New energy equipment high-frequency wire harness dynamic connection optimization method based on reinforcement learning

Abstract

The invention aims to provide a reinforcement learning-based dynamic connection optimization method for high-frequency wire harnesses of new energy equipment, and relates to the technical field of new energy, and the method comprises the steps: 1, collecting the operation data of each battery module of the new energy equipment, and building a state vector in combination with the high-frequency wire harness characteristic data of each high-frequency wire harness in the new energy equipment; 2, calculating a predicted temperature distribution vector based on the state vector and the temperature distribution vector; based on the state vector, the high-frequency wire harness stress and the predicted temperature distribution vector, constructing a state space and an action space of a reinforcement learning model; 3, defining a reward function of the reinforcement learning model; and completing dynamic connection optimization of the high-frequency wire harness in an iterative optimization mode. According to the invention, the stability, the energy efficiency and the safety of the system are obviously improved, and the service life of the high-frequency wire harness and the connection point is effectively prolonged.

Description

Technical Field

[0001] The present invention relates to the field of new energy technology, and in particular to a method for dynamic connection optimization of high-frequency wiring harnesses of new energy equipment based on reinforcement learning. Background Art

[0002] In the field of new energy equipment, especially in high-frequency and high-power application scenarios such as electric vehicles, wind power generation, and solar energy storage systems, the use of high-frequency current is becoming more and more widespread because it has significant advantages in improving power density and reducing equipment volume. However, high-frequency current brings problems such as temperature rise, power loss, and mechanical stress in practical applications. These problems make the stability and life of high-frequency wiring harness connections the focus of the industry. The existing technical solutions still have shortcomings in dealing with the rapid changes in temperature and stress in high-frequency environments, and it is difficult to meet the operating requirements of new energy equipment in complex dynamic environments. Traditional high-frequency wiring harness optimization methods mainly rely on static optimization solutions based on electrical parameters. For example, existing studies often reduce power loss by reducing resistance or improving wiring harness materials. However, under high-frequency conditions, the influence of the current skin effect causes the current density on the surface of the wiring harness to increase sharply, resulting in an increase in effective resistance, which in turn exacerbates temperature rise and power loss. Although some studies have alleviated this problem by improving the conductivity and heat dissipation properties of materials, this improvement solution has limited effect when facing variable current loads and ambient temperatures, and it is difficult to adapt to dynamic changes under different working conditions. This static material optimization method cannot respond in real time to the temperature rise and increased power loss caused by load changes. Especially in an environment with drastic high-frequency load fluctuations, the adaptability of the traditional method is significantly insufficient.

[0003] In terms of temperature management, existing technologies mainly deal with the temperature rise caused by high-frequency current by improving heat dissipation materials or adding cooling devices. For example, liquid cooling, air cooling and high-efficiency thermal conductive materials are widely used in electric vehicles and energy storage systems to achieve heat dissipation of wiring harnesses and connection points. However, these solutions usually require a significant increase in the volume and power consumption of the cooling system, which is difficult to further optimize in a limited space. At the same time, most of the existing cooling solutions are passive heat dissipation, which cannot be dynamically adjusted according to real-time temperature changes, resulting in the heat dissipation performance being difficult to keep up when the current or ambient temperature suddenly rises, thereby causing overheating risks. In high-frequency environments, due to rapid and drastic temperature changes, passive heat dissipation systems are difficult to meet heat dissipation requirements, and overheating problems still plague high-frequency applications of new energy equipment. Especially in high-power density scenarios, traditional passive heat dissipation methods make it difficult to control temperature rise, affecting the stability and service life of the equipment. In addition, existing technical studies have also explored the impact of stress control on high-frequency wiring harnesses. The electromagnetic force fluctuations caused by high-frequency currents will induce mechanical stress inside the material and gradually accumulate into fatigue stress, which may cause the wiring harness material to break or the connection point to fail. Existing solutions usually use mechanical reinforcement methods or material improvements to increase the tensile strength of the wiring harness to cope with the stress caused by high-frequency currents. However, traditional material reinforcement solutions have limited effects under frequent electromagnetic fluctuations, especially when mechanical stress fluctuates frequently, the problem of material fatigue accumulation is still difficult to avoid. Due to the periodicity and irregularity of the electromagnetic stress caused by high-frequency currents, it is difficult for the stress enhancement methods in the existing technology to be adjusted in real time according to the changes in actual stress. Therefore, in practical applications, traditional stress control methods are difficult to adapt to the changing high-frequency environment, which can easily lead to stress accumulation in the wiring harness and connection points, thereby affecting the long-term stability of the system. Summary of the invention

[0004] In view of this, the main purpose of the present invention is to provide a method for dynamic connection optimization of high-frequency wiring harnesses of new energy equipment based on reinforcement learning, which realizes efficient temperature rise control, power consumption optimization and stress management by real-time monitoring of the electrical, thermal and stress states of the system and intelligently adjusting parameters such as current path, cooling power and contact resistance. This method utilizes the adaptive ability of the reinforcement learning model to dynamically adapt to high-frequency load changes, significantly improves the stability, energy efficiency and safety of the system, and effectively extends the service life of high-frequency wiring harnesses and connection points.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0006] A method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning, the method comprising:

[0007] Step 1: Collect the operating data of each battery module of the new energy equipment, combine the high-frequency wire harness characteristic data of each high-frequency wire harness in the new energy equipment, the connection point power loss data of each connection point, the temperature distribution vector and the current density vector, and establish a state vector;

[0008] Step 2: Based on the state vector and the temperature distribution vector, calculate the predicted temperature distribution vector; based on the elastic modulus tensor and mechanical strain tensor of the high-frequency wire harness material, calculate the high-frequency wire harness stress; based on the state vector, the high-frequency wire harness stress and the predicted temperature distribution vector, construct the state space and action space of the reinforcement learning model;

[0009] Step 3: Define the reward function of the reinforcement learning model; continuously update the Q value of the reinforcement learning model through iterative optimization until convergence, output the corresponding action space at this time, and complete the high-frequency wire harness dynamic connection optimization based on the action space.

