Low-temperature emergency starting power supply control system

Abstract

This invention relates to the field of power management technology, specifically to a low-temperature emergency start-up power supply control system, comprising: a temperature acquisition unit for real-time acquisition of dynamic temperature field data of the vehicle power supply operating environment through a multi-source temperature sensing network, obtaining ambient temperature field distribution data; a trend prediction unit for predicting the temperature field evolution trend based on the ambient temperature field distribution data, obtaining temperature field evolution trend data; and a risk assessment unit for performing risk coupling assessment based on the temperature field evolution trend data, and obtaining comprehensive risk level data based on power supply health status data and vehicle operating history data. This invention obtains global temperature information by constructing a multi-source temperature sensing network and a dynamic temperature field, and uses algorithms such as wavelet transform and long short-term memory networks to predict the temperature field evolution trend, avoiding the response lag problem of traditional methods and enabling timely and accurate thermal management of the battery.

Description

Technical Field

[0001] This invention relates to the field of power management technology, specifically to a low-temperature emergency start-up power control system. Background Technology

[0002] With the popularization of new energy vehicles and hybrid vehicles, power battery systems based on lithium-ion batteries have become core components. However, the electrochemical characteristics of power batteries are extremely sensitive to operating temperature. In low-temperature environments, the internal resistance of the battery increases sharply, the electrolyte viscosity increases, leading to a decrease in ion migration rate and a decline in the activity of active materials. This results in a significant degradation in the battery's usable capacity and discharge power performance, which directly leads to serious problems such as difficulty in starting, a sharp reduction in driving range, and low charging efficiency in cold climates. To solve this problem, existing technologies generally adopt battery preheating schemes, that is, heating the battery pack with external heating elements before starting the vehicle or during operation, so that its temperature rises to a better operating range.

[0003] However, existing low-temperature preheating control systems often have some drawbacks. The system adopts passive logic based on fixed thresholds, resulting in delayed response and no predictive ability. Its decision-making dimension is singular, relying only on sparse temperature data and ignoring multi-dimensional risk factors such as battery health status and driving history. Its preheating scheme uses a fixed power level, which cannot be finely adjusted according to the real, uneven temperature field and dynamically assessed risk level, ultimately leading to low preheating efficiency, energy waste, and even the risk of damaging battery life. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a low-temperature emergency start-up power supply control system.

[0005] A low-temperature emergency start-up power supply control system, comprising: The temperature acquisition unit is used to collect dynamic temperature field data of the vehicle power supply operating environment in real time through a multi-source temperature sensing network, and obtain ambient temperature field distribution data. The trend prediction unit is used to predict the temperature field evolution trend based on the environmental temperature field distribution data, and obtain temperature field evolution trend data. The risk assessment unit is used to perform risk coupling assessment based on the temperature field evolution trend data, and to obtain comprehensive risk level data based on power health status data and vehicle operation history data. The strategy generation and resource allocation unit is used to generate an adaptive preheating strategy based on the comprehensive risk level data to obtain tiered preheating strategy data; and to dynamically allocate preheating resources based on the tiered preheating strategy data to obtain a preheating resource allocation scheme. An energy optimization unit is used to acquire real-time energy status data of the vehicle power system, and to perform energy constraint optimization on the preheating resource allocation scheme based on the real-time energy status data to obtain an energy-optimized preheating scheme. The control learning and execution unit is used to perform online learning of the control strategy of the energy optimization preheating scheme to obtain control strategy data; and to perform coordinated control for emergency vehicle start-up based on the control strategy data.

[0006] Preferably, the method for acquiring dynamic temperature field data of the vehicle power supply operating environment in real time through a multi-source temperature sensing network to obtain ambient temperature field distribution data includes: Collect raw temperature data; Based on the preset battery structure thermal sensitivity model, regional thermal load analysis is performed on the original temperature data, the temperature measurement points are dynamically divided into core monitoring areas and auxiliary monitoring areas, a weighted temperature measurement point topology map is constructed, and graded temperature measurement topology data is obtained. Based on the hierarchical temperature measurement topology data, the original temperature data is spatiotemporally registered and weighted fused to obtain preliminary fused temperature field data. An environmental interference factor is introduced to dynamically compensate and correct the preliminary fused temperature field data, thereby obtaining environmental temperature field distribution data.

[0007] Preferably, the method for predicting the temperature field evolution trend based on the environmental temperature field distribution data, to obtain temperature field evolution trend data, includes: Multi-scale time-domain fluctuation pattern decomposition was performed on the environmental temperature field distribution data to identify periodic steady-state fluctuations, event-driven transient shocks, and random noise disturbances, thereby obtaining time-domain fluctuation pattern data. Based on the time-domain fluctuation mode data, cross-modal spatiotemporal correlation analysis is performed. For different fluctuation modes, the heat conduction feature vector in the spatial topology of the vehicle power system is extracted to obtain spatiotemporal correlation feature vector data. Based on the spatiotemporal correlation feature vector data, the future state is recursively projected through a long short-term memory network, and different feature vectors are evolved and deduced to construct the temperature field evolution path and obtain temperature field evolution trend data.

