Energy management and safety protection collaborative method for liquid-cooled industrial and commercial energy storage system

Abstract

The application discloses a kind of energy management and safety protection synergic method of liquid-cooled commercial energy storage system, specifically relates to energy storage system management technical field, battery electrochemistry model, thermal distribution map, temperature gradient data and electrical parameter are as input, output battery temperature variation trend curve, and then according to the prediction result sets liquid cooling control strategy;Fusion temperature gradient abnormal parameters, temperature trend risk and multi-modal environmental data abnormal parameters, start fire risk assessment model, predict fire probability and position, calculate fire risk coefficient, generate fire risk report and set safety protection measures;Based on energy efficiency ratio, full life cycle income and battery health degree, construct multi-objective optimization mathematical model, combine energy storage operation data and power grid demand, generate Pareto optimal solution set using non-dominated sorting genetic algorithm, optimize charge-discharge strategy and liquid cooling parameter;Using reinforcement learning to optimize liquid cooling pipeline layout solves the problem that battery temperature cannot be effectively managed.

Description

Technical Field

[0001] This invention relates to the field of energy storage system management technology, and more specifically, to a collaborative method for energy management and safety protection of a liquid-cooled commercial and industrial energy storage system. Background Technology

[0002] Liquid cooling technology, as a highly efficient thermal management method, has been widely used in industrial and commercial energy storage systems. It absorbs and dissipates the heat generated during battery operation through circulating coolant, maintaining the battery within a safe temperature range. However, existing liquid cooling technologies still face many challenges in practical applications, especially in large-scale energy storage scenarios. The electrochemical reactions and thermal behavior of batteries exhibit high complexity and dynamism, leading to uneven temperature distribution and increased risk of thermal runaway. Traditional liquid cooling systems typically rely on static or simple control strategies, such as fixed coolant flow rates and preset temperature thresholds, which are ill-suited to the dynamic thermal management needs arising from battery load fluctuations or changes in environmental conditions. This lag and low adaptability result in insufficient cooling efficiency and reduced temperature control accuracy, not only lowering the energy efficiency of the energy storage system but also potentially creating safety hazards, such as localized overheating or even fire risks.

[0003] Furthermore, existing liquid cooling control systems have significant shortcomings in the synergy between energy management and safety protection. On the one hand, traditional methods lack the ability to predict battery temperature change trends, often relying solely on passive adjustments based on real-time data. This lack of proactive optimization of cooling strategies leads to uneven distribution of cooling resources or over-cooling, ultimately impacting the system's lifecycle benefits and battery health. On the other hand, current fire risk assessments largely rely on monitoring single indicators (such as exceeding temperature limits), failing to effectively integrate multimodal environmental data (such as vibration and gas concentration) for comprehensive analysis. This results in insufficient accuracy in fire prediction and inadequate targeting of protective measures. Under high loads or abnormal operating conditions, this singular approach to thermal management and risk control is insufficient to meet the requirements of modern industrial and commercial energy storage systems for efficient, safe, and intelligent operation. Summary of the Invention

[0004] To overcome the aforementioned deficiencies in the prior art, this invention provides a collaborative method for energy management and safety protection of a liquid-cooled commercial energy storage system, thereby addressing the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a collaborative method for energy management and safety protection of a liquid-cooled commercial and industrial energy storage system, comprising the following steps:

[0006] Simulates battery electrochemical reactions and thermal behavior to generate preset heat distribution maps and heat transfer parameters; real-time acquisition of battery temperature, voltage, current data and multimodal environmental data through sensor networks; noise is filtered and temperature gradient is calculated by edge computing devices to output real-time temperature distribution matrix and temperature gradient data.

[0007] A temperature trend prediction model is built using a long short-term memory network. The model takes a preset heat distribution map, heat transfer parameters, real-time temperature distribution matrix and electrical parameters as input, converts the spatial data into a time series format adapted by LSTM, and outputs a battery temperature change trend curve. Based on the battery temperature change trend curve, a liquid cooling control strategy is set; the coolant flow rate and temperature of the liquid cooling system are dynamically adjusted.

[0008] By integrating abnormal temperature gradient parameters, temperature trend risks, and abnormal parameters from multimodal environmental data, a fire risk assessment model is launched to predict the probability and location of fires, calculate fire risk coefficients, generate fire risk reports, and set up safety protection measures.

[0009] A multi-objective optimization mathematical model is constructed based on energy efficiency ratio, full life cycle benefits, and battery health. Combining energy storage operation data and grid demand, a non-dominated sorting genetic algorithm is used to generate the Pareto optimal solution set, optimize charging and discharging strategies and liquid cooling parameters, and introduce Bayesian optimization technology to dynamically adjust algorithm parameters. Based on the real-time temperature distribution matrix and temperature change trend curve, continuous temperature anomalies are determined (e.g., when the fire risk coefficient continuously exceeds the preset value, it is determined that the temperature is continuously abnormal). If the temperature is continuously abnormal, reinforcement learning is used to optimize the liquid cooling pipeline layout, iteratively adjusting the pipe diameter and flow distribution to ensure cooling uniformity and energy efficiency.

[0010] Preferably, the frequency of battery data acquisition is adaptively adjusted based on the battery temperature change trend curve.

[0011] Preferably, the fire risk assessment model is obtained as follows: a multimodal environmental dataset is constructed, including temperature data, vibration data, sound data, and gas concentration data; features are extracted based on the multimodal environmental information, including at least the rate of temperature change, temperature acceleration, and degree of gas anomaly; an initial fire risk assessment model is built based on a long short-term memory network; the fire risk assessment model is trained based on the training set; cross-validation is used to divide the historical data into training and test sets; the optimal hyperparameter combination is found through grid search or random search, and cross-validation and performance evaluation are performed to obtain the trained fire risk assessment model.

[0012] Preferably, the method further includes a fire risk management step, including:

[0013] Environmental data acquisition: The sensor network also collects multiple environmental data such as battery vibration, sound, and gas concentration (smoke sensor, gas concentration sensor), and fuses them to obtain multimodal environmental data;

[0014] Anomaly detection: A fire risk assessment model is triggered based on abnormal temperature gradient parameters, temperature trend risks, and abnormal parameters of multimodal environmental data. The temperature trend risk refers to the potential overheating risk of the battery temperature in the future based on the predicted temperature change trend. The abnormal parameters of multimodal environmental data refer to the degree to which vibration, sound, and gas concentration exceed the preset range.