[0010] Furthermore, the operating data of each battery module includes: battery module voltage, battery module current and battery module state of charge; the high-frequency wiring harness characteristic data includes: high-frequency wiring harness length, high-frequency wiring harness resistance and high-frequency wiring harness maximum allowable current; the connection point power loss data includes: connection point power loss and connection point rated power; the environmental data includes: temperature distribution vector and current density vector.

[0011] Furthermore, the state vector is expressed using the following formula:

[0012]

[0013] Where X(t) represents the state vector at time t; i, j, and k are all integer subscript indices; SOC i (t) represents the charge state of the ith battery module at time t; V i (t) represents the voltage of the i-th battery module at time t; I i (t) represents the current of the i-th battery module at time t; N b is the number of battery modules; N w is the number of high-frequency wiring harnesses; R j is the high-frequency wiring harness resistance of the jth high-frequency wiring harness; L j is the high-frequency wire bundle length of the j-th high-frequency wire bundle; N is the maximum allowable current of the high-frequency wiring harness of the jth high-frequency wiring harness; c is the number of connection points; is the connection point power loss of the kth connection point; is the rated power of the kth connection point; T(τ) is the temperature distribution vector; J(τ) is the current density vector; τ is the integration time variable.

[0014] Furthermore, in step 2, the predicted temperature distribution vector is calculated based on the state vector and the temperature distribution vector using the following formula:

[0015]

[0016] Wherein, T(t+Δt) is the predicted temperature distribution vector at time t+Δt obtained by calculation; T(t) is the temperature distribution vector at time t; Δt is the time step; C p is the specific heat capacity of the high-frequency wire harness material; ρ is the density of the high-frequency wire harness material; k is the thermal conductivity of the high-frequency wire harness material; is the gradient of T(t); is the gradient operator; is the current of the jth high-frequency wire bundle at time t; A j is the cross-sectional area of ​​the jth high-frequency beam; X2(t) represents the second component of X(t); h is the convection heat transfer coefficient; T amb is the ambient temperature.

[0017] Furthermore, in step 2, the high-frequency harness stress is calculated based on the elastic modulus tensor and mechanical strain tensor of the high-frequency harness material using the following formula:

[0018]

[0019] Where E is the elastic modulus tensor; α is the thermal expansion coefficient; :: represents the double tensor product operation; μ0 is the vacuum permeability; μ r is the relative magnetic permeability; X1(t) represents the first component of X(t); ∈ mech is the mechanical strain tensor; σ(t) is the high-frequency beam stress at time t.

[0020] Furthermore, in step 2, the state space is expressed using the following formula:

[0021]

[0022] Among them, S RL (t) is the state space at time t; T max is the maximum allowable temperature of the high-frequency wiring harness; σ yield is the yield strength of the high-frequency harness material; the action space is expressed using the following formula:

[0023]

[0024] Where A(t) is the action space at time t; ΔI(t) is the current adjustment at time t; ΔR contact (t) is the contact resistance adjustment at time t; ΔP cooling (t) is the cooling power adjustment at time t; represents the second-order L2 norm operation; f is the frequency of the high-frequency line beam; f0 is the preset reference frequency.

[0025] Furthermore, in step 3, the reward function R(t) of the reinforcement learning model is defined as:

[0026]

[0027] Where ||T(t)|| ∞ represents the infinite norm of the temperature distribution vector at time t; η(S RL (t), A(t)) means that in the state space S RL (t), the efficiency when the action space is A(t) is obtained through historical data.

[0028] Furthermore, in step 3, the Q value of the reinforcement learning model is continuously updated by the following formula until convergence:

[0029]

[0030] Among them, Q t+1 (S RL (t+1), A(t+1)) is the updated time t+1, and the state space is S RL (t+1), the action space is the Q value at A(t+1); Q t (S RL (t), A(t)) is time t, the state space is S RL (t), the Q value when the action space is A(t); δ is the learning rate; γ is the discount factor; Q t (S RL When A(t)', A(t)') is time t, the state space is S RL (t)', the Q value when the action space is A(t)'; is the p-order derivative of the state vector X(t).

[0031] By adopting the above technical scheme, the present invention has the following intended effects: First, the present invention realizes the adaptive optimization of high-frequency wiring harnesses in a dynamic environment by introducing a reinforcement learning model. Different from the traditional static parameter optimization method, reinforcement learning can adjust the strategy according to the real-time state change of the system, and through continuous learning and improvement, the system can maintain a stable working state in the electrical, thermal and mechanical load fluctuations. Specifically, the reinforcement learning model of the present invention comprehensively covers the key operating data in the system, including temperature distribution, power loss and stress conditions, through the design of state space and action space. The model continuously updates the Q value based on this information, so that the system can automatically adapt to the dynamic changes of the environment and load during operation, thereby greatly reducing temperature rise, reducing power consumption and effectively controlling stress accumulation. Secondly, the present invention shows excellent effects in temperature control. The skin effect of high-frequency current on the surface of the conductor often leads to power loss and local temperature rise. The present invention monitors the temperature distribution in real time and combines the feedback of the predicted temperature, so that the system can adjust the cooling power and current path in time when the temperature approaches the upper limit to achieve effective control of temperature rise. Different from the traditional passive heat dissipation method, the present invention automatically adjusts the cooling power output according to the current temperature of the system through the cooling power adjustment item in the action space, ensuring that the heat dissipation system can flexibly respond to temperature fluctuations in high-frequency environments. In addition, by predicting temperature changes, the present invention can also identify temperature hot spots in advance and further optimize the current path to avoid local overheating, which significantly improves the temperature control efficiency of the high-frequency wiring harness system. The present invention also has significant advantages in stress control. In a high-frequency current environment, the fluctuation of electromagnetic force often causes mechanical stress inside the material, especially the accumulation of fatigue stress may cause material fracture or connection point failure. The present invention realizes dynamic management of stress by introducing stress monitoring in the state space and combining the feedback control function of the reinforcement learning model. Specifically, the system will automatically reduce the current or optimize the current path when the stress is close to the yield strength of the material to reduce the local stress load. This effectively extends the service life of the high-frequency wiring harness and the connection point, while ensuring the overall stability of the system. Compared with the static material enhancement method used in the prior art to improve the tensile strength, the present invention can make dynamic adjustments based on the real-time stress situation, which is more flexible and adaptable. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0033] Figure 1 A schematic diagram of a method flow of a method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The method of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments of the present invention.