[0008] Preferably, the method for performing risk coupling assessment based on the temperature field evolution trend data, and obtaining comprehensive risk level data based on power supply health status data and vehicle operation history data, includes: Based on the temperature field evolution trend data, potential failure modes are mapped to identify and quantify various low-temperature physicochemical failure risks, thereby obtaining failure mode scenario data. Based on the failure mode scenario data and the power supply health status data, the current health status of the power supply is projected onto the multi-dimensional failure mode space, and its safety margin with the boundary of each failure mode is quantified to obtain scenario-related vulnerability data. Based on the scenario-related vulnerability data and the vehicle operation history data, the historical operation mode is analyzed and probability weighted to obtain risk scenario probability data. Based on the probability data of the risk scenarios, nonlinear risk coupling calculations are performed on the vulnerability data associated with the scenarios to generate comprehensive risk level data.

[0009] Preferably, an adaptive preheating strategy is generated based on the comprehensive risk level data to obtain tiered preheating strategy data, including: The comprehensive risk level data is deconstructed to identify the core driving factors that lead to the current risk level, and these factors are quantified into multi-dimensional risk feature vectors to obtain risk feature vector data. Based on the risk feature vector data, the optimal preheating strategy prototype is matched and selected from the preset strategy knowledge base to obtain preheating strategy prototype data. Based on the prototype data of the preheating strategy, an adaptive strategy parameter space is constructed with preheating power, area of ​​action, and intervention duration as key dimensions, and strategy parameter space data is obtained. Based on the strategy parameter space data, a risk-energy consumption dual-objective Pareto optimization is performed to determine the optimal parameter combination and generate graded preheating strategy data.

[0010] Preferably, a preheating resource allocation scheme is obtained by dynamically allocating preheating resources based on the tiered preheating strategy data, including: The hierarchical preheating strategy data is analyzed in time and space to decompose it into the target energy injection distribution in the power system topology, thereby obtaining the preheating energy demand distribution data. Based on the preheating energy demand distribution data, a topology map of available preheating resources is constructed with preheating execution units as nodes and heat conduction paths as weighted edges. The status of each node is calibrated in real time to obtain available preheating resource topology data. Based on the available preheating resource topology data, an ant colony optimization algorithm is used for multi-path collaborative energy scheduling to plan the optimal resource matching path for the preheating energy demand distribution data, thereby obtaining resource scheduling path planning data. Based on the resource scheduling path planning data, an instruction set is generated to form a preheating resource allocation scheme, wherein the instruction set includes specific execution units, start-stop timing sequences, and power curves.

[0011] Preferably, the method for acquiring real-time energy state data of the vehicle power system and optimizing the preheating resource allocation scheme based on the real-time energy state data to obtain an energy-optimized preheating scheme includes: By integrating real-time electrochemical measurements and aging model deduction with a multi-dimensional state observer, the state of charge, health, and power state of the power system are comprehensively evaluated to obtain real-time energy state data. Based on the real-time energy state data, a forward-looking simulation of energy consumption is performed on the preheating resource allocation scheme to predict the instantaneous impact and long-term effects of implementing the preheating resource allocation scheme on the key performance indicators of the power system, and to obtain preheating energy cost data. Based on the preheating energy cost data and the minimum energy threshold for emergency vehicle start-up, a multi-level energy safety constraint boundary is dynamically constructed to obtain dynamic energy constraint data. Based on the dynamic energy constraint data, the power allocation, action timing and resource combination in the preheating resource allocation scheme are iteratively optimized to minimize preheating energy consumption while ensuring start-up capability, thereby generating an energy-optimized preheating scheme.

[0012] Preferably, the method for performing online learning of control strategies for the energy-optimized preheating scheme to obtain control strategy data includes: The energy optimization preheating scheme is deconstructed, the real-time system state is fused, reinforcement learning state features are constructed, and state feature vector data is obtained. Based on the state feature vector data, control actions are generated through a policy network and interacted with the system in real time to calculate the execution efficiency and target deviation, thereby obtaining real-time reward signal data. Based on the real-time reward signal data, the control strategy network is iteratively optimized online to obtain iteratively optimized control strategy data. The final control logic is extracted from the iteratively optimized control strategy data to form control strategy data that can autonomously decide the optimal control sequence under any state.

