Regulation and control system and method based on heat energy of energy storage cabinet

Abstract

The invention discloses a regulation and control system and method based on heat energy of an energy storage cabinet, and belongs to the technical field of intelligent heat management. Temperature data in the energy storage cabinet are collected, a temperature distribution matrix is constructed, and a thermal gradient vector and an overheating risk area are extracted based on a difference algorithm; further establishing a thermal power load mapping model, and dynamically generating an air circulation path optimization diagram; the opening angle of a flow deflector and the rotating speed of a fan are adjusted in combination with an optimized path, and self-adaptive regulation and control of a thermal field in the cabinet body are achieved; meanwhile, a machine learning model is trained by utilizing historical operation data, an optimal regulation and control strategy is output, and an execution layer is driven to realize collaborative response of a fan and a flow deflector; the method has the advantages of high regulation and control precision, high response speed, low energy consumption and the like, and is suitable for heat management requirements of various types of energy storage cabinets.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent thermal management, in particular to a regulation system and method based on thermal energy of an energy storage cabinet. BACKGROUND

[0002] Under the background of rapid development of energy storage technology, the energy storage cabinet as the core equipment undertakes the task of energy storage and release. However, with the continuous rise of energy storage power density, the internal of the energy storage cabinet is in a high heat load operation state for a long time, which leads to a significant temperature rise on the surface of the cabinet, and easily triggers battery thermal runaway, system efficiency decline and even safety accidents.

[0003] Traditional heat dissipation designs are mostly focused on passive heat dissipation structures (such as air cooling, natural convection) or local heat exchange units, but such solutions generally have problems such as response delay, uncertain heat accumulation position, and low control precision. Especially in the backpack-type energy storage cabinet deployed outdoors or running in extreme environments, the day-night temperature difference is severe, and the environmental fluctuation is strong. Relying on only a single ventilation or heat dissipation mechanism can easily cause local overheating or uneven heat distribution of the energy storage module, and further affect the stability of the whole system.

[0004] On the other hand, the energy storage cabinet itself generates a large amount of sensible heat (thermal energy) during operation. If this thermal energy can be used as a control variable for "dynamic feedback control" of internal air flow, heat dissipation path and fan behavior, it can realize self-adaptive control of the heat source and further improve the energy efficiency closed-loop regulation capability. However, there is currently no mature technology to realize the closed-loop path of "data extraction - modeling analysis - feedback control" of the internal thermal energy state of the energy storage cabinet. SUMMARY

[0005] The purpose of the present application is to provide a regulation system and method based on thermal energy of an energy storage cabinet to solve the problems in the background art.

[0006] In order to achieve the above purpose, the present application provides the following technical solution: a regulation method based on thermal energy of an energy storage cabinet, comprising: collecting real-time temperature data of different positions of the energy storage cabinet to obtain a temperature distribution matrix , the element represents the current temperature value of the i-th layer and the j-th module of the energy storage cabinet; based on the temperature distribution matrix , calculating the thermal gradient vector G of each local region of the cabinet body, and identifying the potential overheating risk area R through a difference algorithm; According to the thermal gradient vector G and the overheating risk area R, a local thermal power load mapping model H of the cabinet body is constructed, which is used to represent the unit volume heat generation rate of each region; In combination with the heat power load mapping model H, an air flow path optimization graph P is dynamically generated and used to update the opening angle of the air guide vanes and the rotation speed of the fan inside the energy storage cabinet; The historical operation data D of the cabinet body is acquired, and a regulation and control model M is trained using a machine learning algorithm. The regulation and control model M takes the current temperature distribution and the historical operation data D as inputs, and outputs an optimal regulation and control strategy. The output optimal regulation and control strategy is applied to the execution layer to drive the air guide vane mechanism and the fan group to respond cooperatively.

[0007] Preferably, real-time temperature data at different positions of the energy storage cabinet is collected to obtain a temperature distribution matrix T m , including: A temperature sensor array is arranged at multiple hierarchical positions inside the energy storage cabinet, and the temperature sensor array is arranged in a grid manner along the vertical and horizontal directions to form a spatial distribution structure; The temperature sensor array is polled in real time to obtain the temperature values of each sensing node and record the spatial coordinate index (i, j); The temperature value corresponding to each sensor node is mapped to its spatial position one by one to construct a two-dimensional temperature data set Traw(i, j); Based on Traw(i, j), data filtering processing is performed to generate a temperature distribution matrix after removing abnormal temperature points.

[0008] Preferably, the thermal gradient vector G of each local region of the cabinet body is calculated, and a potential overheating risk region R is identified through a difference algorithm, including: The temperature distribution matrix is calculated by using a first-order difference operator in the horizontal and vertical directions to calculate the local temperature gradient for any node (i, j) and , and the two components are combined into a thermal gradient vector G(i, j); Spatial smoothing filtering is performed on G(i, j) to suppress measurement noise, and the gradient module length and the second-order difference are calculated to identify the thermal convex feature; According to a preset threshold set, including a gradient threshold θg, a module length threshold θm, and a Laplace threshold θl, each node is scored, and if and (∂T / ∂x or ∂T / ∂y) > θg, or Laplace > θl, the candidate overheating pixel is marked; The candidate pixel is subjected to connected component analysis and confidence evaluation combined with short-time temperature rise rate, and the connected component set with a confidence higher than a threshold τ constitutes the final potential overheating risk region R, and the region boundary and priority ranking are output.