[0035] Example 1: Reference Figure 1 , a method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning, the method comprising:

[0036] Step 1: Collect the operating data of each battery module of the new energy equipment, combine the high-frequency wire harness characteristic data of each high-frequency wire harness in the new energy equipment, the connection point power loss data of each connection point, the temperature distribution vector and the current density vector, and establish a state vector;

[0037] In the high-frequency wiring harness system of new energy equipment, the operating data of the battery module, such as voltage, current, SOC (state of charge), etc., directly affect the overall power output of the system. Due to the voltage and current fluctuations between battery modules, especially under dynamic high-frequency conditions, these fluctuations can cause local temperature rise, increase system power consumption, and even cause overheating or short circuit in extreme cases. Therefore, the operating data of the battery module is included in the state vector in order to track these electrical parameters in real time during the connection optimization process to ensure that the system can operate stably with minimal power consumption. In particular, in the reinforcement learning algorithm, these parameters of the battery module provide the model with an important basis for the current energy state of the system, enabling it to weigh the relationship between current conduction and power consumption in the control strategy. Secondly, the characteristic data of the high-frequency wiring harness is also an important part of the state vector. In the wiring harness system, characteristic data such as resistance and inductance will change with frequency, resulting in significant changes in the electrical conductivity and temperature rise characteristics of the wiring harness under high frequency conditions. High-frequency signals can cause skin effects, causing current to tend to the surface of the wiring harness, thereby increasing the effective resistance. This will not only aggravate power loss, but also bring additional thermal effects, affecting the temperature distribution and mechanical stress of the wiring harness and connection points. By incorporating these characteristic data of the high-frequency wiring harness into the state vector, the method of the present invention can more accurately capture the impact of high-frequency signals on the performance of the wiring harness, and actively adjust the connection method during the optimization process to reduce the power loss caused by the high-frequency current. In addition, the power loss data of the connection point is also an important part of the state vector. Due to the power loss generated by the contact resistance and high-frequency current, the connection point is prone to become an area where heat is concentrated in the system, thereby affecting the overall temperature distribution. Especially under high-frequency conditions, the loss of the connection point is not just a simple ohmic loss, it may also be affected by the scattering effect of electromagnetic waves, resulting in a complex heat accumulation phenomenon. The power loss data of the connection point helps the method of the present invention to reflect the concentrated area of ​​the thermal effect of the connection point in the state vector, providing a key reference for heat loss for the reinforcement learning model. Combined with this power loss information, the model can regulate the connection point during the optimization process to reduce heat accumulation, thereby extending the overall service life of the system.

[0038] Next, the temperature distribution vector, as an important component of the state vector, can effectively reflect the thermodynamic properties of each part of the new energy equipment during high-frequency operation. Current conduction under high-frequency conditions can cause a sharp rise in temperature, especially when the thermal diffusion rate of the material is low, heat is easily accumulated in a specific area, causing thermal stress. The present invention incorporates the temperature distribution vector into the state vector to achieve precise control and regulation of temperature during the optimization process. This not only helps to avoid safety hazards caused by excessive temperature, but also provides a real-time thermodynamic feedback for the dynamic connection optimization of the wiring harness, ensuring that the system remains within the appropriate temperature range at all times. At the same time, the temperature distribution vector can help the reinforcement learning model identify which areas are more likely to have temperature anomalies, so as to regulate these parts in advance in action selection to avoid connection failure caused by overheating. Finally, the introduction of the current density vector is to supplement the detailed local information in the electrical conduction process. The current density reflects the distribution of current per unit cross-sectional area and is an important basis for the calculation of temperature rise and electromagnetic stress. Under high-frequency conditions, the phenomenon of uneven current distribution is more obvious. The increase in local current density will not only cause temperature rise, but also generate additional electromagnetic stress inside the material. By incorporating the current density vector into the state vector, the reinforcement learning model can identify the concentration of current in each area, thereby avoiding local overheating and material deformation caused by high-density current during the optimization process. During the dynamic connection optimization process, the distribution of current density enables the model to more accurately control the conduction path of the current, preventing high-frequency current from excessively dissipating energy in local concentration, thereby affecting the overall performance of the system.

[0039] Step 2: Based on the state vector and the temperature distribution vector, calculate the predicted temperature distribution vector; based on the elastic modulus tensor and mechanical strain tensor of the high-frequency wire harness material, calculate the high-frequency wire harness stress; based on the state vector, the high-frequency wire harness stress and the predicted temperature distribution vector, construct the state space and action space of the reinforcement learning model;