[0013] Preferably, the coordinated control for vehicle emergency start based on the control strategy data includes: Based on the control strategy data, the control logic is instantiated in real time to generate the optimal control timing and execution path for the current vehicle operating condition, thereby obtaining real-time control instruction sequence data. Based on the real-time control instruction sequence data, multi-system resource collaborative arbitration is performed to dynamically allocate control rights and priorities to each preheating and startup execution unit, thereby obtaining collaborative execution logic gating data. Based on the collaborative execution logic gating data, closed-loop tracking and state feedback correction of instructions are performed, and control parameters are dynamically fine-tuned according to the instantaneous changes in system response to obtain adaptive correction instruction data. Based on the adaptive correction instruction data, the final instruction is sent to the underlying hardware execution unit and the startup success flag is monitored to generate collaborative control task completion status data.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This invention obtains global temperature information by constructing a multi-source temperature sensing network and a dynamic temperature field. It uses algorithms such as long short-term memory networks to predict the evolution trend of the temperature field, transforming the control logic from a passive mode of "reheating when the temperature is below the threshold" to an active mode of "intervening in advance when the risk of low temperature is predicted in the future". This avoids the response lag problem of traditional methods and enables more timely and accurate thermal management. This invention integrates temperature field evolution trends, battery state of health (SOH), and vehicle operating history data to concretize the abstract low-temperature environment into quantifiable physicochemical failure risks such as lithium plating and internal resistance surges, thereby enabling multi-dimensional comprehensive evaluation. This allows the system to accurately identify the actual risk levels faced by batteries with different aging degrees under different operating conditions, thereby generating differentiated preheating strategies, improving the effectiveness of preheating, and ensuring the long-term safety and lifespan of the battery. This invention employs a dual-objective Pareto optimization based on risk and energy consumption, energy constraint optimization, and deep reinforcement learning. This allows the system to move beyond executing fixed heating levels and instead plan preheating power, area, and duration based on real-time comprehensive risk levels. Furthermore, it utilizes an ant colony optimization algorithm to achieve optimal resource allocation, ensuring that unnecessary battery energy consumption is minimized without sacrificing emergency start-up capabilities. Additionally, the online learning capability of the near-end strategy optimization algorithm ensures long-term high efficiency and reliability. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0016] In the diagram: 1. Temperature acquisition unit; 2. Trend prediction unit; 3. Risk assessment unit; 4. Strategy generation and resource allocation unit; 5. Energy optimization unit; 6. Control learning and execution unit. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This invention provides a technical solution: a low-temperature emergency start-up power supply control system, comprising: Temperature acquisition unit 1 is used to acquire dynamic temperature field data of the vehicle power supply operating environment in real time through a multi-source temperature sensing network to obtain ambient temperature field distribution data. Trend prediction unit 2 is used to predict the evolution trend of the temperature field based on the environmental temperature field distribution data, and obtain the temperature field evolution trend data. Risk assessment unit 3 is used to conduct risk coupling assessment based on temperature field evolution trend data, and to obtain comprehensive risk level data based on power health status data and vehicle operation history data. The strategy generation and resource allocation unit 4 is used to generate adaptive preheating strategies based on comprehensive risk level data to obtain hierarchical preheating strategy data; and to dynamically allocate preheating resources based on hierarchical preheating strategy data to obtain a preheating resource allocation scheme. Energy optimization unit 5 is used to acquire real-time energy status data of the vehicle power system, and to perform energy constraint optimization on the preheating resource allocation scheme based on the real-time energy status data to obtain an energy-optimized preheating scheme. The control learning and execution unit 6 is used to learn the control strategy of the energy optimization preheating scheme online and obtain control strategy data; and to perform coordinated control for emergency vehicle start-up based on the control strategy data.

[0019] In this invention, the power health status data is obtained from the vehicle's battery management system (BMS), while vehicle operating history data is collected from the vehicle control unit (VCU) or the onboard data bus. The power health status data refers to a set of key indicators characterizing the aging degree and performance degradation of the battery system, not a single numerical value. Its core indicators include State of Health (SOH), which is the ratio of the battery's current maximum usable capacity to its factory rated capacity; internal resistance, reflecting the degradation of the battery's power output capability; and self-discharge rate, etc. This data is typically obtained by the battery management system (BMS) through analysis of long-term charge and discharge history and the use of algorithms such as extended Kalman filtering. Online estimation is a key basis for assessing the battery's "quality"; vehicle operation history data refers to time series data that records the vehicle's recent operating modes. Key information includes charging history (such as the last charging time and whether the charging method is fast or slow charging), driving mode (such as whether the recent driving was high-intensity or gentle), and resting time (such as how long the vehicle has been parked in low temperatures). The significance of this data is that it can reflect the battery's energy and thermal stress history before the current moment. For example, a car that has just finished a long-distance drive and a car that has been parked in ice and snow for three days will have completely different internal thermodynamic states and starting risks, even if the current external temperature is the same.

[0020] In one optional embodiment, dynamic temperature field data of the vehicle power supply operating environment is collected in real time through a multi-source temperature sensing network to obtain ambient temperature field distribution data, including: Raw temperature data is collected simultaneously using thermal imaging sensors, contact temperature sensors, and non-contact infrared sensors. Based on the preset battery structure thermal sensitivity model, regional thermal load analysis is performed on the original temperature data, the temperature measurement points are dynamically divided into core monitoring areas and auxiliary monitoring areas, a weighted temperature measurement point topology map is constructed, and hierarchical temperature measurement topology data is obtained. Based on the hierarchical temperature measurement topology data, the original temperature data is spatiotemporally registered and weighted and fused to obtain preliminary fused temperature field data; An environmental interference factor is introduced to dynamically compensate and correct the preliminary fused temperature field data, resulting in environmental temperature field distribution data.