[0009] Preferably, the construction of the cabinet local thermal power load mapping model H includes: The thermal gradient vector matrix G and the temperature distribution matrix The heat flux density per unit volume Q(i,j) of each node is calculated by superimposing the corresponding spatial coordinates. Q(i,j) is obtained by multiplying the gradient magnitude and the thermal conductivity. Within the identified overheating risk zone R, the heat flux density at each node is integrated over time to obtain the accumulated heat energy E(i,j); Based on the accumulated heat energy E(i,j) and the volume parameters of the module, the heat power load per unit volume P(i,j) is calculated and normalized to form a heat power distribution matrix; The thermal power distribution matrix is ​​transformed into a continuous function model H(x,y) through interpolation and spatial fitting algorithms, which is used to characterize the thermal power density distribution at various locations within the cabinet.

[0010] Preferably, the dynamic generation of the optimized airflow path map P includes: The continuous heat power distribution function H(x,y) is discretized into a set of heat source nodes. A weighted mesh graph Gg is constructed based on the node heat power weights. The mesh nodes represent cabinet space units, the edges represent optional airflow paths, and the edge weights are jointly determined by local resistance, path length, and the temperature drop potential through the edge. Define the path cost function C as a linear combination of edge weights, and specify the constraints including the maximum allowable pressure drop threshold ΔPmax, the minimum ventilation rate Qmin, and the available fan speed range Ωmin to Ωmax. Use the improved A* search algorithm or the shortest path algorithm based on fluid dynamics heuristics to solve for several candidate optimal path sets on Gg. Parallel evaluation based on a fast approximation CFD model is performed on the candidate path set to calculate the expected local temperature drop and flow field uniformity of each path under the current wind turbine operating conditions, and sort them by temperature drop efficiency, energy consumption cost and implementation complexity. The path ranked first is selected as the final optimized airflow path diagram P, and control commands are output to drive the opening angle of the guide vanes and the fan speed.

[0011] Preferably, the opening angle of the guide vanes inside the energy storage cabinet and the fan speed are updated, including: The velocity distribution and pressure gradient of each path segment in the path optimization diagram P are analyzed, and the airflow direction vectors of the flow nodes and high heat load areas are extracted. Based on the angle between the flow vector corresponding to each path segment and the target ventilation direction, the optimal opening angle θopt of the corresponding guide vane is calculated using an angle optimization algorithm, and then adjusted in steps by the actuator. According to the required air volume of the path segment and the path resistance, the required fan pressure ΔP is estimated, and the relationship between the fan pressure and the rotating speed is obtained by table lookup to determine the fan rotating speed Ωopt; The actual adjustment results of the guide vane opening angle θopt and the fan rotating speed Ωopt are fed back in real time, and the fan-guide vane cooperative response model is corrected through deviation analysis.

[0012] Preferably, the output optimal regulation strategy comprises: A multi-dimensional historical data set D of the energy storage cabinet under different operating conditions is collected, including a temperature distribution matrix, a fan rotating speed, a guide vane angle, an environmental temperature and humidity, and an energy consumption record; The data set D is subjected to feature engineering processing, and key feature variables are extracted by using standardization and principal component analysis algorithm; A regulation model M is established based on a supervised learning algorithm, and a gradient boosting tree or a long short-term memory neural network is selected as a core learning structure, and the temperature distribution and the historical control instruction are selected as input samples, and the temperature drop efficiency and the energy consumption are selected as training labels; After the regulation model M converges, the parameters are corrected through cross-validation and online fine-tuning mechanism, so that the regulation model M can output the optimal regulation strategy under real-time input.

[0013] Preferably, the driving guide vane mechanism cooperates with the fan group to respond, comprising: Receiving the optimal control instruction output by the regulation model M, including the target guide vane opening angle θopt and the fan rotating speed Ωopt; Controlling the fan driving circuit through the PWM pulse width modulation mode, converting the target rotating speed Ωopt into a motor control signal, and monitoring the rotating speed feedback value Ωact in real time to correct the error in a closed loop; Driving the guide vane rotating mechanism through the stepper motor controller, positioning the angle according to the target angle θopt, and correcting the initial zero point; During the response process of the guide vane and the fan, the control rhythm is dynamically adjusted according to the execution state and the environmental feedback.

[0014] The application also provides a regulation system based on the thermal energy of the energy storage cabinet, comprising: The data acquisition module acquires real-time temperature data of different positions of the energy storage cabinet to obtain a temperature distribution matrix , wherein the element represents the current temperature value of the i-th layer and the j-th module of the energy storage cabinet; The risk area identification module calculates the thermal gradient vector G of each local area of the cabinet body based on the temperature distribution matrix , and identifies the potential overheating risk area R through a difference algorithm; A model construction module: according to the thermal gradient vector G and the overheating risk area R, a cabinet local thermal power load mapping model H is constructed, which is used to represent the unit volume heat generation rate of each area; An optimization update module: in combination with the thermal power load mapping model H, an air flow path optimization graph P is dynamically generated, and the opening angle of the energy storage cabinet internal guide vane and the fan speed are updated according to the air flow path optimization graph P; A regulation strategy output module: historical operation data D of the cabinet is obtained, and a regulation model M is trained by using a machine learning algorithm, wherein the regulation model M takes the current temperature distribution and the historical operation data D as input, and outputs an optimal regulation strategy; An execution module: the output optimal regulation strategy is applied to an execution layer to drive the guide vane mechanism and the fan group to respond cooperatively.

