A smart power consumption optimization method and system for network construction type energy storage

Abstract

The present application belongs to the field of network type energy storage cabinet, especially relates to a kind of intelligent power consumption optimization method and system for network type energy storage, the method is by real-time acquisition operating state data, external dispatching instruction and system operation mode, constructs dynamic hypergraph model to integrally represent the coupling relationship between power scheduling task, heat load demand and heat management energy consumption;Further, with the minimum system total operating energy consumption as the goal, based on the model collaborative optimization solution, generate the joint operation strategy of power scheduling and heat management control in future period;Finally, execute the strategy on digital twin energy storage cabinet, realize the closed-loop joint control of power conversion and heat management subsystem, thereby solve the energy efficiency bottleneck problem caused by the split optimization of electric heating system, improve overall operation economy and safety.

Description

Technical Field

[0001] This invention belongs to the field of grid-type energy storage cabinets, and particularly relates to a smart power consumption optimization method and system for grid-type energy storage. Background Technology

[0002] Currently, research on energy consumption optimization for grid-type energy storage cabinets has shifted from basic heat dissipation to energy efficiency management of the thermal management subsystem itself. Existing technologies, such as Chinese Patent No. CN118248999B, disclose a thermal management method and system for grid-type energy storage cabinets. By employing direct cooling and heating cycles to reduce heat exchange stages, it directly reduces the inherent energy consumption of the thermal management subsystem. Furthermore, another approach, such as Chinese Patent No. CN119298843B, discloses a dynamic balancing configuration method and system for heat dissipation power consumption of photovoltaic energy storage cabinets. By deploying a temperature sensor network and constructing a predictive model, it achieves dynamic control of heat dissipation equipment such as fans and liquid-cooled pumps to optimize their operating energy consumption. However, these methods all focus on optimizing the energy efficiency of the thermal management subsystem in isolation, without analyzing its energy consumption within the framework of the overall power consumption of the energy storage system. Specifically, these solutions lack consideration for the following issues: During dynamic processes such as grid frequency regulation and charging / discharging, the operating conditions of the core power circuit of the energy storage cabinet change rapidly, and the generated heat fluctuates accordingly. Existing thermal management energy consumption optimization strategies rely solely on temperature feedback or prediction, failing to proactively coordinate with the power commands and operating modes of the main circuit. This results in the thermal management subsystem's power consumption changes potentially lagging or failing to optimally match actual heat dissipation needs, leading to overcooling / heating or untimely response, thus hindering further improvements in overall system energy efficiency. Therefore, establishing a dynamic collaborative analysis model between the energy consumption of the thermal management subsystem and the power of the energy storage main circuit, and achieving joint optimization of both on both time and energy scales, has become an urgent technical problem to be solved. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes an intelligent power consumption optimization method and system for grid-type energy storage, which solves the energy efficiency bottleneck caused by the disconnect between electricity and heat in traditional control, and achieves systematic energy saving and safe and reliable operation.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A smart energy consumption optimization method for grid-based energy storage includes:

[0006] Acquire operational status monitoring data, external dispatch commands, and system operation modes of grid-type energy storage cabinets;

[0007] Based on the aforementioned operational status monitoring data, scheduling instructions, and system operation modes, a dynamic hypergraph model is constructed to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand, and thermal management energy consumption in the grid-type energy storage cabinet.

[0008] With the goal of minimizing the total operating energy consumption of the grid-type energy storage cabinet, a collaborative optimization solution is performed based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles.

[0009] On the pre-defined twin energy storage cabinet, the cooperative operation strategy is executed to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet.

[0010] Specifically, the construction process of a dynamic hypergraph model includes:

[0011] Based on real-time monitoring data from sensors within the grid-type energy storage cabinet, the output power of the power conversion subsystem, the state of charge and temperature of the battery clusters, the temperature of the H-bridge, the temperature of the cabinet environment, and the real-time power consumption of the thermal management subsystem are obtained.

[0012] The communication messages collected from the energy management subsystem are parsed to obtain the power scheduling task instructions of the grid-type energy storage cabinet. The power scheduling task instructions include the target power value and the effective time window.

[0013] Based on the local controller status of the network-type energy storage cabinet or the mode identifier bit in the communication message, the current system operating mode is obtained;

[0014] Based on the target power value in the power scheduling task instruction, the real-time state of charge of the battery cluster, and the real-time output power of the power conversion subsystem, the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions is obtained by querying a preset loss-power-state of charge mapping table.

[0015] Specifically, the construction process of the dynamic hypergraph model also includes:

[0016] Based on the instantaneous total heat source power, the average temperature of the battery cluster, the real-time junction temperature of the H-bridge, and the thermal resistance-capacity parameter network pre-calibrated according to the physical structure of the energy storage cabinet, the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge within a future preset time period is obtained by solving the state-space equation corresponding to the thermal resistance-capacity parameter network. The theoretical temperature rise trajectory constitutes a working condition coupling hyperedge in the dynamic hypergraph model. The nodes associated with the working condition coupling hyperedge include: a command node representing the target power value, a node representing the real-time state of charge, a node representing the average temperature of the battery cluster, and a node representing the junction temperature of the H-bridge.

[0017] Specifically, the construction process of the dynamic hypergraph model also includes:

[0018] Based on the theoretical temperature rise trajectory, at least one operating condition coupled hyperedge is generated in the dynamic hypergraph model, wherein the operating condition coupled hyperedge is associated with the instruction node representing the target power value, the node representing the real-time state of charge, the node representing the average temperature of the battery cluster, and the node representing the H-bridge junction temperature.

[0019] Based on the real-time temperature of the cabinet environment, the real-time power consumption of the thermal management subsystem, the system operation mode, the preset cooling energy efficiency ratio characteristic curve, and the theoretical temperature rise trajectory, the theoretical cooling power consumption sequence is obtained.

[0020] Based on the theoretical cooling power consumption sequence, a regulation response hyperedge is generated in the dynamic hypergraph model, wherein the regulation response hyperedge is associated with at least two nodes among the nodes representing the average temperature of the battery cluster, the nodes representing the H-bridge junction temperature, the nodes representing the ambient temperature inside the cabinet, the nodes representing the system operating mode, and the nodes representing the total power consumption of the thermal management subsystem.

[0021] The dynamic hypergraph model is constructed based on the operating condition coupling hyperedge, the control response hyperedge, and the corresponding nodes with related relationships.

[0022] Specifically, the process of obtaining the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions includes:

[0023] Based on the target power value, the real-time state of charge, and the real-time temperature of the battery cluster, an online parameterized loss generation model is invoked to generate a baseline loss power value under the current operating condition; wherein, the online parameterized loss generation model uses recursive least squares identification with the actual total heat source power feedback value of the previous moment and the current input parameters, and updates its internal polynomial coefficients in real time.

[0024] Based on the real-time output power of the power conversion subsystem, the target power value, and the current switching frequency and modulation ratio obtained in real time by the power conversion subsystem controller, the instantaneous conversion efficiency at the current operating point is obtained by bilinear interpolation through a pre-constructed two-dimensional efficiency cloud map, and an efficiency compensation factor is constructed.

[0025] Specifically, the process of obtaining the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions also includes:

[0026] Based on the real-time temperature of the battery cluster and the current battery health status value obtained through the battery management subsystem, the preset loss compensation surface that integrates aging and temperature characteristics is queried to obtain the compensation coefficient.

[0027] Based on the dynamically generated baseline loss power value, the efficiency compensation factor, and the compensation coefficient, a weighted fusion is performed using a Kalman filter to obtain the optimal estimate of the instantaneous total heat source power.

[0028] Specifically, obtaining the average temperature of the battery cluster and the junction temperature of the H-bridge over a predetermined future time period includes:

[0029] Based on the pre-calibrated thermal resistance-thermal capacity parameter network, the topology of the thermal resistance-thermal capacity parameter network is obtained. The topology includes the position of each thermal capacity node, the connection relationship of each thermal resistance node, and the injection position of each heat source node.

[0030] Based on the topology and the thermal capacity value of each thermal capacity node and the thermal resistance value of each thermal resistance node in the thermal resistance-thermal capacity parameter network, a set of thermal balance differential equations with the temperature of each thermal capacity node as the state variable is constructed.

[0031] The thermal equilibrium differential equations are transformed into the standard form of state-space equations to obtain state matrix A, input matrix B, output matrix C, and direct transfer matrix D. The elements of state matrix A are composed of the reciprocals of the thermal capacity and thermal resistance, used to characterize the thermal coupling strength between each thermal capacity node. The elements of input matrix B characterize the heat injection relationship between each heat source node and its corresponding thermal capacity node. Output matrix C is used to extract the average temperature of the battery cluster and the H-bridge junction temperature from all state variables. Direct transfer matrix D is set to a zero matrix.

[0032] Specifically, obtaining the average temperature of the battery cluster and the junction temperature of the H-bridge over a predetermined period of time also includes:

[0033] Based on the target power value in the power scheduling task instruction and the system operation mode, the preset heat generation allocation ratio table is queried to obtain the battery heat generation allocation coefficient and the H-bridge heat generation allocation coefficient.

[0034] Based on the battery heat generation distribution coefficient and the H-bridge heat generation distribution coefficient, the instantaneous total heat source power is decomposed into battery heat source power and H-bridge heat source power.

[0035] Based on the battery heat source power and the H-bridge heat source power, an input vector u is constructed according to the heat injection relationship defined by the input matrix B, where each element of the input vector u corresponds to the injection power of a heat source node at the current moment;

[0036] Based on the average temperature of the battery cluster and the real-time junction temperature of the H-bridge, optimal estimation is performed using a Kalman filter to obtain the state estimates of the temperature of all thermal capacity nodes in the thermal resistance-thermal capacity parameter network.

[0037] Specifically, obtaining the average temperature of the battery cluster and the junction temperature of the H-bridge over a predetermined period of time also includes:

[0038] Based on the state estimate, the initial values ​​of the state vector are constructed;

[0039] Based on the state-space equation, a numerical integration algorithm is used, with the initial value as the initial condition and the input vector u as the input, to calculate and obtain the change sequence of the state vector within the future preset time period;

[0040] Based on the change sequence of the state vector, a linear transformation is performed through the output matrix C to extract the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H bridge within the future preset time period;

[0041] The average temperature of the battery cluster and the real-time junction temperature of the H-bridge are obtained by real-time monitoring and compared with the predicted value at the corresponding time in the theoretical temperature rise trajectory to calculate and obtain the temperature prediction error sequence.

[0042] Based on the temperature prediction error sequence, a preset parameter adaptive update algorithm is invoked to fine-tune at least one parameter in the thermal resistance-heat capacity parameter network online, thereby obtaining the updated thermal resistance-heat capacity parameter network parameters; wherein, the parameter adaptive update algorithm is a recursive least squares algorithm or an extended Kalman filter algorithm.

[0043] Based on the updated thermal resistance-heat capacity parameter network parameters, the theoretical temperature rise trajectory extraction process is re-executed to obtain the updated theoretical temperature rise trajectory, and the updated theoretical temperature rise trajectory is used as the final output theoretical temperature rise trajectory for the subsequent construction of the dynamic hypergraph model.

[0044] A smart power consumption optimization system for grid-based energy storage includes:

[0045] The data acquisition module obtains operational status monitoring data, external dispatch commands, and system operation modes of the grid-type energy storage cabinet;

[0046] The construction module, based on the operation status monitoring data, scheduling instructions and system operation mode, constructs a dynamic hypergraph model to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand and thermal management energy consumption in the grid-type energy storage cabinet;

[0047] The optimization solution module takes minimizing the total operating energy consumption of the grid-type energy storage cabinet as the optimization objective, and performs collaborative optimization solution based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles.

[0048] The execution module executes the cooperative operation strategy on the preset twin energy storage cabinet to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet.

[0049] Compared with the prior art, the beneficial effects of the present invention are:

[0050] This invention addresses the shortcomings of existing technologies by constructing and applying a dynamic hypergraph collaborative optimization model. This systematically solves the overall energy efficiency bottleneck caused by the fragmented optimization of power scheduling and thermal management in traditional grid-type energy storage cabinets. Through refined modeling such as online parameterized loss generation, two-dimensional efficiency cloud map interpolation, and compensation surfaces that integrate aging and temperature, high-precision adaptive estimation of instantaneous total heat source power is achieved. Furthermore, a thermal resistance-capacitance network and state-space equations are used to accurately predict the theoretical temperature rise trajectory, which is then encapsulated as a coupled hyperedge of associated commands, states, and temperatures. Simultaneously, based on the cooling energy efficiency characteristic curve, the theoretical cooling power consumption sequence required to offset the temperature rise is calculated and encapsulated as an associated temperature and model... The core innovation lies in the regulation and response of power consumption to hyperedges. By dynamically quantifying the importance of these heterogeneous coupling relationships through a hypergraph attention mechanism, and using this as a basis for global collaborative optimization, a power and thermal management collaborative control strategy that minimizes total operating energy consumption is generated. This enables the thermal management subsystem to shift from passively responding to temperature to actively coordinating power planning, significantly reducing ineffective energy consumption caused by response lag or overcooling while ensuring safety. Finally, with the help of online rolling updates and offline training mechanisms of the model, the system gains the ability to adapt to equipment aging, environmental changes, and scheduling needs throughout its entire life cycle, achieving a dual improvement in systemic energy saving and consumption reduction as well as operational reliability. Attached Figure Description

[0051] Figure 1 This is a flowchart of a smart power consumption optimization method for grid-type energy storage according to the present invention;

[0052] Figure 2 This is a block diagram of an intelligent power consumption optimization system for grid-type energy storage according to the present invention. Detailed Implementation

[0053] To facilitate understanding of the background and application scope of this application, the following description of specific implementation methods is provided in conjunction with typical grid-type energy storage system operation scenarios.