[0021] It should be noted that the multi-source temperature sensing network refers to the use of a combination of various types of sensors to achieve three-dimensional temperature monitoring of the battery pack at points, surfaces, and space, in order to construct a high-resolution temperature field and overcome the shortcomings of traditional sparse temperature measurement points that cannot reflect the uneven temperature distribution. The battery structure thermal sensitivity model is a pre-established digital model that describes the different sensitivities to temperature changes at different locations inside the battery pack due to differences in structure and electrochemical characteristics. This model is usually based on finite element analysis (FEA) thermal simulation and calibrated with a large amount of experimental data. The core monitoring area and auxiliary monitoring area are dynamically divided based on this model, aiming to make the system focus more on key areas such as the thermal runaway initiation point and high-power cells, and to achieve optimized allocation of resources and computing power. Environmental interference factors refer to external factors such as outside temperature, wind speed, and solar radiation that have a real impact on the heat exchange of the battery pack. Introducing their quantitative data is to make the final temperature field more realistically reflect the comprehensive thermal environment of the vehicle. A topology graph consists of "nodes" and "edges" connecting the nodes, used to represent the relationships between entities. In this system, a node represents a pixel area of ​​each physical temperature sensor (whether it's a contact, infrared, or thermal imaging sensor), and an edge represents the heat conduction path and efficiency between two temperature sensing points (nodes). For example, the "edge" between two adjacent temperature sensing points on a battery cell is very strong, while the "edge" between a temperature sensing point in the center of the battery pack and one on the outer casing might be weak because heat transfer is slow. The weight of a node represents the importance of that temperature sensing point. Based on the "battery structure thermal sensitivity model," the system divides certain areas into "core monitoring areas" (e.g., cell locations that are prone to overheating or most sensitive to low temperatures). Temperature measurement points located in these core areas have their nodes assigned a high weight. This means that the system will "trust" or "pay more attention" to the data of these high-weight nodes when performing any analysis. The weight of an edge represents the tightness of the thermal connection between two nodes. The higher the weight, the faster and easier the heat is transferred between the two points. This weight is determined by factors such as the physical structure of the battery pack and the thermal conductivity of the materials. Spatiotemporal registration includes spatial registration and temporal alignment. Spatial registration maps readings from different sensors to a unified three-dimensional coordinate system. For example, a frame of a thermal imaging camera contains the temperature of tens of thousands of pixels; spatial registration precisely maps each pixel to its physical location on the surface of a battery pack. Similarly, the readings of contact sensors must be located to the precise coordinates of the cell they are in contact with. Since different sensors may have different sampling frequencies, temporal registration uses interpolation or synchronization algorithms to align the data from all sensors to a unified time axis. After all data has been aligned through spatiotemporal registration, additional... The weight fusion step is the process of merging calculations. The "weights" here come from the temperature measurement point topology map generated in the first step. When the system calculates the precise temperature of a specific location, it will comprehensively consider the readings of multiple sensors around it. For example, the readings from sensors in the "core monitoring area" (with high node weights) have a greater say in the calculation. Readings from more reliable or more accurate sensors can also be given higher weights. It will also utilize the "edge" information in the topology map. If there are no sensors in a certain area, the system can perform more accurate interpolation estimation based on the readings of its adjacent, thermally closely connected (with high edge weights) sensors.

[0022] In an optional embodiment, temperature field evolution trend prediction is performed based on ambient temperature field distribution data to obtain temperature field evolution trend data, including: Wavelet transform algorithm is used to decompose the environmental temperature field distribution data into multi-scale time-domain fluctuation patterns, identify periodic steady-state fluctuations, event-driven transient shocks and random noise disturbances, and obtain time-domain fluctuation pattern data; Based on time-domain fluctuation mode data, cross-modal spatiotemporal correlation analysis is performed. For different fluctuation modes, the heat conduction feature vectors in the spatial topology of the vehicle power system are extracted to obtain spatiotemporal correlation feature vector data. Based on spatiotemporal correlated feature vector data, the future state is recursively projected through a long short-term memory network, and different feature vectors are used to perform evolutionary deduction to construct the temperature field evolution path and obtain temperature field evolution trend data.

[0023] It should be noted that the wavelet transform algorithm is used here to decompose the complex and disordered original temperature time series data to distinguish different types of temperature changes (e.g., slow diurnal variations and rapid, accelerated heat generation), facilitating subsequent targeted modeling. The heat conduction feature vector is a dataset containing physical parameters such as thermal conductivity, thermal capacity gradient, and spatial adjacency, describing the inherent characteristics of heat conduction within the battery pack according to physical laws. The Long Short-Term Memory (LSTM) network is a recurrent neural network particularly adept at processing and predicting time series data. This network was chosen because it can learn the long-term dependence of temperature changes, thereby making more accurate recursive predictions of future temperature field evolution. The initial training of this model relies on historical temperature field data collected from a large number of different vehicle models under various climatic conditions, including driving and stationary conditions, and a clear prediction time window (e.g., the next 30 minutes) is set. At the same time, the model will periodically use updated parameters from the cloud server for online adaptive learning to cope with model drift problems caused by vehicle aging or environmental changes.

[0024] In one optional embodiment, risk coupling assessment is performed based on temperature field evolution trend data, and comprehensive risk level data is obtained based on power supply health status data and vehicle operation history data, including: Based on temperature field evolution trend data, potential failure modes are mapped to identify and quantify various low-temperature physicochemical failure risks, resulting in failure mode scenario data. Based on failure mode scenario data and power supply health status data, the current health status of the power supply is projected onto a multi-dimensional failure mode space, and its safety margin with the boundary of each failure mode is quantified to obtain scenario-related vulnerability data. Based on scenario-related vulnerability data and vehicle operation history data, a Markov model is used to analyze historical operation patterns and assign probabilistic weights to obtain risk scenario probability data. Based on risk scenario probability data, a pre-set neural network model is used to perform nonlinear risk coupling calculations on scenario-related vulnerability data to generate comprehensive risk level data.