[0015] In the above technical solution, the present application provides technical effects and advantages: 1. The present application constructs a data model taking the temperature distribution matrix, the thermal gradient vector and the thermal power load as the core, combines the path optimization algorithm, the intelligent control strategy and the closed-loop response mechanism of the execution layer, and forms a full-process, full-structure and self-adaptive thermal energy regulation method for the energy storage cabinet. The scheme can dynamically generate the optimal air flow path according to the internal thermal field change of the cabinet, and adjust the guide vane angle and the fan speed in real time, so as to realize the spatial balance of the temperature distribution, the rapid suppression of local high temperature and the optimization control of the system overall energy consumption.

[0016] 2. The present application is significantly improved in terms of thermal sensing accuracy, control response speed and intelligent regulation strategy. The system has the ability of self-learning and continuous optimization, can adapt to various environmental conditions and operating states, effectively prolongs the service life of the energy storage system, improves the energy efficiency ratio, and significantly enhances the stable operation ability and application breadth of the energy storage cabinet in the complex power system. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0018] Figure 1 The method flowchart of the present application.

[0019] Figure 2 The system module flowchart of the present application. DETAILED DESCRIPTION

[0020] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0021] Embodiment 1, please refer to Figure 1 The energy storage cabinet thermal energy-based regulation method described in the embodiment includes: Collecting real-time temperature data of different positions of the energy storage cabinet to obtain a temperature distribution matrix , the element represents the current temperature value of the jth module of the ith layer of the energy storage cabinet; Based on the temperature distribution matrix , calculating the thermal gradient vector G of each local area of the cabinet body, and identifying the potential overheating risk area R through a difference algorithm; According to the thermal gradient vector G and the overheating risk area R, a local thermal power load mapping model H of the cabinet body is constructed, which is used to represent the unit volume heat generation rate of each area; Combined with the thermal power load mapping model H, an air flow path optimization graph P is dynamically generated, and the opening angle of the air guide fin inside the energy storage cabinet and the rotating speed of the fan are updated according to the air flow path optimization graph P; Obtaining historical operation data D of the cabinet body, training a regulation model M by using a machine learning algorithm, the regulation model M taking the current temperature distribution and the historical operation data D as inputs, and outputting an optimal regulation strategy; The output optimal regulation strategy is applied to an execution layer to drive the air guide fin mechanism and the fan set to respond cooperatively.

[0022] A plurality of temperature sensors are preset inside the energy storage cabinet to form a two-dimensional temperature sensing network. The temperature sensors are arranged in an array form along the hierarchical direction (i.e. the height direction) and the horizontal distribution direction of the energy storage cabinet, forming a regular grid structure. Each layer corresponds to a plurality of modules on a horizontal plane, and each module is fixedly installed with a temperature sensor, thereby realizing comprehensive coverage of each key heating area inside the energy storage cabinet.

[0023] Taking a specific implementation as an example, it is assumed that the energy storage cabinet contains 5 layers of battery modules, and each layer contains 6 battery modules, so 5x6=30 temperature sensors need to be configured. These sensors can adopt thermal resistance type, thermocouple type or digital temperature acquisition devices, and digital sensors with industrial precision and response time less than 1 second are preferred to meet the requirements of fast and stable data acquisition.

[0024] To realize efficient access and data synchronization collection of each sensor node, a set of distributed temperature collection units is deployed. The collection unit includes a master collector (such as an MCU or an ARM processor) and a multi-channel polling module, which sequentially polls each node through multiple data collection channels. Each polling cycle is controlled within 500 ms to ensure that the sampling refresh rate meets the system thermal change response requirements.

[0025] In each collection process, the master unit sequentially accesses each sensor, records its temperature value T, and binds the corresponding spatial coordinate index (i, j) of the value, where i represents the vertical level where the sensor is located, and j represents the module number in the level. For example, That is, the current temperature of the 3rd layer and the 5th module.

[0026] All temperature values collected by polling are constructed into a two-dimensional temperature data set T_raw(i, j) to represent the original temperature measurement results of each node. However, considering that sensors may produce drift values or abnormal high or low values due to interference or failure, a data filtering mechanism is introduced.

[0027] The specific method includes: Upper and lower limit value filtering is performed on each data point in Traw to eliminate abnormal values outside the reasonable physical range (such as -20℃~80℃); The spatial neighborhood average method is used to correct single isolated abnormal values, that is, for a point whose temperature value deviates from the surrounding by more than 10℃, the average value of its four adjacent valid points is used to replace it; For consecutive missing or drifting data points, they are preferentially accessed and redundantly sampled in the next collection cycle.

[0028] After filtering and abnormality elimination, the temperature data is formally constructed into a temperature distribution matrix , which is a two-dimensional array , and the element represents the current stable temperature value of the i-th layer and j-th module of the energy storage cabinet.

[0029] The energy storage cabinet temperature distribution matrix is a two-dimensional matrix, and the element represents the real-time temperature value of the i-th layer and j-th module of the energy storage cabinet. To obtain the temperature change trend of the local area inside the energy storage cabinet, the present application uses a first-order difference algorithm to calculate the thermal gradient of each node.

[0030] Taking node as the center point, the first-order difference value in the horizontal and vertical directions is calculated respectively as the gradient estimate of the temperature in that direction. Specifically: The horizontal temperature gradient ∂T / ∂x is defined as: ; The vertical temperature gradient ∂T / ∂y is defined as: ; The two gradient values ​​mentioned above together form the thermal gradient vector G(i,j), where G(i,j) is a two-dimensional vector used to describe the heat transfer trend at that point along two directions. To further analyze the thermal gradient intensity, the magnitude of G(i,j), i.e., the amplitude of the thermal gradient vector at that point, is calculated to assess the degree of heat concentration. The magnitude is defined as the square root of the sum of the squares of the horizontal and vertical components of G(i,j).