[0054] In the tasks of frequency regulation, voltage regulation, and inertia support for power systems with a high proportion of renewable energy, the operation of grid-based energy storage cabinets exhibits complex characteristics of strong dynamics, multiple objectives, and high safety requirements: within the same dispatch cycle, they need to respond quickly to the power command from the grid to maintain stability, while ensuring that the temperature of the internal battery clusters and power devices (such as H-bridges) remains within safe limits; the output and losses of the power conversion subsystem change drastically, directly driving transient changes in thermal load, while the response of the thermal management subsystem is constrained by ambient temperature, equipment efficiency, and operating mode. In practice, operators typically expect energy storage systems to minimize their total operating energy consumption while meeting all grid functional and safety constraints. The aforementioned complexity is amplified by a common phenomenon: there is a strong coupling between the heat loss generated by power dispatch and the cooling energy consumption required for thermal management, but the control decisions for both are often fragmented in traditional architectures or only involve simple end-point linkage, resulting in implicit conflicts and energy waste in terms of time scale and optimization objectives.

[0055] Existing methods generally employ two approaches: First, power dispatch is prioritized, and after satisfying grid commands and electrical safety constraints, the thermal management subsystem performs independent closed-loop control based on measured temperature. Second, a simplified joint optimization model is constructed to attempt global optimization under fixed efficiency assumptions and a static thermal model. The former often results in passive response, delayed cooling, or overly conservative thermal management when faced with rapidly changing power commands, commonly exhibiting temperature control oscillations, response delays, and ineffective energy consumption. While the latter possesses theoretical synergistic potential, its model accuracy, computational complexity, and online adaptability are insufficient to meet real-time control requirements when dynamic nonlinearities and uncertainties such as equipment loss characteristic drift, aging effects, and environmental disturbances are superimposed. Furthermore, it struggles to achieve stable replanning under rolling dispatch and feedback deviations.

[0056] The common dilemma is that once the problem of co-optimization of strongly coupled electrothermal systems is postponed to the control end or relies on static simplified models, the power planning and thermal management strategies at the front end have become fixed or inaccurate. No matter how much local adjustment is made afterward, it is difficult to avoid the risks of excessive safety margin, insufficient energy efficiency potential, or dynamic mismatch.

[0057] The core processing logic of this application is based on the aforementioned common dilemma. It does not rely on pre-fixed efficiency curves or static thermal resistance-capacity parameter network models. Instead, it first identifies the heat source power caused by power scheduling as the key coupling variable connecting the electrical and thermal sides, and pre-defines high-precision, adaptive estimation of this variable as the triggering factor and anchor point for collaborative solution. To this end, after acquiring real-time monitoring data, scheduling instructions, and operating modes, a dynamic hypergraph model is constructed, using various states and instructions as nodes and "operating condition coupling" and "regulation response" as hyperedges. This transforms the instantaneous total heat source power from a passive loss result into an active collaborative anchor point. Around this anchor point, a forward and backward quantization mapping is triggered: forward, through the online updated thermal resistance-capacity parameter network state space equation, the anchor point (heat source power) is mapped to the theoretical temperature rise trajectory of the battery and H-bridge, encapsulated as an operating condition coupling hyperedge; backward, through the cooling energy efficiency characteristic curve and operating mode constraints, the heat load required to offset this temperature rise is mapped to a theoretical cooling power consumption sequence, encapsulated as a regulation response hyperedge. Unlike conventional optimization methods that prioritize electrical and then thermal approaches or simple joint optimizations, this embodiment places "high-precision anchor point prediction - bidirectional relationship quantification - hypergraph weight dynamic learning - global collaborative solution - feedback verification and update" in a rolling closed loop. After each generation of a collaborative strategy, the model parameters are simultaneously fine-tuned online using actual operational feedback, dynamically improving the accuracy of prediction and optimization in the next cycle, thus avoiding long-term performance degradation or control risks caused by model mismatch.

[0058] It is important to note that, due to the stringent requirements for engineering reliability and global optimality, this application employs a weight learning and closed-loop re-optimization logic based on a dynamic hypergraph attention mechanism. Under different operating modes and conditions, the prioritization of temperature rise safety and cooling energy consumption naturally differs. Simple fixed-weight optimization can easily lead to policy rigidity and an inability to adapt to dynamic changes. In this embodiment, after the hypergraph model is constructed, a hypergraph attention network is introduced to dynamically learn and weight the relationship strength between nodes and hyperedges, enabling the optimization objective and constraints to adaptively adjust with the system state. Simultaneously, after the control command is executed, the deviation between the actual power, temperature, power consumption, and predicted values ​​is used as a trigger signal. Once a threshold is exceeded, immediate local re-optimization is initiated, ensuring that each control action closely matches the current actual system state and mitigating solution failures caused by disturbances or model errors. Furthermore, the online optimization results of total operating energy consumption and long-term historical data serve as feedback for offline model training, ensuring that the entire collaborative system continuously evolves during its rolling process, guaranteeing stable, efficient, and safe operation throughout its entire lifecycle.

[0059] It is important to emphasize that the above description is not an exhaustive list of algorithmic details, but rather clarifies the technical starting point and reproducible implementation boundaries of this embodiment: Given the availability of real-time data, scheduling instructions, and operating modes, and the ability to calibrate key equipment losses and thermal characteristics through experiments and online learning, this embodiment constructs a hypergraph model with instantaneous total heat source power as a dynamic anchor point. Based on this model, it achieves global collaborative optimization and closed-loop adaptive control of power scheduling and thermal management strategies. This fundamentally resolves multiple challenges in existing technologies, including electrothermal separation control, model mismatch, insufficient dynamic adaptability, and global energy efficiency bottlenecks, providing a unified solution framework for the efficient, reliable, and intelligent operation of grid-based energy storage systems. Please refer to [link to relevant documentation]. Figure 1 The present invention provides an embodiment of a smart power consumption optimization method for grid-based energy storage, comprising the following steps:

[0060] S1. Obtain operational status monitoring data, external dispatch instructions, and system operation modes of the grid-type energy storage cabinet;

[0061] S2. Based on the operating status monitoring data, scheduling instructions and system operating mode, construct a dynamic hypergraph model to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand and thermal management energy consumption in the grid-type energy storage cabinet;

[0062] S3. Taking the minimum total operating energy consumption of the grid-type energy storage cabinet as the optimization objective, perform collaborative optimization based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles.

[0063] S4. On the preset twin energy storage cabinet, execute the cooperative operation strategy to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet.

[0064] It should be further explained that the construction process of the dynamic hypergraph model in this embodiment includes:

[0065] S201. Based on real-time monitoring data from sensors within the grid-type energy storage cabinet, acquire the output power of the power conversion subsystem, the state of charge and temperature of the battery clusters, the temperature of the H-bridge, the temperature of the cabinet environment, and the real-time power consumption of the thermal management subsystem. It should be further noted that this embodiment synchronously acquires key electrical, thermal, and status parameters of the system operation through a multi-source sensor network. Specifically, the real-time output power of the power conversion subsystem is collected to calculate the external power flow and efficiency; the real-time state of charge and average temperature of the battery clusters are collected to assess the battery energy state and thermal safety; the real-time junction temperature of the H-bridge is collected to monitor the reliability of power devices; the real-time temperature of the cabinet environment is collected to assess heat dissipation conditions; and the real-time power consumption of the thermal management subsystem is collected to directly quantify temperature control energy consumption.

[0066] S202. Collect communication messages from the energy management subsystem and parse them to obtain the power scheduling task instruction of the grid-type energy storage cabinet. The power scheduling task instruction includes the target power value and the effective time window.

[0067] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the following will describe the process of a grid-type energy storage cabinet parsing power scheduling task instructions from communication messages, using a specific example:

[0068] Suppose that one morning, the energy management subsystem of a grid-type energy storage unit sends a standard dispatch command message via the IEC 104 protocol. The local controller of the energy storage unit first receives and verifies the integrity of the message, confirming that its frame structure, checksum, and protocol version conform to the established standard.

[0069] Subsequently, the controller parses the message data field according to the format definition of the IEC104 protocol. For example, in the "body address" and "information element" fields of the message, the controller extracts the value representing the power command. Assuming the parsed target power value is 500 kilowatts, this value indicates that the power grid requires the energy storage cabinet to output 500 kilowatts of power.

[0070] Meanwhile, the controller parses the timestamp field of the instruction from the same message. Assuming the parsed instruction's effective time is 10:00:00 and its expiration time is 10:15:00, this means the effective time window for this scheduled task is 15 minutes, starting at 10:00 AM.

[0071] After analysis, the target power value of 500 kW will serve as the core attribute of an "instruction node" in the dynamic hypergraph model. This node will become the starting point for all subsequent calculations: on the one hand, it is directly used to determine the planned output baseline of the power conversion subsystem in the future; on the other hand, it is also a key input for calculating the instantaneous total heat source power that the system may generate when performing this task, and thus it is associated with the "condition-coupled hyperedge" for predicting the temperature rise of the battery and H-bridge.

[0072] The extracted effective time window (10:00:00 to 10:15:00) precisely defines the prediction period for this collaborative optimization calculation. The optimization algorithm will collaboratively plan the power output and cooling consumption for each future step within this 15-minute timeframe, ensuring that the generated operating strategy is completely synchronized with the grid dispatch instructions in the time dimension, neither prematurely ending the task nor unnecessarily extending its operation. Through this precise message parsing and information injection, external high-level dispatch instructions are accurately converted into specific input parameters driving the internal intelligent collaborative optimization model.

[0073] S203. Based on the local controller status of the network-type energy storage cabinet or the mode identifier bit in the communication message, obtain the current system operating mode;

[0074] It should be further explained that the process of obtaining the current system operating mode in this embodiment includes:

[0075] The operating mode identifier can be read from the internal status register of the local controller of the grid-type energy storage cabinet; or the operating mode identifier can be extracted from a predefined field in the data field of the communication message.

[0076] Based on a preset operation mode encoding mapping table, the read or extracted operation mode identifier is parsed into the corresponding system operation mode. The system operation mode is used to characterize the macroscopic working stage of the grid-type energy storage cabinet. It is input into the dynamic hypergraph model as a mode node attribute, directly affecting the construction of the control response hyperedge and the constraints of the optimization problem. Specifically, different system operation modes correspond to different thermal management strategy priorities, equipment availability combinations, and energy consumption optimization objectives. For example, the system operation modes of the grid-type energy storage cabinet include grid-connected charging mode, grid-connected discharging mode, frequency regulation mode, off-grid support mode, economic dispatch mode, and standby maintenance mode. In this embodiment, the setting of each mode is achieved by defining its unique optimization objective priority and key constraints. Taking the economic dispatch mode as an example, this mode is set with the minimum total system operating cost as the core optimization objective. Its constraints allow the thermal management subsystem to operate with the minimum necessary power while meeting the upper temperature limit. At the same time, the optimization algorithm will smoothly adjust the output of the power conversion subsystem within the time window of meeting the grid dispatch command to take advantage of the low electricity price, thereby generating a coordinated operation strategy that takes into account electricity cost and self-loss on a long-term scale.

[0077] S204. Based on the target power value in the power scheduling task instruction, the real-time state of charge of the battery cluster, and the real-time output power of the power conversion subsystem, the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions is obtained by querying a preset loss-power-state of charge mapping table.

[0078] It should be further explained that the process of obtaining the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions in this embodiment includes:

[0079] S2041. Based on the target power value, the real-time state of charge, and the real-time temperature of the battery cluster, call the online parameterized loss generation model to generate the reference loss power value under the current operating condition; wherein, the online parameterized loss generation model uses the feedback value of the actual total heat source power at the previous moment and the current input parameters to perform recursive least squares identification and update its internal polynomial coefficients in real time.