[0025] It's important to note that the potential failure mode mapping (PMMA) correlates a predicted temperature value (e.g., -20°C) with a probabilistic model of specific electrochemical side reactions that might occur at that temperature, such as lithium plating, electrolyte solidification, and a sharp increase in internal resistance leading to a sudden drop in usable power. This transforms physical prediction into risk prediction. "Safety margin" is not a single value but a set of quantitative indicators. For example, for the risk of "insufficient power," the safety margin could be the ratio of predicted usable power to the minimum start-up threshold power. For the risk of "lithium plating," the safety margin could be the difference between the current negative electrode potential and the lithium plating potential threshold. The safety margin can also include a time dimension, such as "how long is expected to reach the failure boundary under the current temperature decrease trend (Time-to-Failure)," which provides a direct calculation basis for the "lead time" of the preheating strategy. The preset neural network model preferably adopts a graph neural network (GNN) or a Transformer architecture. GNN can handle the topological relationships (heat conduction paths) between various components within the battery system well, while Transformer excels at capturing the mutual influence weights between different risk factors (attention mechanism). Scene-related vulnerability data refers to the ability of a battery to resist specific failure modes after comprehensively assessing its current state of health (SOH). For example, a severely aged battery is much more vulnerable at -10°C than a new battery. Markov models are used to analyze vehicle historical data to identify vehicles that may be in a state of "long-term inactivity". Different scenarios, such as "first start after initial startup" or "restart after short-distance driving," are assigned different probabilities. Nonlinear risk coupling calculation refers to the comprehensive calculation of risk information (temperature, health status, scenario) from multiple dimensions through neural networks. It can capture the complex mutual reinforcement effects between various risk factors, thereby generating a more accurate comprehensive risk level than simple weighted summation. This comprehensive risk level data is not a simple single value, but a structured data package, which includes a main risk score (such as 0-100 points) and decomposed scores of various sub-risks (such as lithium plating risk, insufficient power risk), providing a more refined decision-making basis for subsequent strategy generation.

[0026] In an optional embodiment, an adaptive preheating strategy is generated based on comprehensive risk level data to obtain tiered preheating strategy data, including: Risk feature deconstruction is performed on comprehensive risk level data to identify the core driving factors that lead to the current risk level, and these factors are quantified into multi-dimensional risk feature vectors to obtain risk feature vector data. Based on risk feature vector data, the optimal preheating strategy prototype is matched and selected from the preset strategy knowledge base to obtain preheating strategy prototype data. Based on the prototype data of the preheating strategy, an adaptive strategy parameter space was constructed with preheating power, area of ​​action, and intervention duration as key dimensions, and the strategy parameter space data was obtained. Based on the strategy parameter space data, Pareto optimization with risk and energy consumption as dual objectives is performed to determine the optimal parameter combination and generate graded preheating strategy data.

[0027] It should be noted that the risk feature deconstruction aims to explore the root causes behind high-risk levels (e.g., whether the overall temperature is too low or the local temperature difference is too large) in order to address the problem effectively. The preheating strategy prototype is a typical preheating pattern pre-existing in the knowledge base, such as "pulsed high-power preheating" or "continuous low-power heat preservation". The system selects the most suitable prototype as the basis by matching the risk feature vector. This knowledge base is scalable and can be updated through the OEM's OTA (over-the-air) service to introduce more advanced or climate-adaptive preheating strategies. The adaptive strategy parameter space is a multi-dimensional space composed of key preheating control variables (power, area, duration). The final generated "tiered preheating strategy data" is not just a set of fixed parameters, but a phased action plan. For example: the first phase uses high-power pulses to quickly raise the temperature of key areas; the second phase uses medium power to act on a wider area to reduce the temperature difference; and the third phase uses low power to maintain the temperature. Risk-energy bi-objective Pareto optimization is an optimization method used to find a series of optimal solutions in a parameter space. These solutions achieve the best balance between the two conflicting objectives of "risk reduction degree" and "energy consumption amount". It includes the quantification of "risk reduction degree" and "energy consumption amount". The quantification of "risk reduction degree" can be calculated through a "forward-looking risk simulation". That is, for each strategy combination (power P, region A, duration T) in the parameter space, the system calls the trend prediction unit 2 and risk assessment unit 3 to perform a rapid virtual simulation to predict "if this strategy is implemented, what will be the new comprehensive risk level after duration T". The quantification of "energy consumption amount" can not only directly calculate power × duration, but also introduce a concept of "comprehensive cost". For example, a weight of "component health loss" can be added. Frequent use of maximum power for preheating, although fast, may accelerate the aging of heating elements. In this case, the comprehensive energy cost = power consumption + α × equivalent loss of component aging, where α is an adjustable weight coefficient.

[0028] In an optional embodiment, preheating resources are dynamically allocated based on tiered preheating strategy data to obtain a preheating resource allocation scheme, including: Spatiotemporal analysis was performed on the data of the graded preheating strategy to decompose it into the target energy injection distribution in the power system topology, thus obtaining the preheating energy demand distribution data. Based on the distribution data of preheating energy demand, a topology map of available preheating resources is constructed with preheating execution units as nodes and heat conduction paths as weighted edges. The status of each node is marked in real time to obtain the topology data of available preheating resources. Based on the available preheating resource topology data, an ant colony optimization algorithm is used for multi-path collaborative energy scheduling to plan the optimal resource matching path for the preheating energy demand distribution data, and obtain resource scheduling path planning data. Based on resource scheduling path planning data, an instruction set containing specific execution units, start-stop timings, and power curves is generated to form a preheating resource allocation scheme.