[0015] Fire risk assessment: The fused multimodal environmental information is input into the fire risk assessment model to predict the probability and potential location of fire, calculate the fire risk coefficient, and the probability, potential location and fire risk coefficient constitute the fire risk report. Based on the fire risk report, safety protection measures are set; and intelligent responses (such as local isolation or fire extinguishing agent spraying) are automatically triggered.

[0016] Preferably, a dynamic optimization response strategy is adopted to ensure both resource utilization efficiency and safety. The target monitoring battery is divided into several areas, and maintenance priorities are determined based on fire risk coefficients. The fire risk coefficients are obtained as follows:

[0017] Obtain the fire probability P fire The probability of a fire occurring in the future time frame predicted by the LSTM model, with a value range of [0, 1].

[0018] Joint temperature trend risk T risk Multimodal environment data anomaly parameter H y The fire risk coefficient, C, is obtained from the fire probability using the following formula. risk =1-exp[-k·(P fire ·(1+T risk )·(1+H y ))], where k represents a constant controlling the overall risk sensitivity, ranging from [1, 10].

[0019] Preferably, the process also includes operation optimization management steps, including: constructing a multi-objective optimization mathematical model based on energy efficiency ratio, full life cycle benefits, and battery health; using the multi-objective optimization mathematical model, combined with energy storage system operation data and grid demand, to quantify operation characteristics by constructing the multi-objective optimization mathematical model, initializing a non-dominated sorting genetic algorithm population to generate a preliminary strategy, and generating a Pareto optimal solution set through evolutionary calculation to achieve multi-objective collaborative optimization, optimizing charging and discharging strategies and liquid cooling system parameters, and generating an economical, reliable, and environmentally friendly Pareto optimal solution set; and introducing Bayesian optimization technology to dynamically adjust algorithm parameters, improving optimization efficiency and global search capability.

[0020] Preferably, the operation optimization management steps include the following:

[0021] Step S31: Construct a multi-objective optimization mathematical model: Extract the grid demand and key parameters of the energy storage system from the operating data. The grid demand data includes at least load and frequency regulation demand, and the key parameters include energy efficiency ratio, life cycle revenue and battery health. Jointly establish a multi-objective optimization mathematical model; output the multi-objective optimization mathematical model and grid demand data.

[0022] Step S32: Initialize the non-dominated sorting genetic algorithm population: Based on the multi-objective optimization mathematical model and power grid demand data, the non-dominated sorting genetic algorithm is used to initialize the population, and the initial running strategy set is output.

[0023] Step S33: Generate Pareto optimal solution set through evolution: Based on the initial running strategy set, perform evolutionary calculations using a non-dominated sorting genetic algorithm to output the Pareto optimal solution set;

[0024] The process involves identifying the Pareto front layer, where the first layer represents the current optimal strategy. A tournament selection mechanism is used to select parent strategies from the population based on crowding distance and dominance level. Simulated binary crossover and polynomial mutation are employed to adjust decision variables, generating offspring strategies and expanding the diversity of the solution space. The sorting, selection, and mutation processes are iteratively executed until convergence conditions are met (e.g., the number of iterations reaches a preset value or the front layer stabilizes), generating a Pareto optimal solution set. Each solution set includes optimized charge / discharge power and liquid cooling parameters, corresponding to target values ​​for energy efficiency ratio, total lifecycle benefits, and battery health.

[0025] Step S34, Bayesian optimization adjustment: Based on the Pareto optimal solution set, Bayesian optimization technology is introduced to adjust the crossover probability and mutation index of the non-dominated sorting genetic algorithm, and the final running optimization control strategy is generated to improve the solution set performance and global optimization capability.

[0026] Preferably, it also includes the step of optimizing the liquid cooling pipeline layout:

[0027] If any of the following factors still exist under the optimized control strategy scenario: abnormal temperature gradient parameters, temperature trend risks, or abnormal parameters in multimodal environmental data, the step of optimizing the liquid cooling pipeline layout will be triggered.

[0028] Preferably, the process of optimizing the liquid cooling pipeline layout includes the following steps:

[0029] The battery temperature change trend curve and real-time temperature distribution matrix are obtained and combined with heat transfer parameters to input the heat flow simulation model. Reinforcement learning is used to iteratively adjust the pipe diameter, branches and flow distribution. The reward objectives for pipe optimization are set based on cooling uniformity, liquid cooling efficiency and maximum temperature. Optimized pipe design parameters are generated to ensure that cooling uniformity and energy efficiency are both taken into account.

[0030] Obtain pipeline design parameters, adjust liquid cooling pipeline layout, and verify battery temperature change trend curves and real-time temperature distribution matrix to ensure the reliability of pipeline optimization.

[0031] Preferably, the reward target is calculated as follows:

[0032] To obtain cooling uniformity, a higher value indicates better uniformity; to obtain cooling efficiency Q. remored With energy consumption E cooling The liquid cooling efficiency is calculated by comparing the two; the reward target is calculated using the following formula.

[0033]

[0034] Among them, T i This represents the temperature at a certain location in time and space, where w1, w2, and w3 represent the weighting coefficients for each item, and T. max The highest temperature, T safe As a safety threshold, T ref This is a reference temperature.

[0035] Preferably, the method further includes a fault location step:

[0036] If any of the following are present: abnormal temperature gradient parameters, temperature trend risks, or abnormal multimodal environmental data parameters, the fault location step will be triggered.

[0037] By utilizing time-series anomaly detection algorithms, operational data is monitored in real time, and abnormal behavior is automatically identified. Combined with causal inference technology, the causal relationship between anomalies and components (such as batteries and liquid cooling units) is analyzed to quickly locate the root cause. A fault knowledge graph is constructed to automatically classify anomaly patterns and recommend solutions based on historical data.

[0038] The system acquires operational data from the energy storage system, applies a recurrent neural network based on an attention mechanism to analyze time-series characteristics, generates abnormal behavior sequences, and identifies potential fault signals (such as sudden temperature increases or voltage drops).