[0080] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the steps in this embodiment to generate the reference loss power value under the current operating condition include:

[0081] Using target power, real-time state of charge, and real-time battery cluster temperature as input variables, a polynomial regression structure is constructed, comprising first- to third-order monomials of each variable and pairwise cross terms, forming the basic framework of the online parameterized loss generation model. Five hundred sets of historical time data covering different operating conditions are extracted from the historical operating database. The target power range is set to 0 to 200 kW, the real-time state of charge ranges to 0.2 to 0.9, and the real-time battery cluster temperature ranges to 15 to 45°C. These parameter data sets are integrated into a historical input sample set. Simultaneously, the actual total heat source power feedback values, corresponding one-to-one with the historical input sample times and obtained from sensor measurements or energy conservation inverse calculations, are extracted to construct a historical output sample set. Based on the historical input and output sample sets, regression calculations are performed using the publicly available batch least squares method to solve for the initial polynomial coefficient vector of the model and the corresponding initial error covariance matrix, ultimately completing the offline initialization of the online parameterized loss generation model.

[0082] Secondly, during the online operation of the system, the online adaptive parameter update step of the model is executed; this embodiment is triggered within each sampling period, and specifically includes the following implementation process:

[0083] A1. Data preparation and update trigger judgment: After determining whether it is the initial sampling moment based on the system clock, if so, obtain the actual total heat source power feedback value, target power value, real-time state of charge, and real-time temperature of the battery cluster from the system cache of the previous sampling period; based on these obtained values, construct the input vector of the previous moment according to the structure defined by the model, and explicitly use the actual total heat source power feedback value of the previous moment as the output observation value, wherein the system sampling period is set to 10 seconds.

[0084] A2 dynamically determines the forgetting factor based on the current operating mode identifier obtained from the system monitoring unit, the power command change rate calculated from the current and previous target power values, and the ambient temperature change rate calculated from ambient temperature sensor readings. Specifically, it performs lookup and interpolation calculations based on a preset adaptive rule mapping table to obtain an adaptive forgetting factor value specific to the current sampling time. The adaptive rule mapping table is pre-calibrated through experimental analysis and system optimization. Its core rule is configured as follows: when the system operates in an operating mode with frequent power command changes, or when the calculated power command change rate exceeds the first threshold of 15% of the rated power, a smaller forgetting factor value in the range of 0.95 to 0.98 is output, aiming to enable the model to quickly track new operating conditions; when the system operates in a stable power operating mode, or when the power command change rate is lower than the second threshold of 5% of the rated power, a larger forgetting factor value in the range of 0.990 to 0.999 is output, aiming to enhance the model's memory of long-term steady-state characteristics and suppress measurement noise.

[0085] A3 maintains a first-in, first-out (FIFO) data cache queue in memory as a sliding window, based on a preset fixed window length of 100 groups. This sliding window stores the most recent groups of valid data consisting of input vectors and output observations in chronological order. In the current period, the latest data pair constructed in A2 is added to the end of this sliding window. Subsequently, the total number of data pairs stored in the sliding window is checked; if it exceeds the fixed window length, the oldest data pair is automatically deleted from the head of the window to ensure that only data from the most recent period is retained within the window.

[0086] A4, based on the input vector and output observations from the previous time step obtained in A2, the adaptive forgetting factor for the current time step calculated in A3, and the stored model parameters from the previous time step (i.e., the polynomial coefficient vector and the error covariance matrix), calls the existing publicly available recursive least squares algorithm with a forgetting factor to perform calculations. The initial gain vector in the recursive formula is set to the identity matrix, and the initial prediction error threshold is set to 0.01kW. This calculation process sequentially calculates the gain vector and prediction error, and finally updates the polynomial coefficient vector and the error covariance matrix, thereby obtaining the updated model parameters applicable to the current time step.

[0087] A5, while updating model parameters, simultaneously monitors signals from device status flags, battery management system messages, and external command interfaces. When a predefined event characterized by a step change in system loss characteristics is detected, a model reset process is immediately triggered. These events include: a flag change upon power-on after device shutdown for maintenance; a decrease in battery health status estimate reported by the battery management system exceeding 3% within five consecutive sampling periods; or receiving a clear reset command. The reset process includes: immediately clearing the entire sliding data window maintained in A3; subsequently, the system automatically enters a temporary state, waiting for new valid data to refill the sliding window. When the amount of data within the window reaches the minimum sample size of 50 required for batch processing calculation, a batch least squares calculation process, similar to offline initialization, is automatically triggered. Based on the new data within the window, a completely new initial polynomial coefficient vector and error covariance matrix are calculated, and these are used to completely replace the currently recursively maintained model parameters, thus completing the model reinitialization and preventing old data from contaminating the model in the new state.

[0088] Based on the target power value, real-time state of charge, and real-time temperature of the battery cluster obtained from the real-time data bus at the current sampling moment, the current input vector is constructed according to the model structure. The current input vector is then multiplied by the latest polynomial coefficient vector obtained through the online adaptive parameter update step to output the baseline loss power estimate under the current operating conditions. The motivation for this step is to overcome the inherent limitation of traditional static loss models in adapting to the dynamic operating conditions of grid-type energy storage cabinets. By constructing and updating a parameterized loss generation model online, the core purpose of this embodiment is to utilize real-time system feedback data to enable the loss estimation model to continuously track and learn the loss characteristic drift caused by equipment aging, environmental changes, and different load characteristics. This elevates the estimation of the baseline loss power value from relying on fixed empirical formulas or tables to a dynamic estimation process that can autonomously evolve and dynamically match the current actual loss characteristics of the system.

[0089] S2042. Based on the real-time output power of the power conversion subsystem, the target power value, and the current switching frequency and modulation ratio obtained in real time by the power conversion subsystem controller, bilinear interpolation is performed using a pre-constructed two-dimensional efficiency cloud map to obtain the instantaneous conversion efficiency at the current operating point, and an efficiency compensation factor is constructed.

[0090] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, in this embodiment, the process of obtaining the instantaneous conversion efficiency at the current operating point by performing bilinear interpolation through a pre-constructed two-dimensional efficiency cloud map includes: based on a laboratory test platform, performing full-condition efficiency calibration on the power conversion subsystem to obtain a raw efficiency dataset covering the load rate from 0 to 100% and the switching frequency from 5kHz to 20kHz. The input voltage of the test platform is stabilized at DC 600V, and the output voltage range is set to AC 380V; based on the raw efficiency dataset, extracting each test... The test points use a first-dimensional parameter and a second-dimensional parameter. The first-dimensional parameter is the ratio of the output power to the rated power of the power conversion subsystem at each test point, used to characterize the load rate. The load rate test step size is set to 5%, and the rated power is set to 500kW. The second-dimensional parameter is the switching frequency corresponding to each test point, used to characterize the switching loss level of the power device. The switching frequency test step size is set to 1kHz. Based on the extracted first-dimensional and second-dimensional parameters, a continuous efficiency surface is constructed using a Gaussian radial basis function surface fitting algorithm, with the load rate as the abscissa, the switching frequency as the ordinate, and the instantaneous conversion efficiency as the elevation. The penalty factor of the fitting algorithm is set to 0.001, and the iterative convergence threshold is set to... Based on the continuous efficiency surface, discrete sampling is performed at a set load rate sampling interval of 2% and a switching frequency sampling interval of 0.5kHz to obtain a two-dimensional efficiency lookup table composed of discrete efficiency data points, i.e., the pre-constructed two-dimensional efficiency cloud map. This cloud map contains 51 load rate sampling points and 31 switching frequency sampling points, totaling 1581 discrete efficiency data points. Based on the real-time output power of the power conversion subsystem and the rated power corresponding to the target power value, the current real-time load rate is calculated. Based on the power conversion subsystem controller, the current switching frequency is obtained through the CAN bus with a communication cycle of 100ms. Based on the current real-time load rate and the current switching frequency, the grid cell in the two-dimensional efficiency cloud map is located, and the coordinates of the four vertices of the grid cell and their corresponding known efficiency values ​​are obtained. Based on the normalized offset of the current real-time load rate and the current switching frequency relative to the lower left vertex of the grid cell, and the known efficiency values ​​of the four vertices, bilinear interpolation is performed, retaining five significant decimal places during the interpolation calculation, and finally obtaining the instantaneous conversion efficiency of the current operating point.

[0091] The motivation behind this process is to address the problem of insufficient accuracy in estimating power conversion losses under dynamic operating conditions caused by traditional efficiency models that use fixed average efficiency or single-variable lookup tables. Its core principle is to pre-construct a two-dimensional continuous efficiency model with load rate and switching frequency as independent variables through full-condition experimental calibration and surface fitting, thus elevating discrete measured data into a continuously analyzable mathematical surface. Then, during real-time operation, a bilinear interpolation algorithm is used to leverage the linear characteristics of this surface in local regions, based on the current real-time load rate and switching frequency, to quickly and accurately calculate the corresponding instantaneous conversion efficiency. This method, through a combination of scientific experiments and mathematical modeling, achieves a refined and adaptive characterization of the efficiency of the power conversion subsystem, a key factor influencing losses.

[0092] S2043. Based on the real-time temperature of the battery cluster and the current battery health status value obtained through the battery management subsystem, query the preset loss compensation surface that integrates aging and temperature characteristics, and obtain the compensation coefficient.

[0093] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility of the calculation and does not represent the actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the process of obtaining the compensation coefficient by querying the preset loss compensation surface that integrates aging and temperature characteristics in this embodiment is as follows: Based on long-term historical data, a historical dataset covering the entire battery life cycle and typical temperature conditions is obtained. The dataset size is 2000 sets of historical data points. Each historical data point contains the battery health status value at the historical time, the battery temperature value at the historical time, and the true value of the historical loss compensation coefficient obtained by back-calculating the actual total heat source power at the corresponding historical time. The battery health status value ranges from 0.6 to 1.0, and the battery temperature value ranges from -10℃ to At 55℃, the true value of the historical loss compensation coefficient ranges from 0.85 to 1.35. Based on this historical dataset, an input matrix and output vector are constructed for surface fitting. Each row of the input matrix corresponds to a historical data point, containing the battery health status value and battery temperature value as two dimension parameters. The input matrix has a dimension of 2000×2, and the output vector is a 2000×1 column vector, with each element corresponding to the true value of the historical loss compensation coefficient for a historical data point. Based on the input matrix and output vector, a continuous compensation surface function is constructed using a bivariate cubic polynomial regression fitting algorithm. This function uses the battery health status value as the first independent variable, the battery temperature value as the second independent variable, and the loss compensation coefficient as the dependent variable. The regularization parameter of the fitting algorithm is set to 0.0001, and the iteration termination condition is that the difference between two adjacent fitting errors is less than 0.0001. The fitting error is calculated using the mean square error. Based on the continuous compensation surface function, discrete sampling is performed at a set battery health status value sampling interval of 0.01 and battery temperature value sampling interval of 1℃. A two-dimensional lookup table composed of discrete compensation coefficient data points is obtained, which is the preset loss compensation surface integrating aging and temperature characteristics. This surface contains 41 battery health status sampling points, increasing from 0.6 in steps of 0.01 to 1.0, and 66 battery temperature sampling points, increasing from -10℃ in steps of 1℃ to 55℃, for a total of 2706 discrete compensation coefficient data points. Based on the battery health status estimation results reported by the battery management subsystem, the current battery health status value is obtained with an update cycle of 500ms, and the estimation accuracy is retained to three decimal places. Based on the temperature measurement values ​​of 12 evenly distributed sampling points within the battery cluster, through... The arithmetic mean algorithm is used to calculate the average value to obtain the real-time temperature value of the battery cluster, with a temperature measurement accuracy of ±0.1℃. Based on the current battery health status value and the real-time temperature value, the grid cell in which the battery cluster is located is located in the preset loss compensation surface. The coordinates of the four vertices of the grid cell and their corresponding known compensation coefficient values ​​are obtained. The coordinate positioning accuracy is retained to three decimal places, and the accuracy of the known compensation coefficient values ​​is retained to four decimal places. Based on the normalized offset of the current battery health status value and the real-time temperature value relative to the lower left vertex of the grid cell and the known compensation coefficient values ​​of the four vertices, bilinear interpolation is performed. The accuracy of the normalized offset calculation is retained to four decimal places, the intermediate calculation results during the interpolation process are retained to five decimal places, and the accuracy of the final output compensation coefficient is retained to four decimal places. This is how the compensation coefficient is obtained.

[0094] The motivation behind this process is to overcome the long-term prediction bias caused by neglecting the coupling effect of battery aging and temperature in traditional heat loss models. Its basic principle is to quantify the coupled influence of two key factors—battery health status and operating temperature—on heat generation characteristics into a continuous mathematical surface through historical operational data inversion. During actual operation, based on the current real-time battery health status and temperature, accurate compensation coefficients are obtained from this pre-defined surface through querying and bilinear interpolation. This method unifies the slow time-varying factors of aging and the fast time-varying factors of temperature within a dynamic correction framework, enabling the estimation of battery heat generation power to adapt to the performance degradation throughout the battery's entire life cycle and efficiency changes at different operating temperatures. This provides highly accurate and long-term adaptable key correction parameters for estimating the core heat source power.

[0095] S2044. Based on the dynamically generated baseline loss power value, the efficiency compensation factor, and the compensation coefficient, the instantaneous total heat source power is obtained by weighted fusion through a Kalman filter.