[0029] It should be noted that the core of "spatiotemporal analysis" is a reverse calculation process of physical modeling. The input strategy is "heat region A of the battery pack from -20℃ to -5℃ and region B to -10℃ within time T". The system will accurately calculate, based on the pre-calibrated battery thermodynamic model (including the specific heat capacity, mass, thermal conductivity, etc. of each region), the net energy injection of X joules into region A and Y joules into region B required to achieve this temperature increase (ΔT). The final "preheating energy demand distribution data" is not a total energy value, but an "energy demand heat map". This digital map accurately marks the number of joules that each three-dimensional spatial unit (Voxel) in the battery pack topology needs to absorb. Some regions may require a large amount of energy injection (highlighted demand), while some regions require little or no energy (dim demand). The available preheating resource topology map is a digital system model where "nodes" represent all available heating elements on the vehicle (such as PTC heating elements and heating films), including reversible components such as the drive motor or the heat pump of the air conditioning system. "Weighted edges" represent the efficiency and paths of heat transfer from these elements to different locations in the battery pack. The ant colony optimization algorithm, a heuristic algorithm simulating ants finding food paths, is used here to find the optimal energy delivery path in this topology map. That is, how to combine different heating elements to accurately transfer the required heat to the target area with the highest efficiency and speed, achieving intelligent and refined scheduling of preheating resources. The output of the ant colony algorithm is not "select heater A," but rather "allocate 60% of the energy to heater A, 30% to heater B, and 10% to the drive motor." This resource combination scheme addresses the core problem of how to combine multiple heat sources so that the combined heat output perfectly matches the "energy demand heat map" generated in the first step. While it's possible that using any single heater alone cannot efficiently cover all areas requiring heating, through the algorithm's collaborative scheduling, the most uniform and accurate heating target can be achieved with minimal overall energy consumption and in the shortest time. Real-time calibration of each node's status includes not only its availability but also dynamic parameters such as its current power limitations, thermal response delay, and aging decay coefficient, ensuring that the algorithm plans based on the most realistic system state.

[0030] In an optional embodiment, the method involves acquiring real-time energy state data of the vehicle's power system, and optimizing the preheating resource allocation scheme based on the real-time energy state data using energy constraints to obtain an energy-optimized preheating scheme, including: By integrating real-time electrochemical measurements and aging model deduction with a multi-dimensional state observer, the state of charge, health, and power state of the power system are comprehensively evaluated to obtain real-time energy state data. Based on real-time energy state data, a forward-looking simulation of energy consumption is performed on the preheating resource allocation scheme to predict the instantaneous impact and long-term effects of implementing the preheating resource allocation scheme on the key performance indicators of the power system, and to obtain preheating energy cost data. Based on the preheating energy cost data and the minimum energy threshold for emergency vehicle start-up, a multi-level energy safety constraint boundary is dynamically constructed to obtain dynamic energy constraint data. Based on dynamic energy constraint data, the power allocation, action timing and resource combination in the preheating resource allocation scheme are iteratively optimized to minimize preheating energy consumption while ensuring start-up capability, and generate an energy-optimized preheating scheme.

[0031] It should be noted that the multidimensional state observer is an algorithm that accurately estimates the internal state of a battery by fusing sensor data and battery models, and its results are more reliable than single sensor readings. Real-time electrochemical measurement is not just about reading a percentage of charge; the observer receives a series of the most primitive and fundamental physical signals in real time, mainly including: the instantaneous value of the battery terminal voltage, the precise ampere of the inflow / outflow current, and the surface temperature of key temperature measurement points in the battery pack. The aging model extrapolation is a dynamic digital archive stored in the BMS that records the battery's "life cycle" from the time it leaves the factory to the present (total charge and discharge ampere-hours, extreme temperatures experienced, number of high-rate charge and discharge cycles, etc.). Based on this history, it extrapolates the "physical" parameters of the current battery that cannot be directly measured, such as how much the current actual usable capacity has decayed and how many milliohms the internal resistance has increased compared to when it left the factory. The three are then fused using an extended Kalman filter estimation algorithm. Energy consumption prospective simulation refers to a rapid virtual exercise conducted internally by the system before preheating to predict how much electricity the preheating scheme will consume. The multi-level energy safety constraint boundary is a set of dynamic safety rules. The "hard constraint" is the bottom line that the remaining electricity after preheating must be sufficient to start the vehicle, while the "soft constraint" is to avoid excessive consumption to protect battery life while meeting the hard constraint. The iterative optimization process is to repeatedly fine-tune the preheating scheme under the supervision of this safety boundary until a final scheme that meets the preheating target, strictly adheres to the safety bottom line, and has the lowest energy consumption is found. The minimum energy threshold for emergency vehicle start-up is itself a dynamic variable. It is calculated by the system based on the current battery SOH, the estimated internal resistance at ambient temperature, and the load characteristics of the starter motor (or drive system), rather than a fixed value.

[0032] In an optional embodiment, the method for performing online learning of control strategies for an energy-optimized preheating scheme to obtain control strategy data includes: The energy optimization preheating scheme is deconstructed into objectives, the real-time state of the system is integrated, reinforcement learning state features are constructed, and state feature vector data is obtained. Based on state feature vector data, control actions are generated through a policy network and interacted with the system in real time to calculate execution efficiency and target deviation, thereby obtaining real-time reward signal data. Based on real-time reward signal data, the control policy network is iteratively optimized online using a near-end policy optimization algorithm to obtain iteratively optimized control policy data. The final control logic is extracted from the iteratively optimized control strategy data to form control strategy data that can autonomously decide the optimal control sequence under any state.