[0031] This thermal gradient calculation operation iterates through all valid points within the matrix, ultimately generating a set of thermal gradient vector matrices with the same dimensions as the original temperature matrix. .

[0032] Considering that temperature sensors may experience sampling fluctuations or short-term noise during the data acquisition process, direct differential grading may lead to abnormally large gradient values. Therefore, it is necessary to... Spatial filtering is performed to improve stability.

[0033] This invention employs a Gaussian filtering method to smooth the thermal gradient vector matrix. Specifically, a 3×3 Gaussian kernel is used to perform a weighted average of each gradient point and its eight surrounding points to suppress sharp local changes, thereby eliminating misjudgments caused by sensor errors.

[0034] At the same time, it also calculates The second-order spatial difference, or Laplace operator, is used to identify spatially elevated regions of heat. The Laplace difference is calculated by multiplying the temperature value at the center point by -4 and adding the sum of the temperatures of its four adjacent points in all directions, to determine whether the point is located in a temperature "plateau" or a local peak region.

[0035] thermal gradient With Laplace matrix After the calculation is completed, the present invention sets a set of thermal risk assessment thresholds, including: Gradient magnitude threshold θg: When the thermal gradient magnitude exceeds this value, it indicates that the local temperature changes drastically. Laplace threshold θl: When the Laplace difference result is positive, it indicates that heat accumulates at this point; Temperature difference threshold θt: Temperature difference between adjacent points exceeding this value also constitutes a risk factor.

[0036] In this invention, the default settings are as follows: θg is set to 4 degrees Celsius per unit distance, θl is set to 2 degrees Celsius, and θt is set to 5 degrees Celsius.

[0037] For each node (i,j), execute the following judgment logic: If the thermal gradient modulus at that point is greater than θg; And the temperature difference with any adjacent point is greater than θt; And the Laplace value at that point is greater than θl; The node is then marked as a hot anomaly and used as a seed pixel for a candidate risk region.

[0038] This composite decision-making strategy can effectively eliminate atypical risk areas such as high temperature or high gradient, thereby enhancing the engineering reliability of the algorithm.

[0039] After candidate point identification, this invention further performs spatial connectivity analysis. Using the 8-adjacency connectivity principle, adjacent candidate points are grouped into the same connected region. For each connected region, its area, center coordinates, maximum temperature, and average gradient value are calculated.

[0040] Meanwhile, a dynamic confidence factor is introduced, defined as a weighted combination of the average temperature rise rate (temperature change per unit time) and gradient intensity within the region.

[0041] The formula for calculating the confidence level C is as follows: Where ΔT / Δt represents the average temperature rise rate of the region in the past two sampling rounds, mean(|G|) is the average thermal gradient intensity of the region, and α and β are weighting coefficients. In the actual measurement, α = 0.6 and β = 0.4.

[0042] If the confidence level of a certain connected component is greater than the set threshold τ (e.g., τ = 5), it is finally identified as a potential overheating risk area R, and its spatial boundary, area number, and risk level label (which can be set to low, medium, or high) are output.

[0043] The information from this area will be transmitted as input to the fan control module and the airflow reconfiguration module to drive subsequent thermal regulation actions.

[0044] In heat conduction theory, heat flux density (i.e., the rate of heat transfer per unit area per unit time) can be described by Fourier's law, meaning that heat flux density is proportional to the temperature gradient. Therefore, this invention uses the aforementioned thermal gradient vector G(i,j) as input to calculate the heat flux density Q(i,j) per unit volume at each sampling node, as follows: The effective thermal conductivity λ of each material region in the energy storage cabinet is set, with the unit being watts per meter per degree Celsius (W / m·K), which can be given through experiments or material handbooks; For the j-th module of the i-th layer of the energy storage cabinet, the magnitude of its corresponding thermal gradient vector G(i,j) is known to be... Then, the formula for calculating its heat flux density Q(i,j) is: ;in Let G(i,j) be the Euclidean norm, defined as the square root of the sum of the squares of the gradient components in the horizontal and vertical directions at that point. After this step, the heat flux density matrix can be obtained. Dimensions and temperature matrix Consistent.

[0045] This invention further proposes to integrate the heat flux density Q(i,j) over time to estimate the heat accumulation trend per unit volume of a module in the energy storage cabinet. The heat accumulation E(i,j) reflects the intensity of continuous heat input within a specific time interval, and is defined as follows: Set a time integration window Δt (e.g., 5 minutes or 300 seconds), and integrate Q(i,j) for each node within this time period; Simplifying to a discrete summation form, let's assume the node samples N times within Δt, and Q(i,j,k) represents the heat flux density at the k-th sampling. Then: , k=1 to N; where δt is the interval between each sampling, in seconds. E(i,j) is in joules per cubic meter (J / m³), representing the heat energy accumulation density of this volume region.

[0046] It is important to note that the solution of E(i,j) is performed only within the overheating risk region R to reduce the waste of computational resources and focus on the high-risk region of thermal problems.

[0047] Thermal power load is an important parameter describing the intensity of heat input per unit volume. It represents the average heat output per cubic meter of space at the current time, and its unit is watts per cubic meter (W / m³). In this invention, based on the accumulated heat energy E(i,j), combined with the time window Δt and the module volume V(i,j), the unit volume heat power load P(i,j) of each node is calculated, and the calculation formula is as follows: Where Δt is the integration window time (in seconds), E(i,j) is in joules per cubic meter, and P(i,j) is in watts per cubic meter. If the integration method used has been calculated per unit volume, then the heat power density P(i,j) can be directly regarded as the time average of Q(i,j).