[0096] It should be further explained that the weighted fusion process using a Kalman filter in this embodiment includes:

[0097] B1. Based on the dynamically generated baseline loss power value, efficiency compensation factor, and compensation coefficient, the preliminary fusion value of the instantaneous total heat source power is obtained through multiplication.

[0098] B2. Based on the optimal estimate of the instantaneous total heat source power at the previous moment, the state prediction value of the instantaneous total heat source power at the current moment is calculated using a state transition model. The accuracy of the state prediction value calculation is retained to one decimal place.

[0099] B21. Based on the preset state transition equation, define the state transition model of the system. The system's state variable is the instantaneous total heat source power. The state transition model is expressed as the current state prediction equals the previous optimal state estimate plus the process noise. The state transition matrix of the state transition equation is set to a 1×1 dimension identity matrix, and the process noise follows a Gaussian distribution with a mean of 0 and a variance of 0.01.

[0100] B22. Based on the preset observation equations, define the system's observation model. The system's observation variable is the preliminary fused value of the instantaneous total heat source power. The observation model is expressed as the current observed value equals the current true state value plus the observation noise. The observation matrix of the observation equations is set as a 1×1 dimension identity matrix, and the observation noise follows a Gaussian distribution with a mean of 0 and a variance of 0.001.

[0101] B3. Based on the preliminary fusion value and the state prediction value, a weighted fusion is performed using a Kalman filter to obtain the optimal estimate of the instantaneous total heat source power at the current moment. The process is as follows:

[0102] B31. Based on the state prediction value, Kalman gain, and the difference between the preliminary fusion value and the state prediction value, calculate the optimal estimate of the instantaneous total heat source power at the current moment.

[0103] B32. Calculate the Kalman gain at the current time based on the error covariance prediction of the state prediction value and the preset observation noise covariance; the initial error covariance matrix of the Kalman filter is set to a 1×1 dimension matrix with an initial value of 10; the observation noise covariance matrix is ​​set to a 1×1 dimension matrix with a value of 0.001; the Kalman gain calculation precision is retained to four decimal places.

[0104] B33. Based on the Kalman gain and the error covariance prediction of the state prediction, update the error covariance of the current time step to obtain the optimal state estimate, and use it for the calculation of the next time step; the error covariance update calculation precision is retained to four decimal places, and the updated error covariance matrix is ​​1×1 dimension.

[0105] S205. Based on the instantaneous total heat source power, the average temperature of the battery cluster, the real-time junction temperature of the H-bridge, and the thermal resistance-capacity parameter network pre-calibrated according to the physical structure of the energy storage cabinet, the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge in a future preset time period is obtained by solving the state space equation corresponding to the thermal resistance-capacity parameter network. The theoretical temperature rise trajectory constitutes a working condition coupling hyperedge in the dynamic hypergraph model. The nodes associated with the working condition coupling hyperedge include: a command node representing the target power value, a node representing the real-time state of charge, a node representing the average temperature of the battery cluster, and a node representing the junction temperature of the H-bridge.

[0106] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the process of obtaining the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge within a preset future time period in this embodiment is as follows:

[0107] S2051. Based on a pre-calibrated thermal resistance-thermal capacity parameter network, obtain the topology of the thermal resistance-thermal capacity parameter network. The topology includes the location of each thermal capacity node, the connection relationship of each thermal resistance node, and the injection location of each heat source node. The nodes and their functions in the thermal resistance-thermal capacity parameter network of this embodiment include: thermal capacity nodes are used to characterize physical components with significant heat storage capacity in a grid-type energy storage cabinet. Their node values ​​represent the real-time temperature of the component, such as a battery module or H-bridge substrate, and are the core state variables to be solved in the state-space equations; thermal resistance nodes are used to characterize the heat conduction path between physical components or between a component and the environment. Their node values ​​represent the thermal resistance of this path and are used to construct the thermal coupling relationship between each thermal capacity node in the state matrix A. Heat source nodes are used to characterize the heat injection location. Their node values ​​represent the heat source power converted from power loss, and are mapped to the heat input to a specific thermal capacity node through the input matrix B, serving as the source term driving the change in the thermal state of the system. The thermal resistance-capacity parameter network in this embodiment is a lumped-parameter thermal resistance-capacity parameter network model pre-established based on the physical structure of the energy storage cabinet. Its purpose is to abstract and simplify the complex three-dimensional heat distribution characteristics of the grid-type energy storage cabinet into a mathematical model composed of a finite number of heat capacity, thermal resistance, and heat source elements connected by topology. The thermal resistance-capacity parameter network is the physical basis for constructing the state-space equations to predict the theoretical temperature rise trajectory. The accuracy of its parameters, such as thermal resistance and heat capacity, directly determines the prediction accuracy of the theoretical temperature rise trajectory in this embodiment. The specific parameters are set as follows: the thermal capacity value of the battery module thermal capacity node is 800J / ℃, with one thermal capacity node corresponding to each battery module, for a total of 16 battery module thermal capacity nodes; the thermal capacity value of the H-bridge substrate thermal capacity node is 200J / ℃, with one thermal capacity node corresponding to each H-bridge, for a total of 8 H-bridge substrate thermal capacity nodes; the thermal resistance node between battery modules is 0.05℃ / W, the thermal resistance node between the battery module and the H-bridge substrate is 0.1℃ / W, and the thermal resistance node between each component and the environment is 0.2℃ / W.

[0108] S2052. Based on the topology and the heat capacity value of each heat capacity node and the heat resistance value of each heat resistance node in the thermal resistance-heat capacity parameter network, and according to the law of conservation of energy and combined with the principle of thermal-electric analogy, a set of thermal balance differential equations with the temperature of each heat capacity node as the state variable is constructed. This embodiment constructs the set of thermal balance differential equations through the following steps: Based on the node composition characteristics of the thermal resistance-heat capacity network, each heat capacity node in the thermal resistance-heat capacity network is determined as an independent analysis node. For any independent analysis node, the heat flow through each connected heat resistance node is obtained based on the basic heat flow calculation rules. The basic heat flow calculation rules are to divide the temperature difference between the two ends of the corresponding heat resistance node by the heat resistance value of the heat resistance node itself to obtain the heat flow value flowing through the heat resistance node; the internal energy change rate of the independent analysis node is obtained based on the quantitative correlation established by the law of conservation of energy. The internal energy change rate is obtained by using the heat capacity value and the temperature of the independent analysis node. The calculation result is obtained by multiplying the rate of change of temperature over time. Based on the principle of heat flow continuity, the net heat flow of each independent analysis node is obtained. This principle involves summing all heat flows into the independent analysis node and subtracting the sum of all heat flows out of the node; the difference is the net heat flow of that node. If a heat source directly acts on the independent analysis node, its power is included as an energy input in the calculation. Based on the equivalence between the rate of change of internal energy and the net heat flow of the independent analysis node, a differential equation with the temperature of the corresponding heat capacity node as the variable is obtained. Following this process of constructing the thermal balance differential equation, the following steps are performed sequentially for all heat capacity nodes in the thermal resistance-heat capacity network: determining the independent analysis node, calculating heat flow, obtaining the rate of change of internal energy, calculating net heat flow, and arranging the differential equations. The differential equations corresponding to all heat capacity nodes are then summarized and integrated to obtain a system of simultaneous thermal balance differential equations. The thermal equilibrium differential equations serve as the core dynamic model, used for the subsequent construction of the state-space equations and the dynamic prediction of the temperature of each heat capacity node in the thermal resistance-heat capacity network. The accuracy of heat flux calculations and the accuracy of internal energy change rate calculations are retained to three decimal places.

[0109] S2053. Transform the thermal equilibrium differential equations into the standard form of state-space equations to obtain the state matrix A, input matrix B, output matrix C, and direct transfer matrix D. The state matrix A is a 24×24 matrix, whose elements are composed of the reciprocal of the thermal capacity value and the reciprocal of the thermal resistance value, used to characterize the thermal coupling strength between each thermal capacity node; the input matrix B is a 24×2 matrix, whose elements are used to characterize the heat injection relationship between each heat source node and the corresponding thermal capacity node. Specifically, the elements corresponding to the 16 battery module thermal capacity nodes in the column corresponding to the battery heat source node have a value of 1, and the rest have a value of 0; the elements corresponding to the 8 H-bridge substrate thermal capacity nodes in the column corresponding to the H-bridge heat source node have a value of 1, and the rest have a value of 0; the output matrix C is a 2×24 matrix, used to extract the average temperature of the battery cluster and the H-bridge junction temperature from all state variables. The first row corresponds to the average temperature of the battery cluster, and the element values ​​corresponding to the 16 battery module thermal capacity nodes are all 1 / 16, and the rest have a value of 0; the second row corresponds to the H-bridge junction temperature, and the element values ​​corresponding to the 8 H-bridge substrate thermal capacity nodes are all 1 / 8, and the rest have a value of 0; the direct transfer matrix D is set as a 2×2 zero matrix.

[0110] S2054. Based on the target power value and system operation mode in the power scheduling task instruction, query the preset heat generation allocation ratio table to obtain the battery heat generation allocation coefficient and H-bridge heat generation allocation coefficient.

[0111] It should be further explained that the process of obtaining the battery heat generation distribution coefficient and the H-bridge heat generation distribution coefficient in this embodiment includes:

[0112] C1. Based on the thermal characteristic experimental data of the grid-type energy storage cabinet on the test platform, the original heat generation distribution dataset was obtained. Each data point in the original heat generation distribution dataset includes the system operation mode during the test, the test power value, and the true value of the battery heat generation ratio and the true value of the H-bridge heat generation ratio calculated by inverting the measured temperature data corresponding to the test point. The input voltage of the test platform was stable at DC 600V, and the output voltage was AC 380V. The system operation modes were divided into constant power charging mode, constant power discharging mode, and frequency-modulated standby mode. The three test power values ​​covered 0 to 500kW, with a test step size of 50kW. Each mode and each power was tested three times. The size of the original heat generation distribution dataset was: 3 modes × 11 power points × 3 repetitions, for a total of 99 data points. The temperature measurement accuracy was ±0.1℃.

[0113] C2. Based on the original heat generation and distribution dataset, the data is classified and grouped according to the system operation mode, namely, into three categories: constant power charging mode data group, constant power discharging mode data group, and frequency regulation standby mode data group.

[0114] C3. For each system operating mode, based on the test power value and the corresponding true value of the battery heat generation ratio within each group, a first continuous function relationship between the battery heat generation ratio and power is constructed under this mode through quadratic polynomial curve fitting; based on the test power value and the corresponding true value of the H-bridge heat generation ratio within each group, a second continuous function relationship between the H-bridge heat generation ratio and power is constructed under this mode through quadratic polynomial curve fitting. The regularization parameter of the fitting algorithm is set to 0.0001, and the iteration termination condition is that the difference between two adjacent fitting errors is less than... The fitting error is calculated using the mean square error, and the accuracy of the fitting result is retained to four decimal places.

[0115] C4. Based on the first and second continuous function relationships under all system operating modes, construct a two-dimensional query structure with the system operating mode as the first index and the power value as the second index. This structure is the preset heat generation allocation ratio table. The power value index interval is set to 1kW to cover the range of 0 to 500kW.

[0116] C5. Based on the system operating mode, locate the corresponding mode data branch in the preset heat generation distribution ratio table; based on the target power value, obtain the battery heat generation distribution coefficient and H-bridge heat generation distribution coefficient corresponding to the current power through query or interpolation calculation in the located mode data branch. The interpolation calculation adopts the linear interpolation method, and the calculation accuracy is retained to four decimal places.

[0117] C6. Based on the battery heat generation allocation coefficient and the H-bridge heat generation allocation coefficient, the instantaneous total heat source power is decomposed into battery heat source power and H-bridge heat source power. The core motivation of this step is to solve the problem of inaccurate prediction of fixed allocation models in the thermal management of grid-type energy storage cabinets caused by the dynamic change of the heat generation ratio of batteries and H-bridges with operating conditions. The principle is to establish a quantitative relationship between the heat generation ratio and the system operating mode and output power by inverting the experimental data of previous thermal characteristics; then, by using classification fitting and structured query methods, a function of the heat generation ratio changing continuously with power is constructed for different operating modes, and integrated into an allocation ratio table that can be quickly retrieved by mode and power value. This scheme transforms the complex and coupled physical heat generation mechanism into a mathematical mapping based on measured data that can be adaptively queried. Thus, in real-time applications, the heat source power allocation ratio of batteries and H-bridges can be dynamically and accurately determined according to the current operating mode and power command, providing a key and reliable input decomposition basis for subsequent accurate thermal resistance-thermal capacity parameter network simulation and global energy consumption co-optimization.

[0118] S2055. Based on the battery heat source power and the H-bridge heat source power, construct the input vector u according to the heat injection relationship defined in the input matrix B. The input vector u is a 2×1 dimension vector, where the first element corresponds to the battery heat source power and the second element corresponds to the H-bridge heat source power, with the power values ​​retained to one decimal place.