[0040] The prediction of temperature distribution is the key to achieve thermal control in dynamic connection optimization of the present invention. The conduction of high-frequency current will cause complex thermal effects, including the temperature rise of the wiring harness material, the local heating of the connection point, and the heat conduction and heat dissipation effect inside the device. Under high-frequency conditions, the current will tend to flow on the surface of the conductor due to the skin effect, resulting in an increase in the effective resistance of the material, thereby generating more Joule heat. In order to control the impact of these thermal effects on the connection state of the system, step 2 uses the temperature prediction model to calculate the temperature distribution in the next time step to ensure that the reinforcement learning model can perceive and adapt to temperature changes in advance. When predicting the temperature distribution, the temperature conduction equation combines the properties of the conductor material such as density, thermal conductivity, and specific heat capacity, while considering the local power loss of the connection point and the wiring harness. Through such refined temperature prediction, the reinforcement learning model can not only identify the temperature hot spot area, but also introduce targeted cooling or load reduction measures in the action space to avoid material aging or connection failure caused by temperature accumulation. In addition, the stress calculation of the high-frequency wiring harness also plays an important role in dynamic connection optimization. Under the action of high-frequency current, the inside of the wiring harness will be affected by both temperature changes and electromagnetic forces, resulting in the generation and accumulation of mechanical stress. This stress usually manifests itself as a combination of thermal expansion stress and electromagnetic stress, where the thermal expansion stress originates from the temperature change of the harness material, while the electromagnetic stress originates from the electromagnetic field generated by the high-frequency current in the conductor. By introducing mechanical parameters such as the elastic modulus tensor and the mechanical strain tensor, the stress distribution of the high-frequency harness under the thermoelectric effect is accurately calculated in step 2. In particular, in the reinforcement learning optimization framework proposed in the present invention, the calculation of the stress distribution provides the necessary stress information for the reinforcement learning model, enabling it to avoid high stress areas during the connection optimization process. This stress analysis method is more sophisticated than the traditional optimization scheme that only considers electrical factors, because it can capture the influence of the electromagnetic field on the mechanical structure of the harness under high-frequency current conditions, thereby effectively preventing material deformation, fatigue and damage caused by electromagnetic stress.

[0041] Based on temperature distribution and stress analysis, step 2 further constructs the state space and action space of the reinforcement learning model. The state space includes real-time temperature distribution prediction, wiring harness stress, power loss and other information to fully reflect the state of the system under different operating conditions. Unlike the fixed state space in the traditional method, the state space of the present invention changes dynamically with the update of the operating data, so that the reinforcement learning model can adapt to different working conditions and load conditions. The action space includes dynamic adjustment of the current path, cooling power, and connection resistance. The reinforcement learning model can select the optimal action based on the current state to achieve the comprehensive effect of reducing power consumption, controlling temperature rise, and reducing stress. It is worth noting that the action space design of the present invention is not a simple parameter adjustment, but a comprehensive consideration of the complex response of the high-frequency wiring harness system under various physical factors. For example, under certain high-power loads, the model may choose to reduce the power consumption in the high-temperature area by adjusting the current path, while increasing the cooling power to stabilize the temperature; in other working conditions, the model may adjust the connection resistance to disperse the stress and extend the life of the wiring harness and the connection point. Reinforcement learning can adaptively adjust the strategy to cope with the high-frequency characteristics and complex multi-physical quantity coupling effects in new energy equipment. Compared with traditional connection optimization methods, the present invention relies on real-time state information to generate a dynamic state space under high-frequency conditions, so that the control strategy can quickly respond to changes in electrical and thermal conditions. The action space design achieves optimal control of system performance by dynamically adjusting different parameters, which is difficult to achieve in traditional methods. In addition, the design of the state space and action space enables the reinforcement learning model of the present invention to be applied not only to new energy equipment, but also to other high-frequency application scenarios, such as high-power electronic equipment and high-frequency communication equipment, thus having a wide range of applicability.

[0042] Step 3: Define the reward function of the reinforcement learning model; continuously update the Q value of the reinforcement learning model through iterative optimization until convergence, output the corresponding action space at this time, and complete the high-frequency wire harness dynamic connection optimization based on the action space.

[0043] In step 3, the definition of the reward function is particularly important, which determines the preference of the reinforcement learning model for different states. The reward function integrates multiple factors such as electrical efficiency, temperature control, and stress management to ensure that the model is always optimized in the direction of reducing power consumption, lowering temperature, and reducing stress during the control process. Specifically, the reward function uses the ratio of the electrical output of the battery module to the power consumption of the wiring harness system as an evaluation indicator. The higher the electrical output and the lower the power consumption, the greater the reward value, which helps to encourage the model to minimize unnecessary power loss when selecting actions. At the same time, a temperature control term is introduced in the reward function, which limits the reward value in the high temperature state through the exponential decay coefficient of the temperature state, so that the model automatically adjusts the action in the high temperature area, thereby avoiding system instability caused by heat accumulation. In addition, the reward function also contains a stress management term to constrain the current stress state to ensure that the wiring harness does not exceed the yield strength of the material during operation. Combined with these different weight terms, the reward function provides a comprehensive performance feedback for each action, allowing the model to have a more comprehensive understanding and response to the current state at each moment. In the reinforcement learning process, the iterative update of the Q value is a key step in achieving model optimization. The Q value represents the expected value of the long-term cumulative reward that can be obtained after performing a certain action in a certain state. In order to find the best dynamic connection strategy, the present invention guides the model to gradually move toward the optimal strategy through Q-value updates in step 3. Specifically, the reinforcement learning model selects an action based on the current state, and continuously updates the current Q-value by calculating the reward value of the current action and adding it to the possible optimal reward value in the future, thereby continuously improving the quality of the control strategy in multiple rounds of iterations. The model repeatedly performs this tentative exploration and update in different states to form the best action selection path. This iterative Q-value optimization makes the model not only rely on the immediate reward value, but also focuses on the long-term cumulative rewards, and obtains the global optimal solution by selecting actions with higher cumulative returns in the future.

[0044] The reward function of the present invention is combined with the Q value iteration to form a complete optimization feedback closed loop, so that the reinforcement learning model has the ability to adapt to dynamic states and predict future states. In high-frequency wiring harness systems, the rapid changes in electrical, thermal and stress environments may make traditional optimization methods difficult to adapt, but through the multi-dimensional design of the reward function, the model of the present invention can obtain comprehensive performance feedback of the system in real time. The reward value generated by each action is calculated under the combined influence of multiple factors such as power consumption, temperature and stress, so that the model can accurately balance multiple goals when selecting actions. More importantly, since the reward function design in the present invention has a parametric characteristic, the model can automatically adjust the weights of different factors under different working conditions. For example, under high power loads, the model will pay more attention to temperature control and stress management, while under low power conditions, it will tend to improve electrical efficiency. The present invention realizes self-improvement of dynamic connection strategies through iterative optimization of Q values. Compared with fixed preset strategies, the reinforcement learning model of the present invention can explore the optimal connection method by itself through multiple rounds of Q value updates. This adaptive capability is particularly suitable for the high-frequency operating environment of new energy equipment, because under high-frequency conditions, the system's load conditions and external environment will continue to change, and the preset static strategy often cannot respond in real time. However, the self-learning characteristics brought about by the iterative optimization of the Q value enable the model of the present invention to flexibly adapt to these changes, thereby being able to find the optimal connection strategy under different working conditions. This self-improvement capability significantly improves the applicability and robustness of the present invention.