[0039] Obtain abnormal behavior sequences, combine them with structural causal models to infer the causal relationship between component failures and abnormal behavior sequences, and generate a failure cause-effect graph to locate the root cause of the failure.

[0040] Obtain the cause-effect graph of the fault, combine it with historical data using a graph neural network, generate a fault diagnosis report to classify anomalies and recommend solutions.

[0041] Preferably, the abnormal behavior sequence is identified based on an abnormal behavior detection threshold function, wherein the calculation formula for the abnormal behavior detection threshold function is:

[0042]

[0043] Where, x t Current data value, x pred x represents the predicted value. ref Reference value, where λ represents the sensitivity factor.

[0044] The technical effects and advantages of this invention are as follows:

[0045] (1) The energy management and safety protection collaborative method for liquid-cooled commercial and industrial energy storage systems provided by this invention combines battery electrochemical models, sensor networks, long short-term memory networks, fire risk assessment, and operation optimization management technologies to provide a comprehensive safety protection and energy efficiency optimization strategy. By monitoring battery temperature, current, voltage, and other data in real time, the method predicts temperature change trends and dynamically adjusts the control strategy of the liquid-cooled system to ensure that the energy storage system operates efficiently while avoiding safety risks such as overheating. In addition, the fire risk assessment model combines multimodal environmental data such as temperature, vibration, and gas concentration to trigger intelligent response measures in a timely manner, improving the system's safety and effectively solving the shortcomings of traditional energy storage systems in energy efficiency management and safety protection, thereby improving the system's operating efficiency and safety assurance capabilities.

[0046] (2) The energy management and safety protection collaborative method of liquid-cooled industrial and commercial energy storage system provided by the present invention constructs a multi-objective optimization mathematical model based on energy efficiency ratio, full life cycle benefits and battery health, combines energy storage operation data and grid demand, and uses a non-dominated sorting genetic algorithm to generate Pareto optimal solution set, effectively optimizing charging and discharging strategies and liquid cooling parameters. Attached Figure Description

[0047] Figure 1 This is a flowchart of the temperature control process based on the temperature change trend curve of the present invention.

[0048] Figure 2 This is a flowchart of the temperature control process based on operational strategy optimization according to the present invention. Detailed Implementation

[0049] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0050] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0051] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the scope of this application and its application or use.

[0052] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0053] Example 1, see Figure 1 The present invention provides a temperature control flowchart based on a temperature change trend curve, as shown in the following figure. Figure 1 The method for coordinated energy management and safety protection of a liquid-cooled commercial and industrial energy storage system, as shown, includes the following steps:

[0054] Step S11: Use finite element analysis software combined with the battery electrochemical model to simulate the electrochemical reaction and thermal behavior of the battery under different operating conditions, analyze the temperature distribution and heat transfer characteristics, and generate a preset heat distribution map and heat transfer parameters.

[0055] The explanation is as follows: The heat distribution map shows the temperature distribution of the battery under different operating conditions. By comparing the heat distribution map with the actual measured battery temperature data, the accuracy of the thermal coupling model is calibrated. Heat transfer parameters, such as thermal resistance and thermal capacity, are extracted from the heat distribution map. These heat transfer parameters reflect the heat transfer characteristics of the battery.

[0056] In this embodiment of the invention, finite element analysis software (such as COMSOL Multiphysics) is used to construct a three-dimensional battery model with a mesh generation accuracy of 0.5-1 mm. The battery electrochemical model simulates lithium-ion diffusion and charge transfer based on the Butler-Volmer equation. Input parameters include electrode material properties (such as lithium diffusion coefficient) and operating conditions (such as charging and discharging). The simulation generates a heat distribution map (two-dimensional temperature field, unit: °C) and heat transfer parameters. The heat distribution map is used to calibrate the model by comparing it with actual temperature data to control errors (such as controlling the error within ±2 °C).

[0057] Step S12: Collect the battery's temperature, voltage, and current data in real time through a sensor network, filter noise and calculate the temperature gradient through an edge computing device, and output the real-time temperature distribution matrix and temperature gradient data.

[0058] The temperature gradient data reflects the uniformity of the battery temperature distribution. By analyzing the temperature gradient data, potential local overheating areas can be identified (such as a part where the temperature is significantly higher than other areas), so that timely measures can be taken to avoid the risk of thermal runaway.

[0059] In this embodiment of the invention, the sensor network is deployed with thermistors (accuracy ±0.1℃), voltage sensors (range 0-5V), and current sensors (range 0-200A), covering the positive electrode, negative electrode, and central area of ​​the battery, with a sampling frequency of 1Hz; the edge computing device (such as a Raspberry Pi) uses Kalman filtering to remove noise, calculates the temperature gradient (unit: ℃ / m, formula is the temperature difference between adjacent points divided by the distance); generates a real-time temperature distribution matrix (two-dimensional array, each element is a temperature value), temperature gradient data (vector form), and electrical parameters (voltage and current sequences), and the temperature gradient data is used to identify local overheating (such as gradient > 5℃ / m);

[0060] Step S13: Use a long short-term memory network to build a temperature trend prediction model. Input a preset heat distribution map, heat transfer parameters, real-time temperature distribution matrix and electrical parameters. Convert the spatial data into a time series format adapted by LSTM. Predict the battery temperature change trend curve in the future (e.g., 5-30 minutes). Set liquid cooling control strategy (e.g., time-series control commands for coolant flow rate and temperature setpoint) based on the battery temperature change trend curve.

[0061] In this embodiment of the invention, the long short-term memory network model is recommended to adopt a 3-layer convolutional structure (3×3 convolutional kernels), with the input dimensions being a heat distribution map (two-dimensional), a real-time temperature distribution matrix (two-dimensional), temperature gradient data (one-dimensional vector), and heat transfer parameters (scalar); measures are taken according to the battery temperature change trend curve, such as setting a liquid cooling control strategy if the temperature rise rate is >0.5℃ / min;

[0062] In one possible embodiment, the data acquisition frequency is adaptively adjusted based on the battery temperature change trend curve;

[0063] Step S14: Dynamically adjust the coolant flow rate and temperature of the liquid cooling system according to the liquid cooling control strategy, and monitor the battery temperature in real time. If the battery temperature exceeds the expectation, issue an early warning and provide feedback data to optimize the long short-term memory network model and achieve continuous improvement.