[0119] S2056. Based on the average temperature of the battery cluster and the real-time junction temperature of the H-bridge, optimal estimation is performed using a Kalman filter to obtain the state estimates of the temperature of all thermal capacity nodes in the thermal resistance-thermal capacity parameter network. The process noise covariance matrix of the Kalman filter is set as a 24×24-dimensional diagonal matrix with diagonal elements of 0.01, the observation noise covariance matrix is ​​set as a 2×2-dimensional diagonal matrix with diagonal elements of 0.001 and 0.001 respectively, and the initial error covariance matrix is ​​set as a 24×24-dimensional diagonal matrix with diagonal elements of 10.

[0120] S2057. Based on the state estimate, construct the initial value of the state vector. The state vector is a 24×1 dimension vector, and the initial value is the state estimate output by the Kalman filter. The temperature value is retained to two decimal places.

[0121] S2058. Based on the state-space equation, a fourth-order Runge-Kutta numerical integration algorithm is used. Initial values ​​are used as initial conditions, and the input vector u is used as input to obtain the sequence of changes in the state vector over a preset future time period. The preset future time period is set to 60 minutes, the integration step size is set to 10 seconds, and the input vector u remains constant within each integration step.

[0122] S2059. Based on the change sequence of the state vector, a linear transformation is performed through the output matrix C to extract the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge within a preset future time period. The temperature values ​​are retained to two decimal places, and the trajectory data points are spaced 10 seconds apart, for a total of 360 data points.

[0123] S20591. Based on the average temperature of the battery cluster and the real-time junction temperature of the H-bridge obtained from real-time monitoring, compare them with the predicted values ​​at the corresponding moments in the theoretical temperature rise trajectory to calculate the temperature prediction error sequence. The error calculation uses absolute error, with precision retained to two decimal places, and the length of the error sequence is consistent with the number of data points in the theoretical temperature rise trajectory.

[0124] S20592. Based on the temperature prediction error sequence, a preset parameter adaptive update algorithm is invoked to fine-tune at least one parameter in the thermal resistance-heat capacity parameter network online, obtaining the updated thermal resistance-heat capacity parameter network parameters. The parameter adaptive update algorithm adopts a recursive least squares algorithm, with a forgetting factor set to 0.99, an initial gain vector set to the identity matrix, and an iteration convergence threshold set to... The parameters for fine-tuning include the thermal resistance node values ​​between battery modules and the thermal capacitance node values ​​of the H-bridge substrate.

[0125] S20593. Based on the updated thermal resistance-heat capacity parameter network parameters, re-execute the theoretical temperature rise trajectory extraction process to obtain the updated theoretical temperature rise trajectory, and use the updated theoretical temperature rise trajectory as the final output theoretical temperature rise trajectory for subsequent construction of the dynamic hypergraph model.

[0126] S206. Based on the theoretical temperature rise trajectory, generate at least one operating condition coupling hyperedge in the dynamic hypergraph model, wherein the operating condition coupling hyperedge is associated with the instruction node representing the target power value, the node representing the real-time state of charge, the node representing the average temperature of the battery cluster, and the node representing the H-bridge junction temperature.

[0127] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the process of generating at least one working condition coupled hyperedge in the dynamic hypergraph model in this embodiment is as follows:

[0128] S2061. Based on the theoretical temperature rise trajectory, obtain the trajectory data sequence. The trajectory data sequence includes a predicted sequence of average battery cluster temperature and predicted H-bridge junction temperature at a series of consecutive time points within a future preset time period. The future preset time period is 60 minutes, with a 10-second interval between consecutive time points. The trajectory data sequence contains two sets of predicted value sequences corresponding to 360 time points, with each set containing 360 data points. The numerical precision of the predicted battery cluster temperature sequence is retained to two decimal places, and the numerical precision of the predicted H-bridge junction temperature sequence is also retained to two decimal places.

[0129] S2062. Based on the acquired trajectory data sequence, define and populate the attributes of the working condition coupling hyperedge, specifically:

[0130] D1. Based on the target power value upon which the theoretical temperature rise trajectory is derived, it is defined as the attribute value of the command node associated with the operating condition coupling hyperedge. This command node is used to characterize the external power scheduling command. The target power value ranges from 0 to 500kW, with the numerical precision retained to one decimal place. The attribute value of the command node directly adopts the specific target power value in the current power scheduling task command.

[0131] D2. Based on the real-time state of charge used in calculating the theoretical temperature rise trajectory, it is defined as the attribute value of the state of charge node associated with the operating condition coupling hyperedge. This node is used to characterize the current energy state of the battery. The numerical range of the real-time state of charge is 0.2 to 0.9, and the numerical precision is retained to three decimal places. The attribute value of the state of charge node adopts the real-time state of charge sample value when calculating the theoretical temperature rise trajectory.

[0132] D3. Based on the predicted value sequence of the average temperature of the battery cluster in the trajectory data sequence, generate and associate a battery temperature node. This node is used to characterize and carry the temperature state evolution process of the battery cluster during the prediction period. The attribute data of the battery temperature node is a complete predicted value sequence of the average temperature of the battery cluster. The sequence is stored in array format, with the array index corresponding to the time point. The index range is 0 to 359, and each index position stores the predicted temperature value for the corresponding time point.

[0133] D4. Based on the predicted value sequence of H-bridge junction temperatures in the trajectory data sequence, generate and associate an H-bridge temperature node. This node is used to characterize and carry the temperature state evolution process of the H-bridge during the prediction period. The attribute data of the H-bridge temperature node is a complete predicted value sequence of H-bridge junction temperatures. The sequence storage format is also an array, with the array index corresponding to the time point. The index range is 0 to 359, and each index position stores the predicted junction temperature value for the corresponding time point.

[0134] S2063. Based on the instruction node, state of charge node, battery temperature node, and H-bridge temperature node, a complete operating condition coupling hyperedge record is generated. The data structure of this hyperedge record includes a unique hyperedge identifier, a hyperedge type identifier marked as operating condition coupling, reference identifiers for the four associated nodes, and a storage pointer to the mathematical expression or data sequence of the theoretical temperature rise trajectory. The unique hyperedge identifier uses 16-bit binary encoding, ranging from 0 to 65535, with each hyperedge assigned a unique code. The reference identifiers for the four associated nodes all use the same encoding rule as the unique hyperedge identifier, corresponding to the unique codes of the instruction node, state of charge node, battery temperature node, and H-bridge temperature node, respectively. The storage pointer to the data sequence is a 32-bit memory address pointer, directly pointing to the starting storage address of the trajectory data sequence in memory. The mathematical expression of the theoretical temperature rise trajectory uses a quadratic polynomial fitting expression, fitting based on the time points and corresponding temperature values ​​in the trajectory data sequence. The iteration step size of the fitting algorithm is 0.001, the fitting variable is time, the time unit is seconds, and the value range is 0 to 3600.

[0135] S207. Based on the real-time temperature of the cabinet environment, the real-time power consumption of the thermal management subsystem, the system operation mode, the preset cooling energy efficiency ratio characteristic curve, and the theoretical temperature rise trajectory, obtain the theoretical cooling power consumption sequence.

[0136] Next, a specific and complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. It should be further explained that the process of obtaining the theoretical cooling power consumption sequence in this embodiment is as follows:

[0137] S2071. Based on the laboratory testing platform, perform full-condition performance calibration of the refrigeration unit in the thermal management subsystem to obtain the original performance dataset. Each data point in the original performance dataset includes the ambient temperature inside the cabinet during the test, the operating load rate of the refrigeration unit, and the measured cooling energy efficiency ratio (CERR). The ambient temperature inside the cabinet of the testing platform is adjustable from 10℃ to 45℃, with a test step of 5℃; the operating load rate covers 10% to 100%, with a test step of 10%; each temperature and load rate combination is repeated 3 times, and the original performance dataset consists of 8 temperature points × 10 load rate points × 3 repetitions, totaling 240 data points; the CERR test accuracy is retained to two decimal places.

[0138] S2072. Based on the original performance dataset, a continuous characteristic surface function is constructed using a Gaussian radial basis function surface fitting algorithm. The surface function has the cabinet ambient temperature as the first independent variable, the operating load rate as the second independent variable, and the cooling energy efficiency ratio as the dependent variable. The regularization parameter of the fitting algorithm is set to 0.0001, and the iteration termination condition is that the difference between two adjacent fitting errors is less than [a certain value]. The fitting error is calculated using the mean square error, and the accuracy of the fitting result is retained to three decimal places.

[0139] S2073. Based on a continuous characteristic surface function, discrete sampling is performed at the set ambient temperature sampling interval and operating load rate sampling interval to generate a two-dimensional lookup table composed of discrete data points, i.e., the preset cooling energy efficiency ratio characteristic curve. The ambient temperature sampling interval is set to 2℃, covering the range of 10℃ to 45℃, with a total of 18 sampling points; the operating load rate sampling interval is set to 5%, covering the range of 10% to 100%, with a total of 19 sampling points; the two-dimensional lookup table contains a total of 342 discrete data points.

[0140] S2074. Based on the theoretical temperature rise trajectory, extract the predicted values ​​of the average temperature of the battery cluster and the predicted value of the H-bridge junction temperature at each discrete time point within a preset future time period. The preset future time period is 60 minutes, with a discrete time point interval of 10 seconds, for a total of 360 discrete time points; the extracted temperature prediction values ​​are retained to two decimal places.

[0141] S2075. Based on the heat capacity values ​​of the heat capacity nodes associated with the battery cluster and H-bridge in the thermal resistance-heat capacity parameter network, calculate the heat required to maintain the predicted temperature at each time point below the expected safe temperature, and obtain a time-varying sequence of total cooling demand. The expected safe temperature is set to 40℃, the total heat capacity value of the heat capacity nodes associated with the battery cluster is 12800J / ℃, and the total heat capacity value of the heat capacity nodes associated with the H-bridge is 1600J / ℃. The total cooling demand is calculated using the heat balance formula. The demand at each time point is the product of the corresponding heat capacity value and the difference between the predicted temperature and the expected safe temperature. The calculation accuracy is retained to one decimal place. The total cooling demand sequence contains 360 data points, corresponding one-to-one with discrete time points.

[0142] S2076. Based on the system operating mode, query the preset operating mode-constraint mapping table to obtain the maximum allowable power consumption, minimum allowable power consumption, and response time constant constraints of the thermal management subsystem under the current mode. The system operating modes are divided into three types: constant power charging mode, constant power discharging mode, and frequency regulation standby mode. Under constant power charging mode, the maximum allowable power consumption is 5kW, the minimum allowable power consumption is 0.5kW, and the response time constant is 10 seconds. Under constant power discharging mode, the maximum allowable power consumption is 5kW, the minimum allowable power consumption is 0.5kW, and the response time constant is 10 seconds. Under frequency regulation standby mode, the maximum allowable power consumption is 6kW, the minimum allowable power consumption is 0.8kW, and the response time constant is 5 seconds.

[0143] S2077. Based on the real-time temperature of the cabinet environment and the total cooling demand, query and obtain the instantaneous cooling energy efficiency ratio sequence. The real-time temperature of the cabinet environment is collected by a temperature sensor with a collection period of 10 seconds, synchronized with discrete time points; the instantaneous cooling energy efficiency ratio sequence is obtained through the subsequent step S2078, containing 360 data points, synchronized with the total cooling demand sequence time.

[0144] S2078: Based on the real-time temperature of the cabinet environment and the real-time load rate calculated by dividing the total cooling demand by the rated capacity of the cooling unit, the instantaneous cooling energy efficiency ratio (ELR) value is obtained by querying and bilinear interpolating the preset ELR characteristic curve within a future preset time period. The rated capacity of the cooling unit is set to 10kW; the real-time load rate calculation accuracy is retained to two decimal places; intermediate calculation results during bilinear interpolation are retained to four decimal places, and the final instantaneous ELR value accuracy is retained to three decimal places; the instantaneous ELR sequence contains 360 data points, corresponding one-to-one with discrete time points.

[0145] S2079. Based on the total cooling demand sequence and the instantaneous cooling energy efficiency ratio sequence, the required theoretical cooling power sequence is calculated by dividing the total cooling demand at each time point by its corresponding instantaneous cooling energy efficiency ratio value.

[0146] Based on the operational constraints determined by the system's operating mode, the theoretical cooling power sequence is subjected to a limiting process to generate the final theoretical cooling power consumption sequence. The limiting process rule is as follows: when the theoretical cooling power is greater than the maximum allowable power consumption, the maximum allowable power consumption is taken as the cooling power consumption at that time point; when the theoretical cooling power is less than the minimum allowable power consumption, the minimum allowable power consumption is taken as the cooling power consumption at that time point; when it is in between, the original value remains unchanged. The final theoretical cooling power consumption sequence is accurate to two decimal places.