[0045] Embodiment 2: The operating data of each battery module includes: battery module voltage, battery module current and battery module state of charge; the high-frequency wiring harness characteristic data includes: high-frequency wiring harness length, high-frequency wiring harness resistance and high-frequency wiring harness maximum allowable current; the connection point power loss data includes: connection point power loss and connection point rated power; the environmental data includes: temperature distribution vector and current density vector.

[0046] First, regarding the operating data of the battery module, the present invention introduces the battery module voltage, current and state of charge (SOC), which directly reflects the real-time status and health of the battery. The voltage and current data of the battery module are the basic parameters for calculating the battery power output, and are also important indicators of the battery status. The optimization of the high-frequency harness connection must be based on the current state of the battery module to ensure that the system can stably supply power at any time and avoid increased power loss or abnormal temperature due to battery voltage or current fluctuations. The state of charge (SOC) provides the model with the remaining power information of the battery, so that the reinforcement learning model can understand the energy storage status of the current system and properly adjust the current flow direction in the optimization strategy to achieve reasonable energy distribution and protect the battery life. This multi-dimensional operating data combining voltage, current and SOC can not only enhance the model's grasp of the battery status, but also help optimize the current path distribution and power consumption control. Secondly, the high-frequency harness characteristic data described in Example 2 is also a key component of the present invention, including harness length, harness resistance and maximum allowable current. Since high-frequency wiring harnesses exhibit complex electrical characteristics under high-frequency conditions, especially the length and resistance of the wiring harnesses have an important impact on the conduction efficiency of high-frequency signals. Long wiring harnesses and higher resistance will increase power consumption and heat accumulation during transmission, resulting in a decrease in the system's electrical energy efficiency. Especially in high-frequency states, the current tends to be conducted on the surface of the wiring harness (skin effect), and the effective resistance increases, which not only increases power consumption, but may also cause local temperature rise, thereby increasing the stress load. Therefore, the maximum allowable current of the wiring harness is also an important limiting condition for the model when selecting an action. If this limit is exceeded, the wiring harness may overheat or be damaged. The present invention incorporates the length, resistance and maximum allowable current of the high-frequency wiring harness into the characteristic data to ensure that the model always pays attention to these key parameters during reinforcement learning optimization, and dynamically adjusts the connection method to achieve efficient transmission of electrical energy.

[0047] In addition, the power loss data of the connection point further improves the state vector of the present invention, which includes the power loss and rated power of the connection point. The connection point is an important node for current flow, and its power loss is determined by the contact resistance and the conduction characteristics of the high-frequency current. Especially under high-frequency conditions, the connection point often becomes a concentrated area of ​​power loss and heat accumulation. The ratio of the power loss of the connection point to its rated power provides an evaluation basis for the stability of the system. When the loss is close to the rated power, the connection point may face the risk of overheating or even connection failure. Therefore, the method of the present invention collects the power loss data of the connection point in real time to ensure that the model can identify which connection points have high power consumption and temperature accumulation risks, and can adjust the connection state in a targeted manner in the subsequent optimization process to reduce local power loss and extend the service life of the connection point and the wiring harness. Finally, the introduction of environmental data provides real-time feedback of the external environment of the system for the reinforcement learning model of the present invention, including temperature distribution vectors and current density vectors. The temperature distribution vector reflects the thermal conditions of various parts in the system, which is an important data for dealing with temperature changes in high-frequency environments. High-frequency currents often cause local temperature rises to rise rapidly. If not controlled, thermal stress accumulation may occur, and even material fatigue and failure may occur. The current density vector is used to describe the distribution of current per unit cross-sectional area. The higher the current density, the more obvious the temperature rise in the area. The present invention incorporates the temperature distribution vector and the current density vector into the environmental data, so that the reinforcement learning model can track these temperature and current characteristics at any time, adjust the connection mode of the high-frequency wiring harness in time, avoid local overheating caused by high-density current, and thus ensure the stability and safety of the entire system.

[0048] Embodiment 3: The state vector is expressed using the following formula:

[0049]

[0050] Where X(t) represents the state vector at time t; i, j, and k are all integer subscript indices; SOC i (t) represents the charge state of the ith battery module at time t; V i (t) represents the voltage of the i-th battery module at time t; I i (t) represents the current of the i-th battery module at time t; N b is the number of battery modules; N w is the number of high-frequency wiring harnesses; R j is the high-frequency wiring harness resistance of the jth high-frequency wiring harness; L j is the high-frequency wire bundle length of the j-th high-frequency wire bundle; N is the maximum allowable current of the high-frequency wiring harness of the jth high-frequency wiring harness; c is the number of connection points; is the connection point power loss of the kth connection point; is the rated power of the kth connection point; T(τ) is the temperature distribution vector; J(τ) is the current density vector; τ is the integration time variable.