[0064] In this embodiment of the invention, control parameters are received through a programmable logic controller to adjust the liquid cooling pump flow rate and coolant temperature; the temperature sensor used in step S12 is monitored in real time to record the temperature change within 5-30 minutes after adjustment. If the measured temperature exceeds the predicted value by ±2℃, it is considered to exceed expectations, and the real-time temperature data is fed back to step S13 to update the CNN model weights (it is recommended to use the Adam optimizer with a learning rate of 0.001); the adjustment effect is evaluated through temperature stability, and the optimized model parameters are output to improve long-term thermal management and safety protection capabilities.

[0065] In this embodiment of the invention, it should be further explained that the method also includes a fire risk management step, including:

[0066] Environmental data acquisition: The sensor network also collects multiple environmental data such as battery vibration, sound, and gas concentration (smoke sensor, gas concentration sensor), and fuses them to obtain multimodal environmental data;

[0067] To explain, multimodal environmental data (temperature, vibration, gas concentration) are fused at the feature level: the temporal features of each mode (such as temperature change rate, vibration spectrum energy, and gas concentration gradient) are extracted and weighted by an attention mechanism, with the weight coefficients obtained from training on historical data;

[0068] Anomaly detection: A fire risk assessment model is triggered based on abnormal temperature gradient parameters, temperature trend risks, and abnormal parameters of multimodal environmental data. The temperature trend risk refers to the potential overheating risk of the battery temperature in the future based on the predicted temperature change trend. The abnormal parameters of multimodal environmental data refer to the degree to which vibration, sound, and gas concentration exceed the preset range.

[0069] In one possible embodiment, the fire risk assessment model is triggered when the combined value of the multimodal anomaly parameters exceeds a threshold, or when a single parameter exceeds a safety limit; for example, when the temperature gradient anomaly parameter Wy > 5℃ / m, or the gas concentration anomaly exceeds a preset threshold, the fire risk assessment is triggered immediately.

[0070] Furthermore, the present invention does not impose specific restrictions on the triggering conditions of the fire risk assessment model, which can be any one or a combination of temperature gradient anomaly parameters, temperature trend risks, and multimodal environmental data.

[0071] For ease of understanding, this embodiment is based on a fire risk assessment model triggered by temperature trend risk: the temperature trend risk T risk This refers to the cumulative risk from the current time t to a future time period t+Δt, calculated using the following formula:

[0072]

[0073] in, This indicates the rate of temperature change over time; The acceleration represents the temperature change; β1 is an adjustable parameter used to amplify or reduce the effect of acceleration; σ pred β1 represents the confidence interval for temperature prediction, reflecting the reliability of future trends; β2 represents the uncertainty weight, used to adjust the contribution of prediction error to risk.

[0074] When T risk If the threshold is exceeded, execute fire risk management procedures;

[0075] In one possible embodiment, a fire risk assessment model is triggered based on anomaly parameters in multimodal environmental data, using H... y This parameter represents anomalies in multimodal environmental data, ranging from 0 to 1. The closer the value is to 1, the further the multimodal environmental data deviates from the normal range.

[0076] The quantification formula for the anomaly parameters of the multimodal environmental data is as follows:

[0077] Through formula Process each type of environmental data (vibration, sound, gas concentration) and calculate its anomaly degree; x i This represents the current sensor measurement value (such as vibration acceleration, sound decibels, gas concentration), in μ. i σ represents the mean of historical normal data. i The standard deviation of historical normal data;

[0078] The outlier of each type of environmental data is compressed to 0 to 1 using the Sigmoid function;

[0079] Based on the contribution of each type of environmental data to fire risk, a weight is assigned, and the weighted sum of the normalized anomaly degree and the weight is calculated to obtain the multimodal environmental data anomaly parameter Hy.

[0080] Fire risk assessment: The fused multimodal environmental information is input into the fire risk assessment model to predict the probability and potential location of fire, calculate the fire risk coefficient, and the probability, potential location and fire risk coefficient constitute the fire risk report. Based on the fire risk report, safety protection measures are set; and intelligent responses (such as local isolation or fire extinguishing agent spraying) are automatically triggered.

[0081] For ease of understanding, the quantification threshold for the temperature gradient anomaly parameter and temperature trend risk is defined as follows:

[0082] Based on historical normal operation data, the 99th percentile of the temperature gradient abrupt change amplitude (the rate of change of temperature difference between adjacent monitoring points) and the temperature trend risk slope (the rate of temperature rise per unit time of the LSTM prediction curve) were calculated as the initial anomaly thresholds (i.e., fluctuations exceeding 99% of the normal range were considered abnormal). A joint probability model of multimodal environmental parameters (such as CO concentration and humidity extremes) was constructed using a Bayesian network: by statistically analyzing the distribution of environmental parameters before the fire event, the conditional probability of fire occurrence under each parameter combination was calculated, and P ≥ 0.75 was set as the high-risk threshold. The threshold was further optimized through backtesting: using historical fire events as positive samples, the quantiles (e.g., 98% → 99.5%) and probability thresholds (e.g., 0.75 → 0.8) were adjusted to achieve a recall rate > 90% and a false alarm rate < 5%.

[0083] By dynamically weighting key parameters (such as temperature rise rate weight > CO concentration) through feature importance analysis (such as SHAP value) and embedding a lightweight Bayesian network into edge devices, dynamic threshold calibration and hardware-linked execution are achieved.

[0084] The formula for calculating the temperature gradient anomaly parameter is as follows: Calculate the temperature difference between adjacent sensors and divide it by the distance between them to obtain the local temperature gradient; average the sum of the squares of all local temperature gradients, and then take the square root of the result to obtain the temperature gradient anomaly parameter W. y ;

[0085] The calculation method for the anomaly parameters of the multimodal environmental data is as follows: extract key features, such as vibration frequency, sound intensity, and gas concentration, from the denoised multimodal environmental data; set a normal range threshold for each key feature based on historical data and safety standards; calculate the anomaly degree of each key feature; sum the anomaly degrees of each key feature by weight and perform normalization to obtain the anomaly parameters of the multimodal environmental data.