[0147] The motivation behind this step in this embodiment is to provide a precise quantitative relationship for the control response of the dynamic hypergraph model, that is, to quantify the cooling energy consumption required to offset the temperature rise caused by power tasks. Its core principle is as follows: First, an accurate mathematical model of the cooling unit's energy efficiency ratio as a function of ambient temperature and load rate is established through experimental calibration and surface fitting; second, based on the thermal resistance-heat capacity parameter network theory and energy conservation, the predicted temperature rise trajectory is converted into a sequence of heat loads that need to be removed; then, real-time efficiency is obtained from the energy efficiency model through dynamic querying and interpolation, and combined with operating mode constraints, the heat load sequence is converted into a cooling power consumption sequence. This process unifies multiple factors such as environmental conditions, equipment efficiency, heat load demand, and operating strategies into a single computational framework, achieving a precise and dynamic mapping from temperature rise prediction to cooling energy consumption, providing a crucial data link for subsequent global collaborative optimization.

[0148] S208. Based on the theoretical cooling power consumption sequence, generate a regulation response hyperedge in the dynamic hypergraph model, wherein the regulation response hyperedge is associated with at least two nodes among the nodes representing the average temperature of the battery cluster, the nodes representing the H-bridge junction temperature, the nodes representing the ambient temperature inside the cabinet, the nodes representing the system operating mode, and the nodes representing the total power consumption of the thermal management subsystem.

[0149] Next, a specific complete example will be used to illustrate the entire process. This example is only to illustrate the feasibility at the computational level and does not represent actual values. The specific values ​​can be determined by those skilled in the art through simulation experiments or physical experiments. For example, the process of generating a regulating response hyperedge in the dynamic hypergraph model in this embodiment is as follows:

[0150] S2081. Based on the theoretical cooling power consumption sequence, extract the theoretical cooling power consumption value corresponding to each discrete time point within a future preset time period to form a power consumption data sequence that changes with time, and define the power consumption data sequence as the core response data of the control response superedge; the future preset time period is 60 minutes, the discrete time point interval is 10 seconds, and the power consumption data sequence contains 360 data points, which correspond one-to-one with the discrete time points.

[0151] S2082. Based on the input parameters and intermediate results used to generate the theoretical cooling power consumption sequence, determine the nodes that need to be associated with the control response super-edge, and obtain the attribute values ​​of each node, specifically including:

[0152] E1. Based on the generated theoretical cooling power consumption sequence, obtain the predicted value sequences of the average temperature of the battery cluster and the predicted value sequences of the H-bridge junction temperature. The predicted value sequence of the average temperature of the battery cluster contains 360 data points with a time interval of 10 seconds; the predicted value sequence of the H-bridge junction temperature also contains 360 data points with a time interval of 10 seconds. Based on the predicted value sequence of the average temperature of the battery cluster, generate or associate a battery temperature node, which is used to characterize the temperature state that the battery cluster is expected to maintain during the prediction period; based on the predicted value sequence of the H-bridge junction temperature, generate or associate an H-bridge temperature node, which is used to characterize the temperature state that the H-bridge is expected to maintain during the prediction period. The attribute data of both nodes are stored in array form, with an array index range of 0 to 359, and each index position corresponds to the temperature prediction value at the corresponding time point.

[0153] E2. Based on the real-time temperature of the cabinet environment used when generating the theoretical cooling power consumption sequence, generate or associate an ambient temperature node and set its attribute value to the real-time temperature. The real-time temperature of the cabinet environment is collected by a temperature sensor with a collection period of 10 seconds, synchronized with discrete time points. This node is used to characterize the environmental conditions on which the cooling energy efficiency is calculated.

[0154] E3. Based on the system operation mode used to generate the theoretical cooling power consumption sequence, generate or associate a mode node and set its attribute value to the system operation mode. There are three types of system operation modes: constant power charging mode, constant power discharging mode, and frequency-modulated standby mode. The attribute values ​​are stored in the form of string identifiers. This node is used to characterize the operating constraints and optimization objectives applied to the thermal management strategy.

[0155] E4. Based on the theoretical cooling power consumption sequence itself, generate or associate a thermal management power consumption node and set its attribute value to the power consumption data sequence. The storage format of the power consumption data sequence is consistent with that in S2081, with an array index range of 0 to 359. Each index position stores the theoretical cooling power consumption value at the corresponding time point. This node represents the theoretical energy consumption required by the thermal management subsystem to achieve the temperature control targets of the battery temperature node and the H-bridge temperature node.

[0156] S2083. Based on the battery temperature node, H-bridge temperature node, ambient temperature node, mode node, and thermal management power consumption node, as well as the core response data of the control response superedge, a complete control response superedge record is generated. The data structure of this superedge record includes a unique superedge identifier, a superedge type identifier marked as control response, reference identifiers for at least two of the aforementioned associated nodes, and a storage pointer to the power consumption data sequence. The unique superedge identifier uses 16-bit binary encoding, ranging from 0 to 65535, with each superedge assigned a unique code. The reference identifiers of the associated nodes all use the same encoding rule as the unique superedge identifier, corresponding to the unique code of each node. The storage pointer to the power consumption data sequence is a 32-bit memory address pointer, directly pointing to the starting storage address of the power consumption data sequence in memory. This superedge is used in the dynamic hypergraph model to fully represent the causal relationship of the thermal management power consumption required to maintain the target temperature state of the battery and H-bridge under specific ambient temperatures and operating modes.

[0157] S209. Based on the aforementioned working condition coupling hyperedge, the aforementioned regulation response hyperedge, and the corresponding nodes with associated relationships, a hypergraph algorithm is used to construct the dynamic hypergraph model. It should be further noted that the construction and training process of the dynamic hypergraph model in this embodiment includes:

[0158] F1. Based on all acquired nodes, construct a node set V for the dynamic hypergraph model. The node set V includes instruction nodes representing the target power value, nodes representing the real-time state of charge, nodes representing the average temperature of the battery cluster, nodes representing the H-bridge junction temperature, nodes representing the ambient temperature inside the cabinet, nodes representing the system operating mode, and nodes representing the total power consumption of the thermal management subsystem.

[0159] F2. Based on at least one operating condition coupling hyperedge and at least one regulation response hyperedge generated, construct the set of hyperedges E of the dynamic hypergraph model.

[0160] F3. Based on the node set V and the hyperedge set E, construct the initial hypergraph structure H=(V,E) of the dynamic hypergraph model.

[0161] F4. Assign an initial weight value to each hyperedge in the hyperedge set E; the initial weight of the operating condition coupling hyperedge is normalized based on the target power value of its associated command node; the initial weight of the control response hyperedge is set based on the priority of the operating mode represented by its associated mode node.

[0162] F5. For each node in the node set V, construct its feature vector. The feature vector of a node includes the node's real-time attribute value, the statistical characteristics of the historical attribute value sequence, and the frequency of the node's occurrence in all hyperedges.

[0163] F6. Based on the structure of the dynamic hypergraph model, construct a hypergraph adjacency matrix to quantify the association strength between each pair of nodes through sharing the same hyperedge.

[0164] F7. Based on the hypergraph adjacency matrix and the feature vectors of each node, the updated feature vector of each node after information aggregation and relationship strengthening is obtained through the hypergraph attention network layer. The parameter settings of the hypergraph attention network layer are as follows: 4 attention heads, 64 hidden layer dimensions, ReLU activation function, dropout probability set to 0.1, and Xavier initialization method is used for the weight initialization of the network layer.

[0165] F8. Based on the updated node feature vectors and the updated features of the nodes associated with each hyperedge, the weight value of each hyperedge is updated through a feedforward neural network to obtain the dynamically optimized hyperedge weights. The parameters of the feedforward neural network are set as follows: the network has 3 layers, the input dimension of the first layer is the same as the dimension of the node feature vector, the output dimension of the first layer is 128, the output dimension of the second layer is 64, and the output dimension of the third layer is 1. Each layer uses the ReLU activation function, the dropout probability is set to 0.1, the optimizer is the Adam optimizer, and the learning rate is set to 0.001.

[0166] F9. Based on the updated hyperedge weights and node characteristics in the dynamic hypergraph model, the objective function and constraints of the collaborative optimization problem are reconstructed. In the objective function, the weights of the operating condition coupling hyperedges are used to adjust the degree of emphasis on the temperature rise trajectory under different power tasks, while the weights of the control response hyperedges are used to adjust the degree of emphasis on thermal management energy consumption under different operating modes.

[0167] F10. Based on the reconstructed optimization problem, perform calculations to obtain the updated optimal power adjustment sequence and optimal power consumption adjustment sequence.

[0168] F11. Based on the real-time monitoring data acquired in the new sampling period, the new power scheduling task instructions and the system operation mode, repeat S201 to S208 to generate new operating condition coupling superedge and control response superedge.

[0169] F12. Add the newly generated hyperedge and its associated nodes to the node set V and hyperedge set E of the dynamic hypergraph model.

[0170] F13. Based on the hypergraph structure after adding new data, repeat F1 to F11 to iteratively update the hyperedge weights and node features, and solve for a new collaborative optimization strategy.

[0171] F14. Based on long-term historical optimization results and actual operation data, a loss function is constructed, and the parameters in the hypergraph attention network layer and feedforward neural network are periodically trained and fine-tuned offline to enable the model to continuously evolve to adapt to long-term changes in system characteristics. The loss function adopts the mean squared error loss function, the training batch size is set to 32, the number of training iterations is set to 100 rounds, and periodic offline training is performed once every 24 hours. During fine-tuning, the learning rate is decayed to 0.1 times the initial learning rate.

[0172] This process fundamentally solves the overall energy efficiency bottleneck caused by the fragmented optimization of power scheduling and thermal management subsystems in traditional grid-type energy storage cabinets by constructing and continuously updating a collaborative optimization model based on a dynamic hypergraph. In particular, through a series of refined modeling techniques such as online parameterized loss generation, two-dimensional efficiency cloud map interpolation, and compensation surfaces that integrate aging and temperature, high-precision and adaptive estimation of the instantaneous total heat source power of the system is achieved. Furthermore, by utilizing the thermal resistance-thermal capacity parameter network and state-space equations, the theoretical temperature rise trajectory caused by the execution of power tasks is accurately predicted and encapsulated as a condition coupling hyperedge that associates commands, states, and temperature nodes. At the same time, based on the cooling energy efficiency characteristic curve and operating mode constraints, the theoretical cooling power consumption sequence required to offset the temperature rise is calculated and encapsulated as a regulation response hyperedge that associates temperature, mode, and power consumption nodes. The key lies in the fact that, through algorithms such as hypergraph attention networks, this scheme dynamically learns and quantifies the importance weights of these heterogeneous hyperedges and the complex coupling relationships between their associated nodes. This allows for global collaborative optimization of power curves and cooling power consumption curves over future time periods within a unified hypergraph model framework. This enables the thermal management subsystem to shift from passively responding to temperature to actively coordinating power planning, significantly reducing ineffective energy consumption due to overcooling or delayed response while ensuring safety. Finally, through the model's online rolling update and offline training mechanism, the system gains the ability to continuously adapt to equipment aging, environmental changes, and scheduling needs throughout its entire lifecycle, achieving a dual improvement in systemic energy saving and operational reliability.

[0173] It should be further explained that, in this embodiment, the optimization objective is to minimize the total operating energy consumption of the grid-type energy storage cabinet, and the process of collaborative optimization based on the dynamic hypergraph model includes:

[0174] Based on the effective time window contained in the power scheduling task instruction, the start and end times of the prediction period are obtained.

[0175] Based on the preset optimization calculation discretization step size, a series of equally spaced discrete time points are obtained within the prediction period; the optimization calculation discretization step size is set to 10 seconds, which is consistent with the time interval of the theoretical temperature rise trajectory and cooling power consumption sequence mentioned above.

[0176] Based on nodes of type instruction in the dynamic hypergraph model, obtain their attribute values, i.e., target power values.

[0177] Based on the discrete time point sequence, a first decision variable sequence is defined, where each decision variable corresponds to the planned output power of the power conversion subsystem at a discrete time point.

[0178] Based on the discrete time point sequence, a second decision variable sequence is defined, where each decision variable corresponds to the planned power consumption of the thermal management subsystem at a discrete time point.

[0179] Based on the first decision variable sequence, the planned output power value of the power conversion subsystem at each time point within the prediction period is obtained.

[0180] Based on the planned output power value in the first decision variable sequence, and the corresponding switching frequency obtained in real time by the power conversion subsystem controller or determined according to the power strategy, the corresponding instantaneous conversion efficiency sequence is obtained by querying and bilinear interpolation in the pre-constructed two-dimensional efficiency cloud map. During the bilinear interpolation process, intermediate calculation results are retained to four decimal places, and the final instantaneous conversion efficiency value is retained to three decimal places.

[0181] Based on the planned output power value and the corresponding instantaneous conversion efficiency value in the first decision variable sequence, the total energy consumption of the power conversion subsystem due to power conversion loss during the prediction period is obtained by calculation.

[0182] Based on the second decision variable sequence, the planned power consumption of the thermal management subsystem at each time point within the prediction period is obtained, and its total operating energy consumption is calculated by summation. The calculation precision is retained to one decimal place.