[0051] Specifically, the first term of the state vector X(t) Indicates the overall power state of the battery module at time t. This item is calculated by the state of charge (SOC) and voltage V i (t) and current I i (t) to measure the power output of the system. SOC represents the remaining capacity of each battery, and the voltage V i (t) and current I i (t) directly affects the power transmission capacity of the system. Combining these factors, the current power status of the battery system can be comprehensively evaluated, which is particularly critical for the dynamic optimization of the high-frequency wiring harness of the present invention, because the stability of the battery output will directly affect the temperature rise and power consumption of the wiring harness. Under high-frequency conditions, the output fluctuation of the battery may cause a short-term peak in power, which increases the burden on the connection and wiring harness. Therefore, this comprehensive power status can help the reinforcement learning model predict and prevent potential overheating or overload risks. Secondly, the second term of the state vector X(t) Reflects the power loss characteristics of the high-frequency wiring harness. Here R j is the resistance of the jth high-frequency wire harness, L j For its length, is its maximum allowable current. This item calculates the power consumption of the high-frequency harness at its maximum current. Since the resistance and length of the harness have a direct impact on the power loss, and the skin effect in a high-frequency environment will increase the effective resistance of the harness, further increasing the power loss. This part of the power loss is directly related to the temperature distribution and overall energy efficiency of the system. By accumulating the power loss of each harness, the reinforcement learning model can take into account the power transmission capacity of the harness when selecting actions, ensuring that the harness will not overheat due to excessive power consumption under different load conditions. For high-frequency equipment, optimizing power loss can significantly reduce the temperature rise and energy loss of the system, thereby improving the efficiency and safety of power transmission. The third item of the state vector This indicates the power loss efficiency of the connection point. With its rated power The ratio of is calculated as a negative exponential, reflecting the efficiency status of the connection point under the current working conditions. If the loss of a certain connection point is close to its rated power, the exponential value of this item will be significantly reduced, prompting the model that there may be a risk of heat accumulation at this connection point. As a key node for current transmission, the loss of the connection point will directly affect the overall stability of the system. Especially at high-frequency operation, due to the relationship between contact resistance and current frequency, the connection point may produce a large local temperature rise. By incorporating this item into the state vector, the reinforcement learning model can identify which connection points may overheat due to excessive power loss, and then regulate the current flow path of these connection points in the action space to reduce local heat load and improve the safety and reliability of the system. The last item of the state vector It is used to describe the cumulative coupling effect of temperature and current density, where T(τ) represents the temperature distribution vector, J(τ) represents the current density vector, and τ is the integral time variable. This integral term accumulates the coupling effect of temperature and current density by means of time integration to fully reflect the thermodynamic characteristics of the system over a period of time. In a high-frequency environment, the temperature distribution inside the system will be dynamically affected due to changes in current density, and this cumulative effect of temperature may cause thermal fatigue and stress accumulation of the material. By introducing this term, the model can dynamically evaluate the cumulative effects of temperature and current density to provide support for subsequent action selection. For example, when the cumulative temperature approaches the heat resistance limit of the material, the reinforcement learning model can give priority to cooling measures or dispersed current paths to prevent material damage caused by overheating. The consideration of this coupling effect makes the present invention more practical and safe under high-frequency conditions.

[0052] Embodiment 4: In step 2, the predicted temperature distribution vector is calculated based on the state vector and the temperature distribution vector using the following formula:

[0053]

[0054] Wherein, T(t+Δt) is the predicted temperature distribution vector at time t+Δt obtained by calculation; T(t) is the temperature distribution vector at time t; Δt is the time step; C p is the specific heat capacity of the high-frequency wire harness material; ρ is the density of the high-frequency wire harness material; k is the thermal conductivity of the high-frequency wire harness material; is the gradient of T(t); is the gradient operator; is the current of the jth high-frequency wire bundle at time t; A j is the cross-sectional area of ​​the jth high-frequency beam; X2(t) represents the second component of X(t); h is the convection heat transfer coefficient; T amb is the ambient temperature.

[0055] Specifically, the predicted temperature distribution vector T(t+Δt) is an estimate of the temperature change trend within the time step Δt. In order to accurately describe the temperature rise of the high-frequency wiring harness in a high current and high temperature environment, this formula integrates multiple factors such as heat conduction, power loss, and convection heat dissipation to form a global temperature distribution prediction model. Specifically, the first term T(t) of the formula represents the temperature distribution at the current time t, that is, the initial temperature state of the high-frequency wiring harness system at time t. By adding this term, the model can reflect the temperature of the previous time step in the predicted temperature, ensuring that the temperature prediction is based on the current true temperature distribution, so that the model can capture the dynamic changes in temperature. The time step Δt in the formula represents the time interval used in the prediction process. A shorter time step can make the temperature prediction more accurate, while a longer time step is more suitable for long-term temperature trend analysis. In order to make the formula have a more comprehensive description of the thermal properties of the high-frequency wiring harness material, the present invention introduces the specific heat capacity C p And material parameters such as density ρ. Specific heat capacity C p It represents the heat demand per unit mass of the material during temperature change, which determines the rate at which the temperature rises and falls inside the material; the density ρ reflects the mass density of the material, which is closely related to the rate of temperature change and the conduction characteristics of heat flow. The product of the two is A thermal inertia factor is formed, which indicates the temperature response sensitivity of the system within the time step Δt. The greater the density and specific heat capacity of the high-frequency wire bundle, the slower the system reacts in the heat flow conduction, thus avoiding sudden temperature changes.

[0056] The thermal conductivity k in the formula represents the thermal conductivity of the high-frequency wiring harness material, which determines the speed at which heat is conducted inside the material. The higher the thermal conductivity, the faster the heat transfer inside the material, and the more evenly the temperature diffuses in the material. That is, the heat diffusion term. The model can describe the distribution and conduction effect of temperature inside the high-frequency harness. is the current temperature gradient, indicating the rate of change of temperature within the material. The introduction of this gradient term allows the temperature to be transferred from high temperature to low temperature within the system, achieving spatial temperature balance. The thermal conductivity and temperature gradient of the high-frequency wiring harness are important factors to ensure the stability of the system. By controlling the temperature gradient, the risk of local overheating of the material and accumulation of thermal stress can be avoided. In addition, the formula also takes into account the contribution of the power loss of the high-frequency wiring harness to the temperature rise, that is, the term in is the square of the current in the jth wire bundle, R j is the wiring harness resistance, A j is the cross-sectional area of ​​the wire bundle. This term is obtained through the Joule effect (I 2R) calculates the power loss of each harness and is inversely proportional to the square of the cross-sectional area, indicating the temperature rise per unit area. Since the skin effect of high-frequency current will produce a large power loss on the surface of the conductor, this term estimates the heat source in the system by accumulating the power consumption of each harness, thereby reflecting the local concentration effect of temperature. X2(t) is the second component of the state vector X(t), which plays a weighted role in power loss prediction and is used to further adjust the temperature rise model of the system. In order to more realistically simulate the heat exchange between the high-frequency harness system and the external environment, the convective heat transfer term h(T amb -T(t)), where h is the convective heat transfer coefficient, T amb is the ambient temperature. The heat transfer coefficient h represents the heat transfer efficiency between the material surface and the surrounding air. The ambient temperature T amb It provides a temperature reference that allows heat to diffuse from the high-temperature wiring harness area to the low-temperature external environment. The existence of the heat transfer term helps the system dissipate heat to the outside when the temperature is too high, thereby reducing the overall temperature and ensuring that the high-frequency wiring harness will not cause system failure due to poor heat dissipation during high-power transmission. This term plays an important role in the state prediction and action decision-making of the reinforcement learning model because it reflects the heat dissipation capacity of the system at different ambient temperatures. The model can choose to increase heat dissipation measures under high-temperature conditions to avoid safety problems caused by heat accumulation.