[0086] In this embodiment of the invention, it is necessary to further explain that the dynamic optimization response strategy ensures both resource utilization efficiency and safety. The target monitoring battery is divided into several areas, and maintenance priorities are determined based on a fire risk coefficient. The fire risk coefficient is obtained as follows:

[0087] Obtain the fire probability P fire The probability of a fire occurring in the future time frame predicted by the LSTM model, with a value range of [0, 1].

[0088] Joint temperature trend risk T risk Multimodal environment data anomaly parameter H y The fire risk coefficient, C, is obtained from the fire probability using the following formula. risk =1-exp[-k·(p fire ·(1+T risk )·(1+Hy ))], where k represents a constant controlling the overall risk sensitivity, ranging from [1, 10].

[0089] To explain, in this invention, the LSTM model is used to predict the probability of a fire occurring in the future, and its data comes from historical fire event data and real-time multimodal environmental data.

[0090] Historical data is extracted from the energy storage system's operation logs and accident reports, including parameters such as temperature, vibration, sound, and gas concentration at the time of the fire.

[0091] Real-time data is collected through a sensor network, including multimodal environmental information monitored by temperature sensors, vibration sensors, sound sensors, and gas concentration sensors;

[0092] The collected raw data underwent denoising, normalization, and feature extraction preprocessing, and was organized into a time series format.

[0093] The input at each time step is a vector containing multi-dimensional features such as temperature, temperature gradient, vibration frequency, sound intensity, and gas concentration. The input data is provided to the LSTM model in the form of a three-dimensional tensor ([batch size, sequence length, number of features]).

[0094] The LSTM model is trained based on historical data to learn the relationship between time series and fire events and predict the probability of fire. During operation, real-time data is input into the LSTM model after the same preprocessing and the model outputs the probability of fire occurrence.

[0095] In one possible implementation, the LSTM model time step is set to 1 minute, and the edge computing device processes the historical data of the most recent 30 minutes to form the input tensor; a 2-layer stacked LSTM structure is adopted, with each layer containing 128 hidden units, and the connection layer is combined with the sigmoid activation function to output the probability of fire occurrence.

[0096] In one possible embodiment, the fire risk assessment model is deployed on edge devices to achieve millisecond-level real-time early warning; the cloud is responsible for the continuous training and updating of the model, utilizing large-scale historical data to improve model accuracy, while adapting to the operating characteristics of different energy storage systems; based on the risk assessment results (fire risk coefficient), differentiated fire-fighting strategies are automatically executed, for example:

[0097] Low risk: Localized enhanced cooling;

[0098] Medium risk: Isolate the faulty area;

[0099] High risk: Triggering the spraying of extinguishing agents or emergency cooling.

[0100] In this embodiment of the invention, it is necessary to further explain that the method also includes an operation optimization management step, including: constructing a multi-objective optimization mathematical model based on energy efficiency ratio, full life cycle benefits, and battery health; using the multi-objective optimization mathematical model, combined with energy storage system operation data and grid demand, to quantify operation characteristics by constructing the multi-objective optimization mathematical model, initializing a non-dominated sorting genetic algorithm population to generate a preliminary strategy, and generating a Pareto optimal solution set through evolutionary calculation to achieve multi-objective collaborative optimization, optimizing charging and discharging strategies and liquid cooling system parameters, and generating an economical, reliable, and environmentally friendly Pareto optimal solution set; and introducing Bayesian optimization technology to dynamically adjust algorithm parameters, thereby improving optimization efficiency and global search capability.

[0101] Further explanation is needed in the embodiments of the present invention, see the following: Figure 2 The temperature control process flow chart based on operational strategy optimization includes the following operational optimization management steps:

[0102] Step S31: Construct a multi-objective optimization mathematical model: Extract the grid demand and key parameters of the energy storage system from the operating data. The grid demand data includes at least load and frequency regulation demand, and the key parameters include energy efficiency ratio, life cycle revenue and battery health. Jointly establish a multi-objective optimization mathematical model; output the multi-objective optimization mathematical model and grid demand data.

[0103] The explanation is as follows: the energy efficiency ratio refers to the ratio of heat removed by the liquid cooling system to the input energy; the total life cycle benefit refers to the benefit and cost balance calculated based on the exponential decay discount; and the battery health refers to the life assessment based on the capacity decay rate.

[0104] Step S32: Initialize the non-dominated sorting genetic algorithm population: Based on the multi-objective optimization mathematical model and power grid demand data, the non-dominated sorting genetic algorithm is used to initialize the population, and the initial running strategy set is output.

[0105] The explanation is as follows: multiple sets of decision variable combinations are randomly generated using a non-dominated sorting genetic algorithm to construct an initial population, with each combination corresponding to a potential operating strategy; the decision variables include: charging and discharging power, liquid cooling parameters (coolant flow rate, temperature setpoint), and grid demand;

[0106] Physical constraints (such as power range and flow range) are set to ensure feasibility; based on a multi-objective optimization mathematical model, the target values ​​for each individual in the population are calculated, including energy efficiency ratio, total life cycle benefit and battery health, to form an initial set of operating strategies;

[0107] Step S33: Generate Pareto optimal solution set through evolution: Based on the initial running strategy set, perform evolutionary calculations using a non-dominated sorting genetic algorithm to output the Pareto optimal solution set;

[0108] The process involves identifying the Pareto front layer, where the first layer represents the current optimal strategy. A tournament selection mechanism is used to select parent strategies from the population based on crowding distance and dominance level. Simulated binary crossover and polynomial mutation are employed to adjust decision variables, generating offspring strategies and expanding the diversity of the solution space. The sorting, selection, and mutation processes are iteratively executed until convergence conditions are met (e.g., the number of iterations reaches a preset value or the front layer stabilizes), generating a Pareto optimal solution set. Each solution set includes optimized charge / discharge power and liquid cooling parameters, corresponding to target values ​​for energy efficiency ratio, total lifecycle benefits, and battery health.