[0048] To standardize the evaluation of thermal power, P(i,j) needs to be normalized to become the thermal power density index D(i,j), limiting its range to between 0 and 1. The normalization process is as follows: Let the maximum heat power density be P_max and the minimum be P_min, then we have: This index can be used as a weight input for subsequent regulatory strategies, improving the sensitivity and hierarchy of these strategies.

[0049] To achieve a continuous spatial representation of the overall thermal state of the energy storage cabinet, this invention introduces an interpolation fitting algorithm to smoothly fit the discrete thermal power load values ​​P(i,j) and generate a continuous two-dimensional function model H(x,y) to represent the thermal power density distribution at any location in the cabinet.

[0050] The specific method is as follows: The spatial coordinates (xi, yj) of P(i,j) and the thermal power values ​​form a training sample set; Interpolation methods such as bicubic spline interpolation, Laplace weighted average interpolation, or radial basis function neural network (RBF) are applied to construct a function H(x,y), whose output is the estimated heat power load at any location point; The model H(x,y) satisfies the boundary continuity and smoothness conditions and can be updated and fed back in real time, enabling continuous perception and visualization of the thermal field.

[0051] The continuous function H(x,y) is divided into a rectangular grid of m rows and n columns with a fixed spatial resolution. The grid nodes are denoted as Gg(i,j). Each node represents an independent spatial unit inside the energy storage cabinet, and the thermal power value of the node comes from the function value of H(x,y) at the center point of that unit.

[0052] For every two adjacent grid nodes (vertically or horizontally adjacent), a bidirectional feasible path edge is defined. The weight W(i,j,k,l) ​​of each edge represents the airflow cost from node (i,j) to (k,l). This cost is a linear combination of the following three terms: The local flow resistance value R(i,j,k,l) ​​depends on the presence or absence of structural obstacles and can be obtained from CFD simulation or empirical models. Path length L(i,j,k,l): Euclidean distance between the center points of two nodes; Temperature reduction potential ΔT(i,j,k,l): represents the cooling efficiency potential of the path connecting two regions with different thermal power values, and is a function of the temperature difference between the high-temperature node and the low-temperature node.

[0053] The weight combination method is defined as follows: Where w1, w2, and w3 are adjustment coefficients controlling the weights of each factor, typically set as follows: w1 = 0.5, w2 = 0.3, and w3 = 0.2. Finally, a weighted mesh graph Gg is constructed, where nodes represent spatial points and edges represent walkable paths and their costs.

[0054] To select the optimal ventilation path in Gg, this invention defines a path cost function C, which is the sum of the weights of all edges in the path: , where P is the sequence of edges connected sequentially in the path.

[0055] The following constraints must be met during path search: Maximum pressure drop threshold ΔP_max: The entire path should not cause the fan to exceed this pressure loss; Minimum ventilation rate Q_min: The path must meet the minimum ventilation capacity to ensure heat exchange; The fan's permissible speed range is from Ω_min to Ω_max, which limits the actual wind speed range.

[0056] Based on the above conditions, this invention employs an improved A* search algorithm to generate several candidate optimal paths that satisfy the constraints in the Gg graph, starting from the heat source and ending at the cold source or air inlet. The heuristic function h(n) of the A* algorithm is defined as the distance from the current node n to the target node divided by the cooling potential of the current path, taking into account both target orientation and energy efficiency.

[0057] To select the optimal ventilation path, this invention performs fast approximate CFD simulations on the aforementioned candidate path set to evaluate the following performance metrics: Uniformity of the air velocity field along the path (measured by the root mean square of streamline deviation). Local maximum temperature drop capacity (ΔT_max); Energy cost of implementation (fan power × ventilation time); Complexity of structural adjustments (range of movement angle of the guide vanes, etc.).

[0058] The above indicators are normalized and scored, and the overall path score function is defined: Where s1 to s4 are weight coefficients. The path with the highest score is selected as the final optimized path graph P.

[0059] After the path diagram P is generated, it needs to be converted into control commands for the internal structure of the energy storage cabinet. This invention establishes an airflow structure response mapping mechanism to achieve adaptive adjustment of the guide vane opening angle and the fan speed, as detailed below: The airflow direction vector D(i,j) for each path segment in the analytical path diagram P is calculated and compared with the physical installation direction of the guide vanes inside the energy storage cabinet. For each guide vane position, its optimal opening angle θ_opt is calculated to minimize the angle between the airflow direction and the channel direction.

[0060] The following angle optimization formula is adopted: Where D_target is the current direction vector of the guide vane, and D_path is the airflow direction of the path segment. θ_opt is controlled between 0 and 45 degrees, with fine step control via an electric actuator.

[0061] Based on the expected wind speed and pressure drop in the path segment, estimate the required wind pressure ΔP, consult the fan parameter table, and deduce the required fan speed Ω_opt to achieve this wind pressure. Assume the empirical model for fan pressure and speed is: Then we can obtain: Where k is a coefficient provided by the fan manufacturer. The controller uses Ω_opt as the command value to adjust the fan to the target speed, with a speed control accuracy better than ±100 RPM.

[0062] The actual thermal field changes under the current guide vane angle θ_act and fan speed Ω_act are sampled. If the deviation from the prediction exceeds the threshold ε (e.g., ΔT prediction error is greater than 2), the result is recorded. If the error is corrected by backpropagation, the fan-guide vane coupling response model will be updated.