[0183] Based on the calculated total energy consumption of the power conversion subsystem and the total energy consumption of the thermal management subsystem, a scalar function is constructed with the goal of minimizing the sum of the two, i.e., the objective function for minimizing total operating energy consumption.

[0184] Based on the hyperedges identified as "condition coupling" in the dynamic hypergraph model, the theoretical temperature rise trajectory mathematical model pointed to by their storage pointers is obtained. This model is a state-space equation describing the relationship between the input heat source power and the output temperature change. In this embodiment, based on the data structure of the condition coupling hyperedge, the pointer field stored internally is accessed to obtain the address information pointing to the external storage area. Next, based on the address information, the corresponding external storage area is accessed, and the data structure stored therein is read. This data structure is the encapsulated theoretical temperature rise trajectory mathematical model. Finally, the data structure of this mathematical model is parsed to obtain all the parameters constituting the state-space equation, including the coefficients of the state matrix, input matrix, output matrix, and direct transfer matrix, as well as the nominal parameter set of the thermal resistance-capacity parameter network. This process realizes the transformation from the abstract relational edges in the hypergraph model to the specific executable mathematical equations for simulation calculation, providing an accurate computational kernel for subsequently mapping power decision variables to temperature prediction values.

[0185] Based on the planned output power value in the first decision variable sequence, and the corresponding time-bound heat source power calculation method determined through steps such as online parameterized loss generation model, two-dimensional efficiency cloud map query, and loss compensation surface query, the future heat source power sequence is calculated and obtained as the input to the state-space equation. The numerical precision of the heat source power sequence is retained to one decimal place, and the number of data points is consistent with the discrete time point sequence.

[0186] Based on the heat source power sequence, and using state-space equations for forward simulation calculations, the predicted values ​​of the average temperature of the battery cluster and the H-bridge junction temperature within the prediction period are obtained. The forward simulation calculations employ a fourth-order Runge-Kutta numerical integration algorithm, with an integration step size consistent with the discretization step size of the optimization calculation, which is 10 seconds. The temperature prediction accuracy is retained to two decimal places.

[0187] Based on a preset maximum allowable operating temperature for the battery, a battery temperature safety constraint is constructed, wherein all values ​​in the predicted average temperature sequence of the battery cluster must not exceed this maximum allowable operating temperature. The maximum allowable operating temperature for the battery is set to 45℃.

[0188] Based on the preset maximum allowable junction temperature of the H-bridge, a temperature safety constraint for the H-bridge is constructed, wherein all values ​​in the predicted junction temperature sequence of the H-bridge must not exceed the maximum allowable junction temperature; the maximum allowable junction temperature of the H-bridge is set to 125℃.

[0189] Based on the hyperedges identified as control responses in the dynamic hypergraph model, we obtain the mapping logic framework representing the temperature control demand to cooling energy consumption.

[0190] Based on the predicted average temperature sequence of the battery cluster and the predicted junction temperature sequence of the H-bridge, combined with the heat capacity values ​​of relevant heat capacity nodes in the thermal resistance-heat capacity parameter network, the total cooling demand sequence required to be removed to maintain the temperature below the safe upper limit is calculated. The total heat capacity value of the heat capacity nodes associated with the battery cluster is 12800 J / ℃, and the total heat capacity value of the heat capacity nodes associated with the H-bridge is 1600 J / ℃. The total cooling demand calculation is accurate to one decimal place.

[0191] Based on the real-time monitored ambient temperature inside the cabinet and the real-time load rate sequence calculated by dividing the total cooling demand sequence by the rated capacity of the cooling unit, the corresponding instantaneous cooling energy efficiency ratio sequence is obtained by querying and bilinear interpolating the preset cooling energy efficiency ratio characteristic curve. The rated capacity of the cooling unit is set to 10kW.

[0192] Based on the total cooling demand sequence and the instantaneous cooling energy efficiency ratio sequence, the minimum theoretical cooling power consumption sequence necessary to meet temperature control requirements is calculated. The calculation accuracy is retained to two decimal places.

[0193] Construct energy consumption matching constraints that require each planned power consumption value in the second decision variable sequence to be greater than or equal to the minimum theoretical cooling power consumption value at the corresponding time.

[0194] Based on the system operating mode, the preset operating mode-constraint mapping table is queried to obtain the maximum and minimum allowable power consumption of the thermal management subsystem in the current mode. The maximum allowable power consumption is 5kW and the minimum allowable power consumption is 0.5kW in constant power charging mode and constant power discharging mode; the maximum allowable power consumption is 6kW and the minimum allowable power consumption is 0.8kW in frequency regulation standby mode.

[0195] Construct device capability constraints that require all values ​​in the second decision variable sequence to be within the range formed by the minimum and maximum allowable power consumption.

[0196] Based on the equipment technical specifications of the power conversion subsystem, obtain its maximum and minimum allowable output power; the maximum allowable output power of the power conversion subsystem is set to 500kW, and the minimum allowable output power is set to 0kW.

[0197] Construct power output boundary constraints that require all values ​​in the first decision variable sequence to be within the range formed by the maximum and minimum power.

[0198] Based on the target power value in the power scheduling task instruction, a power tracking constraint is constructed, requiring that the overall trend or cumulative energy of the first decision variable sequence during the prediction period be consistent with the requirements of the scheduling task.

[0199] Based on the output of the dynamic hypergraph model after computation through the hypergraph attention network layer, the dynamic weight value corresponding to each hyperedge is obtained.

[0200] Based on the dynamic weight value of each operating condition coupling hyperedge, its penalty coefficient in the optimization problem corresponding to the temperature safety constraint is adjusted. The higher the weight, the greater the cost of violating the constraint in the optimization.

[0201] Based on the dynamic weight value of each control response superedge, the weighting coefficient of the total energy consumption of the thermal management subsystem in the objective function of minimizing the total operating energy consumption is adjusted. The higher the weight, the higher the optimization priority of reducing this part of the energy consumption.

[0202] The objective function of minimizing total operating energy consumption, all defined decision variables, and all constructed constraints and weighted relationships are integrated into a complete constrained mathematical optimization model.

[0203] Based on the characteristics of the mathematical optimization model and combined with a multi-objective numerical optimization algorithm, a numerical solution for the decision variables that satisfies all constraints and minimizes the objective function value is obtained. The preferred multi-objective numerical optimization algorithm is a sequential quadratic programming algorithm, with the following parameter settings: convergence accuracy is set to... The maximum number of iterations is set to 500, the initial step size factor is set to 0.1, and the line search adopts the Armijo criterion with the criterion parameter set to 0.001.

[0204] Based on the obtained numerical solutions of the decision variables, the optimal value of the first decision variable sequence is output as the optimal power adjustment sequence of the power conversion subsystem.

[0205] Based on the obtained numerical solutions of the decision variables, the optimal values ​​of the second decision variable sequence are output as the optimal power consumption adjustment sequence of the thermal management subsystem.

[0206] It should be further explained that this embodiment involves joint control of the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet, including:

[0207] Based on the optimal power adjustment sequence and optimal power consumption adjustment sequence obtained by solving and arranged in time series, combined with the current system clock, the power setting value and power consumption setting value corresponding to the current control cycle timestamp are obtained. The current control cycle is set to 10 seconds, which is consistent with the discrete time point interval mentioned above, and the system clock synchronization accuracy is at the millisecond level.

[0208] Based on the power setpoint, power control commands containing specific pulse width modulation parameters or power reference values ​​are generated through the control parameter mapping relationship of the power conversion subsystem. Based on the power consumption setpoint, and combined with the power consumption-speed / flow characteristic curves of each actuator in the thermal management subsystem, device control commands containing specific speed setpoints or frequency setpoints are generated. The power consumption-speed characteristic curve of the fan is a quadratic function with fitting coefficients of 0.002, 0.15, and 0; the power consumption-frequency characteristic curve of the compressor is a linear relationship with fitting coefficients of 0.08 and 0.2; the carrier frequency of the pulse width modulation parameter is set to 10kHz, and the duty cycle adjustment accuracy is 0.1%.

[0209] Power control commands are encapsulated into a first standard data frame via a pre-defined CAN bus or Ethernet communication interface and sent to the local controller of the power conversion subsystem. Device control commands are encapsulated into a second standard data frame via the same or another pre-defined communication interface and sent to the local controller of the thermal management subsystem. The CAN bus communication baud rate is set to 500kbps, the data frame format is a standard frame, and the identifier range is 0x000 to 0x7FF; Ethernet communication uses the TCP / IP protocol, the port number is set to 8080, and the data transmission cycle is consistent with the control cycle at 10 seconds.

[0210] In the next sampling cycle, the actual output power of the power conversion subsystem after execution, the actual power consumption of the thermal management subsystem, and the latest temperature data of the battery cluster and H-bridge, acquired by the sensor network, are used as real-time feedback data. The sampling frequency of the sensor network is set to 10Hz, with the actual output power sampling accuracy retained to one decimal place, the actual power consumption sampling accuracy retained to two decimal places, and the temperature data sampling accuracy retained to two decimal places. Simultaneously, based on the digital model of the twin energy storage cabinet, synchronous simulation is performed with the same control commands input to acquire simulation state data. The simulation step size of the digital model is set to 1 second, and the accuracy of the simulation output data is consistent with the sensor sampling accuracy.

[0211] Based on real-time feedback data and simulation status data, the execution effect is verified, specifically including: calculating the tracking deviation between the actual output power and the power setpoint to determine whether it exceeds the first preset threshold; verifying whether the latest temperature of the battery cluster and H-bridge exceeds the safety limit; and calculating the deviation between the actual total operating energy consumption and the expected energy consumption to determine whether it exceeds the second preset threshold. The first preset threshold is set to 5kW, and the second preset threshold is set to 0.3kW·h; the maximum safety limit for the battery is 45℃, and the maximum safety limit for the H-bridge is 125℃.

[0212] If all calculated deviations in the verification results are within the corresponding preset threshold range, and the temperature does not exceed the safety upper limit, then it is determined to meet expectations. In the next control cycle, the above steps are repeated, and the setpoint for the next moment is obtained based on the optimal sequence and issued for execution.

[0213] If any calculation deviation in the verification results exceeds its corresponding preset threshold, or if the temperature exceeds the safety limit, it is determined to be a deviation from expectations, and the instantaneous local re-optimization of the dynamic hypergraph model is immediately triggered. Instantaneous local re-optimization, based on the latest real-time feedback data and the current system state, re-performs rapid co-optimization calculations within a shortened prediction time window, generating corrected power and consumption adjustment segments. The shortened prediction time window is set to 10 minutes, the discretization step size remains 10 seconds, and a total of 60 data points are used. The rapid co-optimization calculation employs a sequential quadratic programming algorithm, with its parameters set to convergence accuracy. The maximum number of iterations is 100, the initial step size factor is 0.2, and the line search adopts the Armijo criterion with a criterion parameter of 0.01.

[0214] The revised adjustment segment is immediately issued and executed as a new control command to replace the optimal sequence of the subsequent part in the original plan, thereby realizing closed-loop and adaptive joint control of power conversion and thermal management of grid-type energy storage cabinet.

[0215] It should be further explained that this embodiment takes the stable operation of a grid-type energy storage cabinet under complex operating conditions with frequent switching between multiple operating modes such as grid-connected charging, frequency regulation, and off-grid support as the specific scenario. Traditional processing methods often first formulate a fixed power output plan based on a preset static loss empirical formula. This plan only focuses on the power tracking requirements of the grid dispatch command and does not consider the dynamic coupling relationship between the heat source generated during the power conversion process and the thermal management energy consumption. Then, it passively starts or adjusts the operating status of thermal management equipment such as fans and compressors based solely on the real-time feedback data of the temperature sensor inside the cabinet. Because the front-end power plan lacks the ability to adapt to changes in operating conditions, aging and decay of equipment after long-term operation, and loss characteristic drift caused by ambient temperature fluctuations, the static model cannot reflect the electrothermal coupling law in real time. This leads to the thermal management strategy either reserving too large a safety margin to avoid the risk of temperature exceeding the limit, causing the thermal management equipment to operate at high load for a long time, resulting in a large amount of ineffective energy waste; or due to inaccurate temperature rise prediction, the adjustment is not timely when temperature changes occur, causing the equipment temperature to fluctuate too much or even approach the safety threshold, affecting the reliability of operation. Even if the speed of the thermal management equipment or the output of the power conversion system are adjusted locally in the future, it can only alleviate the immediate problem and cannot reverse the core contradiction between the lack of full exploitation of energy efficiency potential and safe and stable operation caused by the rigidity of front-end decisions.