[0057] Embodiment 5: In step 2, the high-frequency harness stress is calculated based on the elastic modulus tensor and mechanical strain tensor of the high-frequency harness material using the following formula:

[0058]

[0059] Where E is the elastic modulus tensor; α is the thermal expansion coefficient; :: represents the double tensor product operation; μ0 is the vacuum permeability; μ r is the relative magnetic permeability; X1(t) represents the first component of X(t); ∈ mech is the mechanical strain tensor; σ(t) is the high-frequency beam stress at time t.

[0060] Specifically, the stress σ(t) in the formula represents the stress borne by the high-frequency wire harness at time t. The stress of the high-frequency wire harness is not only affected by the mechanical load, but also by the combined effect of the thermal stress caused by temperature change and the electromagnetic stress generated by the high-frequency current. In order to accurately describe this multi-dimensional stress, the present invention uses the elastic modulus tensor E, the mechanical strain tensor ∈ mech , thermal expansion coefficient α, vacuum magnetic permeability μ0, relative magnetic permeability μ rAnd the important components in the state vector are used to construct this stress model. Such a stress calculation method enables the reinforcement learning model of the present invention to timely perceive the stress changes of the system when dynamically optimizing the high-frequency wiring harness connection, and take effective measures to avoid excessive deformation or fracture of the material, thereby improving the safety and service life of the equipment. The elastic modulus tensor E is one of the key parameters in stress calculation, which describes the material's resistance to deformation under external force. The higher the elastic modulus, the more difficult it is for the material to deform, so the deformation will be smaller under the same stress. In the stress formula, the double tensor product operation:: multiplies the elastic modulus by the composite strain to produce the corresponding stress tensor. This operation shows that the magnitude of stress is not only related to the amplitude of the strain, but also closely related to the elastic properties of the material, so that the stress calculation can fully reflect the material's resistance to deformation in a multi-dimensional mechanical environment.

[0061] Mechanical strain tensor ∈ mech It directly represents the deformation state of the material under mechanical load, which is one of the basic sources of stress. Mechanical strain can be caused by the weight of the device itself or other mechanical external forces. In high-frequency wiring harness connections, mechanical strain may be affected by high-frequency vibration or electromagnetic force. This term is used in the stress formula to describe the static mechanical state of the material, ensuring that the contribution of pure mechanical load to stress can be reflected during calculation. At the same time, combined with the elastic modulus, the mechanical strain tensor brings the amplitude and directionality of mechanical stress into the model, providing a static mechanical reference for the reinforcement learning model to help it better evaluate the impact of external loads on the stability of the wiring harness. The thermal expansion term αΔT(t) in the formula is used to describe the thermal stress caused by temperature changes. Current under high-frequency conditions will cause the temperature of the wiring harness material to rise, especially due to the skin effect, the current is more concentrated on the surface of the conductor, which will cause the temperature in the local area to rise rapidly. When the temperature changes, the material will expand or contract thermally, generating corresponding stress. The thermal expansion coefficient α represents the material's ability to expand with temperature changes. By multiplying it with the temperature difference ΔT(t), this term reflects the expansion stress generated by the material when the temperature changes. This thermal stress term is particularly important in high-frequency applications, because periodic fluctuations in temperature may induce microscopic fatigue within the material, ultimately leading to fracture or performance degradation. Therefore, by incorporating the thermal expansion term into the stress calculation, the present invention can predict the potential impact of temperature changes on the system in real time, thereby helping the reinforcement learning model to take action in advance and reduce the structural risks caused by temperature rise. Electromagnetic stress term The electromagnetic effect of high-frequency current on the material is further considered. High-frequency current will generate an electromagnetic field around the material, which will interact with the current inside the material to generate additional stress. Vacuum magnetic permeability μ0 and relative magnetic permeability μ r Together they determine the material's magnetic response characteristics; their product reflects the material's magnetic permeability in an electromagnetic field. In the equation, X1(t) and X2(t) represent the first and second components of the state vector, i.e., the comprehensive power state of the battery module and the overall power consumption of the wiring harness. By taking the square of the ratio of these two components, we can calculate the contribution of current density to electromagnetic stress. The greater the current density, the stronger the electromagnetic field, and the greater the electromagnetic stress inside the material. This term is used to reflect the influence of the interaction between the electromagnetic field and high-frequency current on the material stress. By considering electromagnetic stress together with other stress sources, the method of the present invention can more accurately describe the mechanical state of the system under high-frequency conditions.

[0062] Embodiment 6: In step 2, the state space is represented by the following formula:

[0063]

[0064] Among them, S RL (t) is the state space at time t; T max is the maximum allowable temperature of the high-frequency wiring harness; σ yield is the yield strength of the high-frequency harness material; the action space is expressed using the following formula:

[0065]

[0066] Where A(t) is the action space at time t; ΔI(t) is the current adjustment at time t;

[0067] ΔR contact (t) is the contact resistance adjustment at time t; ΔP cooling (t) is the cooling power adjustment at time t; represents the second-order L2 norm operation; f is the frequency of the high-frequency line beam; f0 is the preset reference frequency.