[0063] Historical operational data D serves as the foundational data resource for model training, and its dimensions must encompass all key variables affecting thermal regulation results during the operation of the energy storage cabinet. This invention defines the composition of the historical dataset D as follows: Temperature data: including temperature distribution matrices at various times. The dimension is i rows and j columns, recording the temperature changes at different spatial locations of the energy storage cabinet; Control commands include the guide vane opening angle matrix θ(t) and the fan speed Ω(t); Environmental conditions: including ambient temperature T_env(t), relative humidity H_env(t), and inlet air temperature T_in(t); Thermal response index: Record the temperature drop effect ΔT(t) under the current control strategy, the temperature drop value per unit time, and the corresponding energy consumption E(t); Time stamps include sampling timestamps and operation phase numbers.

[0064] The data acquisition cycle is 1 to 10 seconds, depending on the application scenario's response speed requirements. The total data volume should use no less than 10,000 samples as the initial training baseline, and the sampling period should cover different seasons, day and night, and load conditions.

[0065] The raw data D needs to undergo systematic preprocessing before being directly used for model training to improve feature representation capabilities, reduce dimensionality redundancy, and enhance model convergence speed. This invention employs the following feature engineering methods: Missing values ​​are filled using linear interpolation or historical mean; Standardize all numerical features by using the zero-mean, unit-variance method to ensure that feature values ​​fall within a similar range, thereby improving gradient learning efficiency.

[0066] Principal component analysis (PCA) is used to extract the top K principal components for high-dimensional spatial features (such as temperature distribution matrices), which typically retains more than 95% of the cumulative variance explained. Correlation analysis was performed on the control variables (such as angle and rotational speed) and the response results (temperature drop efficiency) to eliminate low-contribution or redundant features; Construct combined features, such as "current heat power density × wind speed ratio" and "number of local hot spots × fan speed", as auxiliary inputs.

[0067] The final constructed feature vector X(t) includes variables such as the current environmental state, the principal component of the temperature distribution, the historical control state and its rate of change.

[0068] After completing the feature vector construction, this invention uses a supervised learning framework for model training, aiming to build a prediction model that takes the current state X(t) as input and outputs the optimal control policy S_opt(t)={θ_opt(t), Ω_opt(t)}.

[0069] Based on the characteristics of the problem and timing requirements, two types of algorithm structures are provided for selection: Gradient Boosting Tree (GBDT) model: suitable for modeling the relationship between static input and labels, with clear feature representation and efficient training, suitable for medium-sized datasets; Long Short-Term Memory Neural Network (LSTM): Suitable for time-dependent sequence prediction tasks, it can handle the delay and feedback effects of temperature control strategies and is suitable for dynamic control scenarios.

[0070] In actual deployment, GBDT can be used as the basic model first, and LSTM can be used for advanced strategy modules that need to consider the influence of past states on future decisions.

[0071] The training objective is to minimize the difference between the predicted control policy and the actual optimal control behavior. This invention employs the following multi-objective loss function: Main loss term: Control strategy deviation loss =Mean squared error (MSE); Auxiliary loss term: Temperature drop efficiency error =The difference between ΔT caused by the current strategy and the historical best ΔT; Energy consumption regularization term =The deviation between the current strategy's corresponding wind turbine power consumption and the average power consumption.

[0072] The total loss function is defined as: ;in, These are weighting coefficients, typically set to 0.5, 0.3, or 0.2.

[0073] A 5-fold cross-validation method is used to evaluate the model's generalization ability, ensuring that the model does not overfit. An Early Stopping mechanism is implemented during training, automatically terminating training if the validation error does not decrease for several consecutive rounds.

[0074] The final output of the model is: M(X) → S_opt(t)={θ_opt(t), Ω_opt(t)}; that is, at any given time, the current state is input, and the optimal angle of the guide vane and the fan speed are output.

[0075] The control model M outputs the optimal control strategy S_opt for the current cycle in real time. The system's main control chip (such as STM32, ARM Cortex-M series, or edge AI processor) needs to receive this control vector immediately and complete data parsing and protocol conversion. The main operations include: Extract the control fields in S_opt into floating-point values ​​via UART, CAN or SPI bus, and assign them to θ_target and Ω_target respectively; For the control instruction formats required by the execution layer control module (such as PWM duty cycle, step pulse frequency, etc.), the target values ​​are linearly mapped and converted to integers. A double-buffer processing mechanism is adopted to prevent signal overwriting or execution lag caused by the model update frequency being faster than the execution frequency.

[0076] Example: If the target fan speed is 2400 RPM, and the fan controller requires a PWM input signal of 0 to 1000 corresponding to 0 to 3000 RPM, then the control signal is: PWM_duty = 2400 ÷ 3000 × 1000 = 800. This value is then sent to the fan drive.

[0077] The fan control employs a PWM drive method. The controller outputs a duty cycle signal to the drive circuit module based on the target speed Ωopt to achieve precise speed regulation. The closed-loop control process is as follows: The controller periodically reads Ωopt and converts it into the corresponding PWM duty cycle; Control signals are sent to the motor drive module via the PWM channel to control the speed. The wind turbine body is equipped with a Hall sensor or photoelectric speed measurement module, and the current actual speed Ωact is returned after each round of sampling; Calculate the speed error If it exceeds the set threshold δΩ (e.g.) If RPM is detected, the PWM output will be adjusted for secondary correction. Moving average filtering is applied to the Ωact sampling data to avoid spurious feedback fluctuations caused by electromagnetic interference. This process ensures that the fan speed regulation accuracy is controlled within ±3%, meeting the dynamic ventilation requirements.