[0109] Step S34, Bayesian optimization adjustment: Based on the Pareto optimal solution set, Bayesian optimization technology is introduced to adjust the crossover probability and mutation index of the non-dominated sorting genetic algorithm, and the final running optimization control strategy is generated to improve the solution set performance and global optimization capability.

[0110] The process involves identifying the hypervolume index (measuring diversity and convergence) of the Pareto optimal solution set, constructing a Gaussian process model, and predicting the effect of parameter adjustment based on historical crossover probabilities, mutation index combinations, and hypervolume index values. The next set of optimal parameters is determined from the prediction results using the expectation improvement function, and applied to a non-dominated sorting genetic algorithm to re-evolve and update the Pareto optimal solution set. This adjustment process is iterated until the increase in the hypervolume index is less than a preset threshold, confirming optimization convergence. Finally, the Pareto optimal solution set is generated and used as the operational optimization control strategy to guide the system's charging / discharging and liquid cooling parameter configuration.

[0111] In one possible implementation, when simultaneously optimizing energy efficiency ratio (short-term indicator), total lifecycle benefit (long-term indicator), and battery health (reliability indicator), the multi-objective optimization mathematical model may contain contradictions (e.g., high energy efficiency may accelerate battery aging). This problem is addressed by balancing these multi-objective conflicts through dynamic weight allocation and hierarchical constraint mechanisms. Specifically, to address the conflict between energy efficiency ratio (short-term), total lifecycle benefit (long-term), and battery health (reliability), a dynamic weight allocation and hierarchical constraint mechanism is introduced within the existing NSGA-II and Bayesian optimization framework to ensure balance among the objectives and algorithm convergence.

[0112] Dynamic weight allocation: Based on the grid demand scenario (such as peak / off-peak load) and battery aging status, the weights of energy efficiency ratio, revenue, and health in the objective function are automatically adjusted; for example, the weight of energy efficiency ratio is increased to 50% during peak load and the weight of revenue over the entire life cycle is increased to 20% during off-peak load; if the battery health declines rapidly due to aging, the weight of energy efficiency is automatically reduced to prioritize the protection of health and avoid overcharging and discharging to reduce lifespan.

[0113] Layered constraint mechanism: Battery health is treated as a hard constraint (non-negotiable indicator), while energy efficiency ratio and revenue are treated as soft targets (optimizable indicators) to avoid excessive sacrifice of battery life; a minimum safety threshold for battery health is set (e.g., capacity decay does not exceed 20%), and all optimization strategies must meet this condition to ensure that the reliability bottom line cannot be breached; under the premise that the health standard is met, energy efficiency and revenue can be flexibly balanced, for example, by adjusting cooling parameters or charging and discharging power to achieve a balance between short-term energy efficiency and long-term revenue;

[0114] Bayesian optimization extension: The weight parameters and constraint thresholds are incorporated into the hyperparameter space of Bayesian optimization to achieve adaptive adjustment of the global policy; when generating Pareto optimal solutions, the solution set that meets the health criteria is retained first to ensure that the final policy takes into account both safety and economy.

[0115] The explanation is that dynamic weights respond to changes in the external scene, hierarchical constraints safeguard the bottom line of safety, and Bayesian optimization coordinates the relationship between the three in a global manner, forming a closed loop of "scene perception → target balancing → policy generation → parameter iteration", which avoids algorithm divergence and meets actual needs.

[0116] In this embodiment of the invention, it should be further explained that the method also includes a fault location step:

[0117] If any of the following are present: abnormal temperature gradient parameters, temperature trend risks, or abnormal multimodal environmental data parameters, the fault location step will be triggered.

[0118] By utilizing time-series anomaly detection algorithms, operational data is monitored in real time, and abnormal behavior is automatically identified. Combined with causal inference technology, the causal relationship between anomalies and components (such as batteries and liquid cooling units) is analyzed to quickly locate the root cause. A fault knowledge graph is constructed to automatically classify anomaly patterns and recommend solutions based on historical data.

[0119] This explanation explains how to use an attention-based RNN model. The input consists of time-series data of temperature, voltage, and current. The output is an anomaly score. When the anomaly score exceeds a preset value, it is marked as abnormal behavior. An example of the anomaly score formula is provided below.

[0120]

[0121] in, σ represents the predicted values ​​of temperature, voltage, and current, λ represents the standard deviation of historical data, and λ represents the sensitivity factor.

[0122] Causal inference: Using causal inference techniques, define nodes in the causal graph (such as "liquid cooling pump failure" or "battery aging"), and analyze the root causes through gradient backpropagation.

[0123] Knowledge graph update: The fault knowledge graph is dynamically updated using a graph neural network. Rule extraction is based on the Apriori algorithm, with a minimum support of 0.1 and a confidence threshold of 0.8.

[0124] The system acquires operational data from the energy storage system, applies a recurrent neural network based on an attention mechanism to analyze time-series characteristics, generates abnormal behavior sequences, and identifies potential fault signals (such as sudden temperature increases or voltage drops).

[0125] Obtain abnormal behavior sequences, combine them with a structural causal model to infer the causal relationship between component failures and abnormal behavior sequences, and generate a failure cause-effect graph to locate the root cause of the failure; including:

[0126] Starting from abnormal behavior sequences and historical operational data, a fault knowledge graph is constructed using a graph neural network. Nodes in the graph represent system components (such as battery modules and liquid cooling units) and abnormality types (such as overheating and short circuits), while edges represent causal relationships (e.g., "battery aging → overheating"). Graph embedding technology is used to map the graph structure to a low-dimensional vector space (e.g., 128-dimensional), and edge weights (ranging from 0 to 1) are trained based on historical data to reflect the correlation strength between various faults. By iteratively optimizing the graph embedding loss function and updating the node vectors, an optimized fault knowledge graph is generated, making similar abnormal patterns closer together and different patterns more distinct.

[0127] Obtain the cause-effect graph of the fault, combine it with historical data using a graph neural network, generate a fault diagnosis report to classify anomalies and recommend solutions.

[0128] The initial fault knowledge graph is constructed based on historical fault cases, and the nodes include fault types (such as overheating and short circuit) and components (such as battery modules and liquid cooling pumps).