[0216] This application employs a more refined technical processing method for this scenario. First, it synchronously collects key electrothermal parameters in real time, such as the output power of the power conversion system, the state of charge and average temperature of the battery cluster, the real-time junction temperature of the H-bridge, the ambient temperature inside the cabinet, and the total power consumption of the thermal management subsystem, through a multi-source sensor network. Then, using an online parameterized loss generation model, combined with the actual heat source power feedback value from the previous moment and the current input parameters, it updates the polynomial coefficients within the model in real time using a recursive least squares identification algorithm to dynamically generate the baseline loss power value under the current operating condition. Simultaneously, based on a pre-constructed two-dimensional efficiency cloud map, combined with the current load rate and switching frequency of the power conversion system, it obtains the instantaneous conversion efficiency through bilinear interpolation to construct an efficiency compensation factor. Then, relying on the loss compensation surface that integrates battery health status and temperature characteristics, it queries and obtains the corresponding compensation coefficients. Finally, it uses a Kalman filter to weightedly fuse the above multi-dimensional data to obtain a high-precision optimal estimate of the instantaneous total heat source power, effectively avoiding the inherent defects of static models. Based on this, a set of thermal balance differential equations is constructed according to Kirchhoff's current law, following the pre-calibrated thermal resistance and thermal capacity parameter network of the energy storage cabinet's physical structure. This system is then transformed into a state-space equation. Combined with dynamically acquired heat distribution coefficients of the battery and H-bridge, the theoretical temperature rise trajectory of the battery cluster average temperature and H-bridge junction temperature within a preset future time period is accurately predicted. Simultaneously, a parameter adaptive update algorithm is invoked to continuously fine-tune the thermal resistance-thermal capacity parameter network parameters based on the real-time temperature prediction error sequence, further improving prediction accuracy. This theoretical temperature rise trajectory is then encapsulated as a condition-coupled hyperedge linking the target power command, battery state of charge, battery temperature, and H-bridge temperature nodes. Based on the cabinet's ambient temperature, system operating mode, and preset cooling efficiency ratio characteristic curve, the theoretical cooling power consumption sequence required to offset the temperature rise is calculated and encapsulated as a control response hyperedge linking the temperature, mode, and thermal management power consumption nodes. A dynamic hypergraph model is then constructed based on these two types of hyperedges. By dynamically learning and optimizing node features and hyperedge weights through a hypergraph attention network, and aiming to minimize total operating energy consumption within a unified model framework, the optimal power regulation sequence and thermal management power consumption regulation sequence for future scheduling cycles are collaboratively solved. Finally, the control commands are synchronously simulated using a twin energy storage cabinet, and the effect is verified by combining real-time feedback data from actual operation. If parameter deviations or temperature anomalies occur, local re-optimization with a shortened prediction time window is immediately triggered, generating corrected control commands and issuing them for execution. The entire process realizes integrated collaborative formulation and dynamic iterative optimization of power planning and thermal management strategies at the front end, fundamentally avoiding the risks of excessive safety margins, energy waste, or dynamic mismatch caused by traditional methods that postpone collaborative optimization to the control end or rely on static models. This ensures that the system maintains both safety and stability and high energy efficiency under complex operating conditions.

[0217] Example 2

[0218] Please see Figure 2Another embodiment of the present invention provides: a smart power consumption optimization system for grid-based energy storage, comprising:

[0219] The data acquisition module obtains operational status monitoring data, external dispatch commands, and system operation modes of the grid-type energy storage cabinet;

[0220] The construction module, based on the operation status monitoring data, scheduling instructions and system operation mode, constructs a dynamic hypergraph model to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand and thermal management energy consumption in the grid-type energy storage cabinet;

[0221] The optimization solution module takes minimizing the total operating energy consumption of the grid-type energy storage cabinet as the optimization objective, and performs collaborative optimization solution based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles.

[0222] The execution module executes the cooperative operation strategy on the preset twin energy storage cabinet to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet.

[0223] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the claims. All of these variations are within the protection scope of the present invention.

Claims

1. A smart power consumption optimization method for grid-based energy storage, characterized in that, include: Acquire operational status monitoring data, external dispatch commands, and system operation modes of grid-type energy storage cabinets; Based on the aforementioned operational status monitoring data, scheduling instructions, and system operation modes, a dynamic hypergraph model is constructed to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand, and thermal management energy consumption in the grid-type energy storage cabinet. With the goal of minimizing the total operating energy consumption of the grid-type energy storage cabinet, a collaborative optimization solution is performed based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles. On the pre-set twin energy storage cabinet, the cooperative operation strategy is executed to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet; The construction process of the dynamic hypergraph model includes: Based on real-time monitoring data from sensors within the grid-type energy storage cabinet, the output power of the power conversion subsystem, the state of charge and temperature of the battery clusters, the temperature of the H-bridge, the temperature of the cabinet environment, and the real-time power consumption of the thermal management subsystem are obtained. The communication messages collected from the energy management subsystem are parsed to obtain the power scheduling task instructions of the grid-type energy storage cabinet. The power scheduling task instructions include the target power value and the effective time window. Based on the local controller status of the network-type energy storage cabinet or the mode identifier bit in the communication message, the current system operating mode is obtained; Based on the target power value in the power scheduling task instruction, the real-time state of charge of the battery cluster, and the real-time output power of the power conversion subsystem, the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions is obtained by querying a preset loss-power-state of charge mapping table.

2. The intelligent power consumption optimization method for grid-type energy storage as described in claim 1, characterized in that, The construction process of the dynamic hypergraph model also includes: Based on the instantaneous total heat source power, the average temperature of the battery cluster, the real-time junction temperature of the H-bridge, and the thermal resistance-capacity parameter network pre-calibrated according to the physical structure of the energy storage cabinet, the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge within a future preset time period is obtained by solving the state-space equation corresponding to the thermal resistance-capacity parameter network. The theoretical temperature rise trajectory constitutes a working condition coupling hyperedge in the dynamic hypergraph model. The nodes associated with the working condition coupling hyperedge include: a command node representing the target power value, a node representing the real-time state of charge, a node representing the average temperature of the battery cluster, and a node representing the junction temperature of the H-bridge.

3. The intelligent power consumption optimization method for grid-type energy storage as described in claim 2, characterized in that, The construction process of the dynamic hypergraph model also includes: Based on the theoretical temperature rise trajectory, at least one operating condition coupled hyperedge is generated in the dynamic hypergraph model, wherein the operating condition coupled hyperedge is associated with the instruction node representing the target power value, the node representing the real-time state of charge, the node representing the average temperature of the battery cluster, and the node representing the H-bridge junction temperature. Based on the real-time temperature of the cabinet environment, the real-time power consumption of the thermal management subsystem, the system operation mode, the preset cooling energy efficiency ratio characteristic curve, and the theoretical temperature rise trajectory, the theoretical cooling power consumption sequence is obtained. Based on the theoretical cooling power consumption sequence, a regulation response hyperedge is generated in the dynamic hypergraph model, wherein the regulation response hyperedge is associated with at least two nodes among the nodes representing the average temperature of the battery cluster, the nodes representing the H-bridge junction temperature, the nodes representing the ambient temperature inside the cabinet, the nodes representing the system operating mode, and the nodes representing the total power consumption of the thermal management subsystem. The dynamic hypergraph model is constructed based on the operating condition coupling hyperedge, the control response hyperedge, and the corresponding nodes with related relationships.

4. The intelligent power consumption optimization method for grid-type energy storage as described in claim 3, characterized in that, The process of obtaining the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions includes: Based on the target power value, the real-time state of charge, and the real-time temperature of the battery cluster, an online parameterized loss generation model is invoked to generate a baseline loss power value under the current operating condition; wherein, the online parameterized loss generation model uses recursive least squares identification with the actual total heat source power feedback value of the previous moment and the current input parameters, and updates its internal polynomial coefficients in real time. Based on the real-time output power of the power conversion subsystem, the target power value, and the current switching frequency and modulation ratio obtained in real time by the power conversion subsystem controller, the instantaneous conversion efficiency at the current operating point is obtained by bilinear interpolation through a pre-constructed two-dimensional efficiency cloud map, and an efficiency compensation factor is constructed.

5. The intelligent power consumption optimization method for grid-type energy storage as described in claim 4, characterized in that, The process of obtaining the instantaneous total heat source power of the power conversion subsystem and the battery cluster under the current operating conditions also includes: Based on the real-time temperature of the battery cluster and the current battery health status value obtained through the battery management subsystem, the preset loss compensation surface that integrates aging and temperature characteristics is queried to obtain the compensation coefficient. Based on the dynamically generated baseline loss power value, the efficiency compensation factor, and the compensation coefficient, a weighted fusion is performed using a Kalman filter to obtain the optimal estimate of the instantaneous total heat source power.

6. The intelligent power consumption optimization method for grid-type energy storage as described in claim 5, characterized in that, The process of obtaining the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge over a future preset time period includes: Based on the pre-calibrated thermal resistance-thermal capacity parameter network, the topology of the thermal resistance-thermal capacity parameter network is obtained. The topology includes the position of each thermal capacity node, the connection relationship of each thermal resistance node, and the injection position of each heat source node. Based on the topology and the thermal capacity value of each thermal capacity node and the thermal resistance value of each thermal resistance node in the thermal resistance-thermal capacity parameter network, a set of thermal balance differential equations with the temperature of each thermal capacity node as the state variable is constructed. The thermal equilibrium differential equations are transformed into the standard form of state-space equations to obtain state matrix A, input matrix B, output matrix C, and direct transfer matrix D. The elements of state matrix A are composed of the reciprocals of the thermal capacity and thermal resistance, used to characterize the thermal coupling strength between each thermal capacity node. The elements of input matrix B characterize the heat injection relationship between each heat source node and its corresponding thermal capacity node. Output matrix C is used to extract the average temperature of the battery cluster and the H-bridge junction temperature from all state variables. Direct transfer matrix D is set to a zero matrix.

7. The intelligent power consumption optimization method for grid-type energy storage as described in claim 6, characterized in that, The method of obtaining the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge over a future preset time period also includes: Based on the target power value in the power scheduling task instruction and the system operation mode, the preset heat generation allocation ratio table is queried to obtain the battery heat generation allocation coefficient and the H-bridge heat generation allocation coefficient. Based on the battery heat generation distribution coefficient and the H-bridge heat generation distribution coefficient, the instantaneous total heat source power is decomposed into battery heat source power and H-bridge heat source power. Based on the battery heat source power and the H-bridge heat source power, an input vector u is constructed according to the heat injection relationship defined by the input matrix B, where each element of the input vector u corresponds to the injection power of a heat source node at the current moment; Based on the average temperature of the battery cluster and the real-time junction temperature of the H-bridge, optimal estimation is performed using a Kalman filter to obtain the state estimates of the temperature of all thermal capacity nodes in the thermal resistance-thermal capacity parameter network.

8. The intelligent power consumption optimization method for grid-type energy storage as described in claim 7, characterized in that, The method of obtaining the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H-bridge over a future preset time period also includes: Based on the state estimate, the initial values ​​of the state vector are constructed; Based on the state-space equation, a numerical integration algorithm is used, with the initial value as the initial condition and the input vector u as the input, to calculate and obtain the change sequence of the state vector within the future preset time period; Based on the change sequence of the state vector, a linear transformation is performed through the output matrix C to extract the theoretical temperature rise trajectory of the average temperature of the battery cluster and the junction temperature of the H bridge within the future preset time period; The average temperature of the battery cluster and the real-time junction temperature of the H-bridge are obtained by real-time monitoring and compared with the predicted values ​​at the corresponding time in the theoretical temperature rise trajectory to calculate and obtain the temperature prediction error sequence. Based on the temperature prediction error sequence, a preset parameter adaptive update algorithm is invoked to fine-tune at least one parameter in the thermal resistance-heat capacity parameter network online, thereby obtaining the updated thermal resistance-heat capacity parameter network parameters; wherein, the parameter adaptive update algorithm is a recursive least squares algorithm or an extended Kalman filter algorithm. Based on the updated thermal resistance-heat capacity parameter network parameters, the theoretical temperature rise trajectory extraction process is re-executed to obtain the updated theoretical temperature rise trajectory, and the updated theoretical temperature rise trajectory is used as the final output theoretical temperature rise trajectory for the subsequent construction of the dynamic hypergraph model.

9. A smart power consumption optimization system for grid-based energy storage, used to implement the smart power consumption optimization method for grid-based energy storage as described in any one of claims 1-8, characterized in that, include: The data acquisition module obtains operational status monitoring data, external dispatch commands, and system operation modes of the grid-type energy storage cabinet; The construction module, based on the operation status monitoring data, scheduling instructions and system operation mode, constructs a dynamic hypergraph model to comprehensively characterize the coupling relationship between power scheduling tasks, heat load demand and thermal management energy consumption in the grid-type energy storage cabinet; The optimization solution module takes minimizing the total operating energy consumption of the grid-type energy storage cabinet as the optimization objective, and performs collaborative optimization solution based on the dynamic hypergraph model to generate a collaborative operation strategy for power scheduling and thermal management control in future scheduling cycles. The execution module executes the cooperative operation strategy on the preset twin energy storage cabinet to jointly control the power conversion subsystem and thermal management subsystem of the grid-type energy storage cabinet.

Images