[0033] It's important to note that the "energy optimization preheating plan" generated in the previous step is a macroscopic, open-loop script (e.g., "Heat the core area of ​​the battery to -5°C within 3 minutes using a total energy of no more than 80Wh"). "Target deconstruction" translates this macroscopic target into a microscopic reference system that the AI ​​can understand at every control instant (e.g., every 100 milliseconds). It doesn't directly hand the plan to the AI ​​for execution, but rather transforms it into a "dynamic bullseye" for the AI ​​to evaluate the quality of its actions. This bullseye information is integrated into the state feature vector. Reinforcement learning state features refer to the input information provided to the AI ​​for decision-making, including a series of key system parameters such as current temperature, target temperature, remaining battery power, and available preheating power. The reward function is the rule used to evaluate the quality of the AI's control actions; for example, quickly and energy-efficiently reaching the target temperature will... High rewards are given, while penalties are imposed for low rewards. The specific form is usually a weighted function that comprehensively considers multiple dimensions such as heating rate, energy consumption, temperature uniformity, and the final error from the target temperature. The weighting coefficients are dynamically adjusted according to the risk level. This function is the core guiding the direction of AI learning. The Proximal Policy Optimization (PPO) algorithm is an efficient and stable reinforcement learning algorithm that allows control policies to be safely and smoothly updated and self-evolved online during the daily operation of the vehicle, rather than using a fixed, unchanging logic from the factory. To ensure the safety of online learning, the update process uses a "shadow mode," where the new policy runs and is evaluated in parallel with the old policy in the background. Only when its performance is verified to be superior to the old policy in multiple simulation scenarios will it be officially deployed, preventing immature policies from affecting driving safety.

[0034] In an optional embodiment, coordinated control for vehicle emergency start-up based on control strategy data includes: Based on the control strategy data, the control logic is instantiated in real time to generate the optimal control timing and execution path for the current vehicle operating condition, and obtain real-time control command sequence data. Based on real-time control instruction sequence data, multi-system resource collaborative arbitration is performed to dynamically allocate control rights and priorities to each preheating and startup execution unit, thereby obtaining collaborative execution logic gating data. Based on the gated data of the collaborative execution logic, closed-loop tracking and state feedback correction of instructions are performed, and control parameters are dynamically fine-tuned according to the instantaneous changes in the system response to obtain adaptive correction instruction data. Based on adaptive correction instruction data, the final instruction is sent to the underlying hardware execution unit and the startup success flag is monitored to generate the number of collaborative control task completion statuses.

[0035] It should be noted that real-time instantiation of control logic refers to applying the learned general control model to the current specific vehicle state to generate a tailored, immediately executable sequence of actions; the collaborative execution logic gating data is a set of dynamic priority and permission allocation rules, which acts like a traffic commander, coordinating multiple units such as the preheating system, starting system, and power management system to ensure that they work together in the right way at the right time and avoid conflicts; closed-loop tracking and state feedback correction of instructions means that the system will continuously monitor the execution effect after issuing instructions (e.g., whether the temperature rises at the expected rate), and once a deviation is detected, it will fine-tune subsequent instructions in real time to ensure that the entire process is always in a state of precise control.

[0036] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A low-temperature emergency start-up power supply control system, characterized in that, include: Temperature acquisition unit (1) is used to collect dynamic temperature field data of the vehicle power supply operating environment in real time through a multi-source temperature sensing network to obtain ambient temperature field distribution data. The trend prediction unit (2) is used to predict the temperature field evolution trend based on the environmental temperature field distribution data, and obtain temperature field evolution trend data. The risk assessment unit (3) is used to perform risk coupling assessment based on the temperature field evolution trend data, and to obtain comprehensive risk level data based on power health status data and vehicle operation history data. The strategy generation and resource allocation unit (4) is used to generate an adaptive preheating strategy based on the comprehensive risk level data to obtain hierarchical preheating strategy data; and to dynamically allocate preheating resources based on the hierarchical preheating strategy data to obtain a preheating resource allocation scheme. The energy optimization unit (5) is used to acquire real-time energy status data of the vehicle power system, and to perform energy constraint optimization on the preheating resource allocation scheme based on the real-time energy status data to obtain an energy-optimized preheating scheme. The control learning and execution unit (6) is used to perform online learning of the control strategy of the energy optimization preheating scheme to obtain control strategy data; and to perform coordinated control of vehicle emergency start based on the control strategy data. Among them, risk coupling assessment is performed based on the temperature field evolution trend data, and comprehensive risk level data is obtained based on power supply health status data and vehicle operation history data, including: Based on the temperature field evolution trend data, potential failure modes are mapped to identify and quantify various low-temperature physicochemical failure risks, thereby obtaining failure mode scenario data. Based on the failure mode scenario data and the power supply health status data, the current health status of the power supply is projected onto the multi-dimensional failure mode space, and its safety margin with the boundary of each failure mode is quantified to obtain scenario-related vulnerability data. Based on the scenario-related vulnerability data and the vehicle operation history data, the historical operation mode is analyzed and probability weighted to obtain risk scenario probability data. Based on the probability data of the risk scenarios, nonlinear risk coupling calculations are performed on the vulnerability data associated with the scenarios to generate comprehensive risk level data. Based on the aforementioned hierarchical preheating strategy data, preheating resources are dynamically allocated to obtain a preheating resource allocation scheme, including: The hierarchical preheating strategy data is analyzed in time and space to decompose it into the target energy injection distribution in the power system topology, thereby obtaining the preheating energy demand distribution data. Based on the preheating energy demand distribution data, a topology map of available preheating resources is constructed with preheating execution units as nodes and heat conduction paths as weighted edges. The status of each node is calibrated in real time to obtain available preheating resource topology data. Based on the available preheating resource topology data, an ant colony optimization algorithm is used for multi-path collaborative energy scheduling to plan the optimal resource matching path for the preheating energy demand distribution data, thereby obtaining resource scheduling path planning data. Based on the resource scheduling path planning data, an instruction set is generated to form a preheating resource allocation scheme, wherein the instruction set includes specific execution units, start-stop timing sequences, and power curves; Acquire real-time energy state data of the vehicle power system, and optimize the preheating resource allocation scheme based on the real-time energy state data to obtain an energy-optimized preheating scheme, including: By integrating real-time electrochemical measurements and aging model deduction with a multi-dimensional state observer, the state of charge, health, and power state of the power system are comprehensively evaluated to obtain real-time energy state data. Based on the real-time energy state data, a forward-looking simulation of energy consumption is performed on the preheating resource allocation scheme to predict the instantaneous impact and long-term effects of implementing the preheating resource allocation scheme on the key performance indicators of the power system, and to obtain preheating energy cost data. Based on the preheating energy cost data and the minimum energy threshold for emergency vehicle start-up, a multi-level energy safety constraint boundary is dynamically constructed to obtain dynamic energy constraint data. Based on the dynamic energy constraint data, the power allocation, action timing and resource combination in the preheating resource allocation scheme are iteratively optimized to minimize preheating energy consumption while ensuring start-up capability, thereby generating an energy-optimized preheating scheme.