[0129] The fault knowledge graph is updated through online learning: when a new fault occurs, features are extracted and matched with existing nodes. If no match is found, a new node is added, and the edge weights are optimized through gradient descent. Example: "If the frequency of 'abnormal liquid cooling pump flow' and 'local overheating' occurring simultaneously exceeds the threshold, a new edge 'liquid cooling pump fault → uneven cooling → abnormal temperature' is added with an initial weight of 0.9."

[0130] In one possible embodiment, an abnormal behavior sequence is identified based on an abnormal behavior detection threshold function, wherein the calculation formula for the abnormal behavior detection threshold function is:

[0131]

[0132] Where, x t Current data value, x pred x represents the predicted value. ref Reference value, where λ represents the sensitivity factor.

[0133] In summary, this invention combines battery electrochemical models, sensor networks, long short-term memory networks, fire risk assessment, and operational optimization management technologies to provide a comprehensive safety protection and energy efficiency optimization strategy. By monitoring battery temperature, current, and voltage data in real time, it predicts temperature change trends and dynamically adjusts the control strategy of the liquid cooling system, ensuring that the energy storage system operates efficiently while avoiding safety risks such as overheating. Furthermore, the fire risk assessment model incorporates multimodal environmental data such as temperature, vibration, and gas concentration to trigger intelligent response measures in a timely manner, improving system safety. This invention effectively addresses the shortcomings of traditional energy storage systems in energy efficiency management and safety protection, enhancing system operating efficiency and safety assurance capabilities.

[0134] Example 2 differs from Example 1 in that it further includes a step of optimizing the liquid cooling pipeline layout.

[0135] If any of the following factors still exist under the optimized control strategy scenario: abnormal temperature gradient parameters, temperature trend risks, or abnormal parameters in multimodal environmental data, the step of optimizing the liquid cooling pipeline layout will be triggered.

[0136] In this embodiment of the invention, it should be further explained that the process of optimizing the liquid cooling pipeline layout includes the following steps:

[0137] The battery temperature change trend curve and real-time temperature distribution matrix are obtained and combined with heat transfer parameters to input the heat flow simulation model. Reinforcement learning is used to iteratively adjust the pipe diameter, branches and flow distribution. The reward objectives for pipe optimization are set based on cooling uniformity, liquid cooling efficiency and maximum temperature. Optimized pipe design parameters are generated to ensure that cooling uniformity and energy efficiency are both taken into account.

[0138] Obtain pipeline design parameters, adjust liquid cooling pipeline layout, and verify battery temperature change trend curves and real-time temperature distribution matrix to ensure the reliability of pipeline optimization.

[0139] The heat flow simulation model is a digital twin model used to simulate the heat transfer and distribution characteristics of battery modules in liquid-cooled commercial and industrial energy storage systems. By dividing the battery modules into finite element meshes, the temperature change of each element is calculated based on the heat conduction equation to simulate the impact of liquid cooling pipeline adjustments (such as pipe diameter and flow rate) on heat flow. The model takes into account the distribution map of uneven cooling areas and heat transfer parameters, and outputs optimized pipeline design parameters and corresponding temperature distribution predictions.

[0140] In this embodiment of the invention, it needs to be further explained that the calculation method of the reward target is as follows:

[0141] To obtain cooling uniformity, a higher value indicates better uniformity; to obtain cooling efficiency Q. remored With energy consumption E coolingThe liquid cooling efficiency is calculated by comparing the two; the reward target is then calculated using the following formula:

[0142]

[0143] Among them, T i This represents the temperature at a certain location in time and space, where w1, w2, and w3 represent the weighting coefficients for each item, and T. max The highest temperature, T safe As a safety threshold, T ref This is a reference temperature.

[0144] In summary, by adding an optimization step to the liquid cooling pipeline layout based on Example 1, the thermal management efficiency of the liquid cooling system is further improved. Specifically, when abnormal temperature gradient parameters, temperature trend risks, or abnormal parameters in multimodal environmental data are detected, liquid cooling pipeline optimization is automatically triggered. A reinforcement learning algorithm is used to iteratively adjust the pipe diameter, branches, and flow distribution to ensure a balance between cooling uniformity and liquid cooling energy efficiency. The heat transfer and distribution characteristics in the liquid cooling system are simulated through a thermal flow simulation model. The optimized pipeline design not only improves the cooling uniformity of the liquid cooling system but also reduces energy consumption, effectively solving the problems of uneven heat distribution and low energy efficiency caused by unreasonable liquid cooling pipeline layout, and improving the safety and operating efficiency of the energy storage system.