[0078] The opening angle θopt of the air guide vane directly determines the airflow directionality and the local heat dissipation path. Its control system includes a stepper motor, a reduction gear set, an angle sensor (or encoder), and a limit detection unit. The control flow is as follows: Convert the target angle θopt into the number of steps for the stepper motor. For example, each step of the stepper motor is an angle of θopt. The target angle is It requires 15 steps; Before each power-on or system reconfiguration, reverse the motor to the physical limit switch and record the current position. To ensure the accuracy of the reference angle; After executing the target number of steps, read the actual angle θact returned by the angle sensor and calculate the error εθ by comparing it with the target value; If the error εθ exceeds the set tolerance δθ (it is recommended to set it to...), If this is the case, then micro-step adjustments will be automatically added to ensure execution accuracy; This mechanism ensures the accuracy and repeatability of airflow guidance control, providing a guarantee for stable local temperature control.

[0079] To improve the dynamic consistency and execution efficiency of temperature response, this invention proposes a coordinated control mechanism between fan speed and guide vane angle. This mechanism uses soft synchronization logic to coordinate the execution rhythm of both, ensuring coordinated response even under drastic changes in heat flux. Specifically: When the heat power density in a certain area increases significantly, it simultaneously triggers the adjustment of the guide vanes and the speed control of the fan. A synchronization coefficient μ∈[0,1] is set to control the response priority. If μ→1, the focus is on wind speed regulation; if μ→0, the focus is on airflow guidance regulation. If the current heat change trend is predicted to continue to rise, the guide vane rotation mechanism will be activated first (fast response); otherwise, the fan speed will be increased first (strong heat dissipation). θact and Ωact are acquired every 5 seconds, and combined with the automatic correction control strategy of the hot zone temperature change rate ΔT / Δt, dynamic iteration is achieved.

[0080] Through the above mechanism, the heat dissipation system of the energy storage cabinet can be adjusted synchronously according to the real-time thermal field status, preventing local heat accumulation caused by the lag of a single component.

[0081] Example 2, please refer to Figure 2 As shown in this embodiment, a control system based on the thermal energy of an energy storage cabinet includes: Data acquisition module: Collects real-time temperature data from different locations within the energy storage cabinet to obtain a temperature distribution matrix. Its elements This represents the current temperature value of the j-th module in the i-th layer of the energy storage cabinet; Risk area identification module: based on temperature distribution matrix Calculate the thermal gradient vector G of each local area of ​​the cabinet, and identify potential overheating risk areas R through a differential algorithm; Model building module: Based on the thermal gradient vector G and the overheating risk area R, a local thermal power load mapping model H of the cabinet is constructed to represent the heat generation rate per unit volume in each area; Optimization and update module: Combined with the thermal power load mapping model H, dynamically generate the air flow path optimization diagram P, and update the opening angle of the guide vanes and the fan speed inside the energy storage cabinet according to it; Control strategy output module: Acquires historical operating data D of the cabinet, trains a control model M using a machine learning algorithm, and the control model M is based on the current temperature distribution. Using historical operating data D as input, the output is the optimal control strategy; Execution module: The output optimal control strategy is applied to the execution layer to drive the guide vane mechanism and the wind turbine to respond in a coordinated manner.

[0082] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for regulating thermal energy in an energy storage cabinet, characterized in that: include: Real-time temperature data is collected from different locations within the energy storage cabinet to obtain a temperature distribution matrix. Its elements This represents the current temperature value of the j-th module in the i-th layer of the energy storage cabinet; Based on temperature distribution matrix Calculate the thermal gradient vector G of each local area of ​​the cabinet, and identify potential overheating risk areas R through a differential algorithm; Based on the thermal gradient vector G and the overheating risk region R, a local thermal power load mapping model H for the cabinet is constructed to represent the heat generation rate per unit volume of each region. Based on the aforementioned thermal power load mapping model H, an optimized airflow path diagram P is dynamically generated, and the opening angle of the guide vanes and the fan speed inside the energy storage cabinet are updated accordingly. Obtain historical operating data D of the cabinet, and train a control model M using a machine learning algorithm. The control model M is based on the current temperature distribution. Using historical operating data D as input, the output is the optimal control strategy; The optimal control strategy output is applied to the execution layer to drive the guide vane mechanism and the wind turbine to respond in a coordinated manner.

2. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The real-time temperature data from different locations of the energy storage cabinet is collected to obtain a temperature distribution matrix T. m ,include: Temperature sensor arrays are installed at multiple levels inside the energy storage cabinet. The temperature sensor arrays are arranged in a grid pattern along the vertical and horizontal directions to form a spatial distribution structure. The temperature sensor array is polled in real time to obtain the temperature value of each sensor node and record its spatial coordinate index (i,j); Map the temperature value corresponding to each sensor node to its spatial location one by one to construct a two-dimensional temperature dataset Traw(i,j); Data filtering is performed based on Traw(i,j) to remove outlier temperature points and generate a temperature distribution matrix. .

3. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The calculation involves determining the thermal gradient vector G of each local area of ​​the cabinet and identifying potential overheating risk areas R using a differential algorithm, including: For the temperature distribution matrix The local temperature gradient is calculated using first-order difference operators in both the horizontal and vertical directions, i.e., for any node (i,j)... and The two components are then combined into a thermal gradient vector G(i,j); Spatial smoothing filtering is applied to G(i,j) to suppress measurement noise, and the gradient magnitude is calculated. And second-order difference to identify thermal protrusion features; Each node is scored based on a preset threshold set, including gradient threshold θg, magnitude threshold θm, and Laplace threshold θl. If (∂T / ∂x or ∂T / ∂y) > θg, or Laplacian > θl, then it is marked as a candidate overheated pixel; Connectivity analysis is performed on candidate pixels, and confidence is evaluated by combining short-term temperature rise rate. The set of connected components with confidence above threshold τ constitutes the final potential overheating risk region R, and the region boundary and priority ranking are output.

4. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The construction of the cabinet's local thermal power load mapping model H includes: The thermal gradient vector matrix G and the temperature distribution matrix The heat flux density per unit volume Q(i,j) of each node is calculated by superimposing the corresponding spatial coordinates. Q(i,j) is obtained by multiplying the gradient magnitude and the thermal conductivity. Within the identified overheating risk zone R, the heat flux density at each node is integrated over time to obtain the accumulated heat energy E(i,j); Based on the accumulated heat energy E(i,j) and the volume parameters of the module, the heat power load per unit volume P(i,j) is calculated and normalized to form a heat power distribution matrix; The thermal power distribution matrix is ​​transformed into a continuous function model H(x,y) through interpolation and spatial fitting algorithms, which is used to characterize the thermal power density distribution at various locations within the cabinet.

5. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The dynamically generated airflow path optimization map P includes: The continuous heat power distribution function H(x,y) is discretized into a set of heat source nodes. A weighted mesh graph Gg is constructed based on the node heat power weights. The mesh nodes represent cabinet space units, the edges represent optional airflow paths, and the edge weights are jointly determined by local resistance, path length, and the temperature drop potential through the edge. Define the path cost function C as a linear combination of edge weights, and specify the constraints including the maximum allowable pressure drop threshold ΔPmax, the minimum ventilation rate Qmin, and the available fan speed range Ωmin to Ωmax. Use the improved A* search algorithm or the shortest path algorithm based on fluid dynamics heuristics to solve for several candidate optimal path sets on Gg. Parallel evaluation based on a fast approximation CFD model is performed on the candidate path set to calculate the expected local temperature drop and flow field uniformity of each path under the current wind turbine operating conditions, and sort them by temperature drop efficiency, energy consumption cost and implementation complexity. The path ranked first is selected as the final optimized airflow path diagram P, and control commands are output to drive the opening angle of the guide vanes and the fan speed.

6. The method for regulating thermal energy of an energy storage cabinet according to claim 5, characterized in that: Update the opening angle of the internal guide vanes and the fan speed of the energy storage cabinet, including: The velocity distribution and pressure gradient of each path segment in the path optimization diagram P are analyzed, and the airflow direction vectors of the flow nodes and high heat load areas are extracted. Based on the angle between the flow vector corresponding to each path segment and the target ventilation direction, the optimal opening angle θopt of the corresponding guide vane is calculated using an angle optimization algorithm, and then adjusted in steps by the actuator. Estimate the required fan pressure ΔP based on the required air volume and path resistance of the path segment, and obtain the relationship between air pressure and speed from the table to determine the fan speed Ωopt; The actual adjustment results of the guide vane opening angle θopt and the fan speed Ωopt are sampled in real time, and the fan-guide vane coordinated response model is corrected through deviation analysis.

7. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The optimal output control strategy includes: Collect multidimensional historical dataset D of the energy storage cabinet under different operating conditions, including temperature distribution matrix, fan speed, guide vane angle, ambient temperature and humidity, and energy consumption records; Feature engineering was performed on dataset D, and key feature variables were extracted using standardization and principal component analysis algorithms. A control model M is established based on a supervised learning algorithm. Gradient boosting tree or long short-term memory neural network is selected as the core learning structure. Temperature distribution and historical control commands are used as input samples, and temperature drop efficiency and energy consumption are used as training labels. After the control model M converges, the parameters are corrected through cross-validation and online fine-tuning mechanisms, so that the control model M can output the optimal control strategy under real-time input.

8. The method for regulating thermal energy of an energy storage cabinet according to claim 1, characterized in that: The drive guide vane mechanism responds in coordination with the wind turbine unit, including: Receive the optimal control command output by the control model M, including the target guide vane opening angle θopt and the fan speed Ωopt; The fan drive circuit is controlled by PWM pulse width modulation, the target speed Ωopt is converted into a motor control signal, and the speed feedback value Ωact is monitored in real time to correct the error in a closed loop. The guide vane rotation mechanism is driven by a stepper motor controller to perform angle positioning based on the target angle θopt and to correct the starting zero point. During the response of the guide vanes and the fan, the control rhythm is dynamically adjusted based on the execution status and environmental feedback.

9. A control system based on the thermal energy of an energy storage cabinet, used to implement the control method based on the thermal energy of an energy storage cabinet as described in any one of claims 1-8, characterized in that: include: Data acquisition module: Collects real-time temperature data from different locations within the energy storage cabinet to obtain a temperature distribution matrix. Its elements This represents the current temperature value of the j-th module in the i-th layer of the energy storage cabinet; Risk area identification module: based on temperature distribution matrix Calculate the thermal gradient vector G of each local area of ​​the cabinet, and identify potential overheating risk areas R through a differential algorithm; Model building module: Based on the thermal gradient vector G and the overheating risk area R, a local thermal power load mapping model H of the cabinet is constructed to represent the heat generation rate per unit volume in each area; Optimization and update module: Combined with the thermal power load mapping model H, dynamically generate the air flow path optimization diagram P, and update the opening angle of the guide vanes and the fan speed inside the energy storage cabinet according to it; Control strategy output module: Acquires historical operating data D of the cabinet, trains a control model M using a machine learning algorithm, and the control model M is based on the current temperature distribution. Using historical operating data D as input, the output is the optimal control strategy; Execution module: The output optimal control strategy is applied to the execution layer to drive the guide vane mechanism and the wind turbine to respond in a coordinated manner.

Images