediction model based on a multi-layer memory network; The fusion vector H is received by the fully connected layer of the prediction model for training, and H is projected into the target output space. The physical constraint embedding mechanism is introduced into the prediction formula, and the expression is: y=σ(W o H+b o )×clip(A,0,1); Among them, y is the predicted value, σ is the Sigmoid activation function, W o is the weight matrix of the fully connected layer, b o is the bias vector of the fully connected layer, A is the physical constraint, and clip(A,0,1) means limiting the physical constraint A to [0,1]; During the prediction model training process, the gradient descent method is used to optimize the prediction model parameters W o and b o ; The prediction performance of the model was evaluated using mean square error and mean absolute error; The trained prediction model receives real-time multi-source data sets and predicts load demand and power supply; Based on the prediction results, combined with the real-time status of the energy storage unit and grid information, preliminary charging and discharging instructions are generated and sent to the edge node; Make preliminary decisions on the initial charge and discharge instructions at the edge node, generate a local charge and discharge plan, and upload it to the cloud server; The cloud server summarizes the local charging and discharging plans, performs global optimization, generates the final charging and discharging instructions, and sends them to the energy storage unit.

2. The method for improving energy storage efficiency combined with energy management according to claim 1, wherein: The sensor network is used to collect energy storage unit status data, environmental data and load data, and grid information is obtained through the Internet and processed in real time. The specific steps are as follows: Real-time monitoring and collection of energy storage unit status data, load demand, and environmental data; Obtain real-time power grid information through the Internet interface; The energy storage unit status data, load demand and environmental data, and grid information are sorted by the timestamp of the acquisition and integrated into a multi-source data set; Filter and normalize multi-source datasets.

3. The method for improving energy storage efficiency combined with energy management according to claim 2, wherein: According to the prediction results, combined with the real-time status of the energy storage unit and the grid information, preliminary charge and discharge instructions are generated and sent to the edge node. The specific steps are as follows: The predicted load demand value and power supply value are obtained through the inverse normalization method; Through sensor networks and Internet interfaces, real-time status data of energy storage units and grid information are obtained to assess the charging and discharging needs of energy storage units. Combine the predicted load demand value and power supply value to generate preliminary charging and discharging instructions; The preliminary charge and discharge instructions are formatted into a standard format and transmitted to the edge nodes of each energy storage unit through a communication protocol.

4. The method for improving energy storage efficiency combined with energy management according to claim 3, wherein: The edge node makes a preliminary decision on the preliminary charge and discharge instructions, generates a local charge and discharge plan, and uploads it to the cloud server. The specific steps are as follows: The edge node receives preliminary charge and discharge instructions through the communication protocol; Extract charging and discharging power and time by parsing preliminary charging and discharging instructions; The edge nodes obtain the status data and environmental parameters of the energy storage unit in real time through the sensor network to obtain a local data set; Optimize preliminary charge and discharge instructions based on local data sets to generate local charge and discharge plans; The local charge and discharge plan is uploaded to the cloud server through the communication protocol.

5. The method for improving energy storage efficiency combined with energy management according to claim 4, wherein: The cloud server summarizes the local charge and discharge plans, performs global optimization, generates the final charge and discharge instructions, and sends them to the energy storage unit. The specific steps are as follows: The cloud server receives the local charge and discharge plans uploaded by all edge nodes and integrates them into a global charge and discharge plan; The constraints are the safe range of the energy storage unit's state of charge, the rated power of charging and discharging, and the energy storage unit's requirements for the fluctuation range of the grid frequency. If the optimized global charge and discharge plan exceeds any constraint, it will be invalidated; The ant algorithm is used to optimize the global charging and discharging plan.

6. The method for improving energy storage efficiency combined with energy management according to claim 5, wherein: The ant algorithm is used to optimize the global charge and discharge plan. The specific steps are as follows: The objective function is defined based on maximizing the benefits of the energy storage unit and minimizing the grid frequency fluctuation; Ants leave pheromones on the path as virtual markers. In the initialization phase, the pheromone values ​​on all paths are set to the same initial value; Define heuristic factors based on the objective function; Each ant starts from the first node and selects the next path based on the pheromone concentration and heuristic factor on the path until it traverses the entire time series; Calculate the probability of an ant choosing a path. The expression is as follows: Among them, Q j,j+1 is the probability of an ant moving from node j to node j+1, φ j,j+1 is the pheromone concentration of path j→j+1, ψ j,j+1 is the heuristic factor of path j→j+1, γ is the coefficient of controlling the influence of pheromone on path selection, and δ is the weight coefficient of controlling the influence of heuristic factor on path selection; Using pheromone updates to record the historical advantages and disadvantages of paths and guide subsequent ants' path selection; The ant algorithm gradually optimizes through multiple rounds of iterations, gradually approaching the optimal global charge and discharge plan until the convergence conditions are met; The maximum number of iterations is set according to the scale of the energy storage unit and the predicted time. After the iterative optimization is completed, the final charge and discharge instructions are output and sent to the edge node for execution on the energy storage unit.

7. A system for improving energy storage efficiency combined with energy management, based on the method for improving energy storage efficiency combined with energy management according to any one of claims 1 to 6, characterized in that: Including data acquisition module, prediction module, preliminary instruction module, preliminary decision module and final instruction module, The data acquisition module is used to collect energy storage unit status data, environmental data and load data through the sensor network, obtain power grid information through the Internet, and process it in real time; The prediction module is used to build a prediction model based on a multi-level memory network to predict future load demand and power supply; The preliminary instruction module is used to generate preliminary charge and discharge instructions based on the prediction results, combined with the real-time status of the energy storage unit and the grid information, and send them to the edge node; The preliminary decision module is used to make preliminary decisions on preliminary charge and discharge instructions at the edge node, generate a local charge and discharge plan, and upload it to the cloud server; The final instruction module is used by the cloud server to summarize the local charge and discharge plans, perform global optimization, generate the final charge and discharge instructions, and send them to the energy storage unit.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for improving energy storage efficiency combined with energy management according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for improving energy storage efficiency combined with energy management according to any one of claims 1 to 6 are implemented.

Images