[0145] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A collaborative method for energy management and safety protection of a liquid-cooled commercial and industrial energy storage system, characterized in that, include: Simulate battery electrochemical reactions and thermal behavior to generate preset heat distribution maps and heat transfer parameters; The battery's temperature, voltage, current data, and multimodal environmental data are collected in real time through a sensor network. Noise is filtered and temperature gradients are calculated by an edge computing device, and the real-time temperature distribution matrix and temperature gradient data are output. A temperature trend prediction model is built using a long short-term memory network. The model takes a preset heat distribution map, heat transfer parameters, real-time temperature distribution matrix and electrical parameters as input, and outputs a battery temperature change trend curve. Based on the battery temperature change trend curve, a liquid cooling control strategy is set; the coolant flow rate and temperature of the liquid cooling system are dynamically adjusted. The fire risk assessment model integrates abnormal temperature gradient parameters, temperature trend risks, and abnormal parameters from multimodal environmental data to calculate fire risk coefficients, generate fire risk reports, and trigger safety protection measures. The triggering conditions for the fire risk assessment model include at least one of the following: abnormal temperature gradient parameters exceeding a preset threshold; temperature trend risk results exceeding a dynamically adjusted threshold; and the temperature trend risk... This refers to a time period from the current time t to a certain future time. The accumulation of risk is calculated using the following formula: in, This indicates the rate of temperature change over time; The acceleration representing the temperature change. These are adjustment parameters used to amplify or reduce the effects of acceleration; This indicates the confidence interval for temperature prediction, reflecting the reliability of future trends; This represents the uncertainty weight, used to adjust the contribution of prediction error to risk; when If the threshold is exceeded, execute fire risk management procedures; Anomaly parameters in multimodal environmental data, obtained by weighted fusion of vibration, sound, and gas concentration anomalies, exceed a threshold. The target monitoring battery is divided into several areas, and the maintenance priority is determined based on the fire risk coefficient. The fire risk coefficient is obtained as follows: Obtain the probability of fire The probability of a fire occurring in the future time frame predicted by the LSTM model, with a value range of [0, 1]. Joint temperature trend risk Anomaly parameters in multimodal environment data The fire risk coefficient is obtained by using the following formula, along with the probability of fire. , where k represents a constant controlling the overall risk sensitivity, ranging from [1, 10]; A multi-objective optimization mathematical model is constructed based on energy efficiency ratio, full life cycle benefits and battery health. Combining energy storage operation data and grid demand, a non-dominated sorting genetic algorithm is used to generate the Pareto optimal solution set, optimize the charging and discharging strategy and liquid cooling parameters, and introduce Bayesian optimization technology to dynamically adjust the algorithm parameters. If the temperature remains abnormal, reinforcement learning is used to optimize the liquid cooling pipeline layout, iteratively adjusting the pipe diameter and flow distribution to ensure cooling uniformity and energy efficiency. If any of the following factors still exist under the optimized control strategy scenario: abnormal temperature gradient parameters, temperature trend risks, or abnormal parameters in multimodal environmental data, the step of optimizing the liquid cooling pipeline layout will be triggered. The battery temperature change trend curve and real-time temperature distribution matrix are obtained and combined with heat transfer parameters to input the heat flow simulation model. Reinforcement learning is used to iteratively adjust the pipe diameter, branches and flow distribution. The reward objectives for pipe optimization are set based on cooling uniformity, liquid cooling efficiency and maximum temperature. Optimized pipe design parameters are generated to ensure that cooling uniformity and energy efficiency are both taken into account. The process involves acquiring pipeline design parameters, adjusting the liquid cooling pipeline layout, and verifying the battery temperature change trend curve and real-time temperature distribution matrix to ensure the reliability of pipeline optimization. The heat flow simulation model is a digital twin model used to simulate the heat transfer and distribution characteristics of battery modules in liquid-cooled commercial and industrial energy storage systems. By dividing the battery modules into finite element meshes, the temperature change of each element is calculated based on the heat conduction equation to simulate the impact of liquid cooling pipeline adjustments on heat flow. The process inputs a distribution map of uneven cooling areas and heat transfer parameters, and outputs optimized pipeline design parameters and corresponding temperature distribution predictions. The reward target is calculated as follows: Cooling uniformity is obtained, with higher values ​​indicating better uniformity; cooling efficiency is also obtained. Energy consumption The liquid cooling efficiency is calculated by comparing the two; the reward target is then calculated using the following formula: in, This represents the temperature at a specific time and location. , , These represent the weight coefficients for each item. The highest temperature, As a safety threshold, This is a reference temperature.

2. The energy management and safety protection collaborative method for a liquid-cooled industrial and commercial energy storage system according to claim 1, characterized in that, Based on the battery temperature change trend curve, the frequency of battery data acquisition is adaptively adjusted; the long short-term memory network converts spatial data into a time series format, including dividing the preset heat distribution map into regions and extracting the average temperature of each region to form a time series input.

3. The energy management and safety protection collaborative method for a liquid-cooled industrial and commercial energy storage system according to claim 2, characterized in that, The method includes fire risk management steps, including: Environmental data acquisition: The sensor network also collects vibration, sound, and gas concentration data of the battery, which are then fused to obtain multimodal environmental data; Anomaly detection: A fire risk assessment model is triggered based on abnormal temperature gradient parameters, temperature trend risks, and abnormal parameters of multimodal environmental data. The temperature trend risk refers to the potential overheating risk of the battery temperature in the future based on the predicted temperature change trend. The abnormal parameters of multimodal environmental data refer to the degree to which vibration, sound, and gas concentration exceed the preset range. Fire risk assessment: The fused multimodal environmental information is input into the fire risk assessment model to predict the probability and potential location of fire, calculate the fire risk coefficient, and the probability, potential location and fire risk coefficient constitute the fire risk report. Safety protection measures are set based on the fire risk report.

4. The energy management and safety protection collaborative method for a liquid-cooled industrial and commercial energy storage system according to claim 1, characterized in that, It also includes operation optimization management steps, including: constructing a multi-objective optimization mathematical model based on energy efficiency ratio, full life cycle benefits, and battery health; using the multi-objective optimization mathematical model, combined with energy storage system operation data and grid demand, to quantify operation characteristics, initialize a non-dominated sorting genetic algorithm population to generate preliminary strategies, and generate Pareto optimal solution sets through evolutionary calculations to achieve multi-objective collaborative optimization, optimize charging and discharging strategies and liquid cooling system parameters, and generate Pareto optimal solution sets; and introducing Bayesian optimization technology to dynamically adjust algorithm parameters, improve optimization efficiency and global search capabilities.

5. The energy management and safety protection collaborative method for a liquid-cooled industrial and commercial energy storage system according to claim 4, characterized in that, The operation optimization management steps include the following: Step S31: Construct a multi-objective optimization mathematical model: Extract the grid demand and key parameters of the energy storage system from the operating data. The grid demand data includes at least load and frequency regulation demand, and the key parameters include energy efficiency ratio, life cycle revenue and battery health. Jointly establish a multi-objective optimization mathematical model; output the multi-objective optimization mathematical model and grid demand data. Step S32: Initialize the non-dominated sorting genetic algorithm population: Based on the multi-objective optimization mathematical model and power grid demand data, the non-dominated sorting genetic algorithm is used to initialize the population, and the initial running strategy set is output. Step S33: Generate Pareto optimal solution set through evolution: Based on the initial running strategy set, perform evolutionary calculations using a non-dominated sorting genetic algorithm to output the Pareto optimal solution set; Step S34, Bayesian optimization adjustment: Based on the Pareto optimal solution set, Bayesian optimization technology is introduced to adjust the crossover probability and mutation index of the non-dominated sorting genetic algorithm to generate the final running optimization control strategy.

Images