2. The low-temperature emergency start-up power supply control system according to claim 1, characterized in that, This is used to collect dynamic temperature field data of the vehicle power supply operating environment in real time through a multi-source temperature sensing network, obtaining ambient temperature field distribution data, including: Collect raw temperature data; Based on the preset battery structure thermal sensitivity model, regional thermal load analysis is performed on the original temperature data, the temperature measurement points are dynamically divided into core monitoring areas and auxiliary monitoring areas, a weighted temperature measurement point topology map is constructed, and graded temperature measurement topology data is obtained. Based on the hierarchical temperature measurement topology data, the original temperature data is spatiotemporally registered and weighted fused to obtain preliminary fused temperature field data. An environmental interference factor is introduced to dynamically compensate and correct the preliminary fused temperature field data, thereby obtaining environmental temperature field distribution data.

3. The low-temperature emergency start-up power supply control system according to claim 2, characterized in that, This is used to predict the temperature field evolution trend based on the environmental temperature field distribution data, resulting in temperature field evolution trend data, including: Multi-scale time-domain fluctuation pattern decomposition is performed on the environmental temperature field distribution data to identify periodic steady-state fluctuations, event-driven transient shocks, and random noise disturbances, thereby obtaining time-domain fluctuation pattern data. Based on the time-domain fluctuation mode data, cross-modal spatiotemporal correlation analysis is performed. For different fluctuation modes, the heat conduction feature vector in the spatial topology of the vehicle power system is extracted to obtain spatiotemporal correlation feature vector data. Based on the spatiotemporal correlation feature vector data, the future state is recursively projected through a long short-term memory network, and different feature vectors are evolved and deduced to construct the temperature field evolution path and obtain temperature field evolution trend data.

4. A low-temperature emergency start-up power supply control system according to claim 3, characterized in that, An adaptive preheating strategy is generated based on the comprehensive risk level data, resulting in tiered preheating strategy data, including: The comprehensive risk level data is deconstructed to identify the core driving factors that lead to the current risk level, and these factors are quantified into multi-dimensional risk feature vectors to obtain risk feature vector data. Based on the risk feature vector data, the optimal preheating strategy prototype is matched and selected from the preset strategy knowledge base to obtain preheating strategy prototype data. Based on the prototype data of the preheating strategy, an adaptive strategy parameter space is constructed with preheating power, area of ​​action, and intervention duration as key dimensions, and strategy parameter space data is obtained. Based on the strategy parameter space data, a risk-energy consumption dual-objective Pareto optimization is performed to determine the optimal parameter combination and generate graded preheating strategy data.

5. A low-temperature emergency start-up power supply control system according to claim 4, characterized in that, Online learning of control strategies for the energy-optimized preheating scheme is used to obtain control strategy data, including: The energy optimization preheating scheme is deconstructed, the real-time system state is fused, reinforcement learning state features are constructed, and state feature vector data is obtained. Based on the state feature vector data, control actions are generated through a policy network and interacted with the system in real time to calculate the execution efficiency and target deviation, thereby obtaining real-time reward signal data. Based on the real-time reward signal data, the control strategy network is iteratively optimized online to obtain iteratively optimized control strategy data. The final control logic is extracted from the iteratively optimized control strategy data to form control strategy data that can autonomously decide the optimal control sequence under any state.

6. A low-temperature emergency start-up power supply control system according to claim 5, characterized in that, Coordinated control for vehicle emergency start-up based on the control strategy data includes: Based on the control strategy data, the control logic is instantiated in real time to generate the optimal control timing and execution path for the current vehicle operating condition, thereby obtaining real-time control instruction sequence data. Based on the real-time control instruction sequence data, multi-system resource collaborative arbitration is performed to dynamically allocate control rights and priorities to each preheating and startup execution unit, thereby obtaining collaborative execution logic gating data. Based on the collaborative execution logic gating data, closed-loop tracking and state feedback correction of instructions are performed, and control parameters are dynamically fine-tuned according to the instantaneous changes in system response to obtain adaptive correction instruction data. Based on the adaptive correction instruction data, the final instruction is sent to the underlying hardware execution unit and the startup success flag is monitored to generate collaborative control task completion status data.

Images