[0068] Specifically, the state space S RL (t) is a description of the current state of the system at time t, which reflects the comprehensive operating status of the high-frequency wiring harness system through temperature change, stress state and state gradient. The first term in the state space represents the relative change of system temperature within the time step Δt, T max is the maximum allowable temperature of the system. This term provides information about temperature stability by comparing the rate of temperature change to the upper limit of the allowable temperature. If the temperature change rate is large and close to the maximum allowable temperature, the system may be at risk of overheating. The reinforcement learning model of the present invention can use this information to adjust the cooling or current in time to prevent the system temperature from rising too quickly. The second term in the state space is It represents the stress state of the system at the current moment. Here, σ(t) is the actual stress value of the high-frequency wire harness at time t, and σ yieldis the yield strength of the harness material. The ratio of this term is used to determine whether the system is close to the stress limit of the material. If the ratio is close to 1, it means that the stress is close to the yield strength and there is a risk of material deformation or damage. By incorporating the stress state into the state space, the model of the present invention can be adjusted in high stress areas to disperse the current flow and reduce the stress burden, thereby extending the service life of the harness and connection points. The third term in the state space is Represents the gradient dot product of the state vector and the temperature distribution, and is used to describe the spatial correlation between temperature and electrical state. This term reflects the spatial interaction between the electrical state and temperature, especially in high-frequency environments, where the current distribution affects the temperature distribution, and temperature changes also affect the electrical state. By incorporating this term into the state space, the model can identify temperature changes in current-concentrated areas and further optimize current paths and cooling solutions to balance electrical and thermal states, thereby reducing the risk of local overheating.

[0069] In the action space A(t), the present invention provides adjustable parameters of the reinforcement learning model, including the current adjustment amount ΔI(t), the contact resistance adjustment amount ΔR contact (t) and cooling power adjustment ΔP cooling (t). Each control parameter in the action space directly affects the thermal and electrical state of the system, allowing the model to dynamically adjust under different environmental and load conditions to maintain the efficiency and safety of the system. The current adjustment ΔI(t) allows the model to adjust the current according to temperature and stress conditions to reduce power density and temperature rise. The contact resistance adjustment ΔR contact (t) is used to optimize the resistance of the connection point to improve the current flow path and reduce power consumption. Especially under high frequency conditions, the adjustment of contact resistance can significantly affect the temperature and power distribution of the system. Cooling power adjustment ΔP cooling (t) controls the power output of the cooling system to quickly reduce the system temperature to a safe level, especially when the load increases or the ambient temperature rises. This term can help the system dissipate heat quickly. The exponential decay term introduced in the action space is The size of the state space is normalized by the L2 norm and combined with the frequency f

[0070] The difference from the reference frequency f0 is used to adjust the execution intensity of the action. The purpose of this term is to adapt the action space to the state space. When the L2 norm of the state space is large, that is, when the system is close to the danger zone, the exponential term will tend to a smaller value, thereby suppressing large adjustments and maintaining the stability of the system. When the frequency f is close to the reference frequency f0, the vibration characteristics and power consumption characteristics of the system tend to be stable, so the adjustment amplitude of the action can be appropriately increased. In this way, through the adaptive adjustment of the exponential decay term, the model can automatically balance the adjustment intensity according to the frequency conditions and the current state, optimize the connection state and improve the overall efficiency without affecting the stability of the system.

[0071] Example 7: In step 3, the reward function R(t) of the reinforcement learning model is defined as:

[0072]

[0073] Where ||T(t)|| ∞ represents the infinite norm of the temperature distribution vector at time t; η(S RL (t), A(t)) means that in the state space S RL (t), the efficiency when the action space is A(t) is obtained through historical data.

[0074] Specifically, the first term of the reward function Used to evaluate the electrical efficiency of the system, where X1(t) represents the first component of the state vector, that is, the comprehensive power output state of the battery module, and X2(t) is the power consumption state of the wiring harness. Through this ratio, the reward function can measure the relationship between the output power and power consumption of the system, reflecting the energy transfer efficiency of the system in the current state. If the ratio is large, it means that the system's power transfer efficiency is high and the power consumption in the wiring harness is low; on the contrary, if the ratio is small, it means that the system's power has been lost during the transmission process. Therefore, the existence of this term encourages the model to select actions that can improve the electrical efficiency of the system, ensure that power consumption is reduced as much as possible during the optimization process, and achieve better energy distribution. In order to avoid the situation of zero denominator, the minimum value of 0.01 is used to ensure the calculation stability of the function. The second term of the reward function is the temperature control term, which evaluates the temperature state by the ratio of the infinite norm of temperature to the maximum allowable temperature. Here ∥T(t)∥ ∞ Indicates the highest temperature value in the system, T maxis the upper temperature limit of the system. The exponential decay form of this term ensures that the reward value will be significantly reduced when the temperature approaches the upper limit, thereby encouraging the model to choose operations that can control the temperature rise and avoid safety risks caused by overheating of the system. In high-frequency wiring harness systems, temperature increases often cause material aging and stress accumulation, so the introduction of this term is very important. It can help the model identify and avoid high temperature conditions and effectively improve the safety and stability of the system. The third term Used to evaluate the current stress state, where ∥σ(t)∥2 represents the second norm of the current stress distribution, σ yield is the yield strength of the material. The purpose of introducing this term is to prevent the stress from exceeding the tolerance of the material. If the current stress is close to the yield strength, the value of this term will be close to 0, resulting in a decrease in the reward value. Therefore, the reinforcement learning model will tend to choose operations that will not cause stress to exceed the limit to ensure the stability of the system under stress. This design is particularly important in high-frequency environments, because fluctuations in high-frequency currents can cause changes in electromagnetic forces, thereby affecting the stress state. Through this stress control term, the model of the present invention can effectively control stress distribution in a dynamic environment to avoid material fatigue or damage. The last term η(S RL (t), A(t)) is the efficiency term, obtained through historical data, expressed in the state space S RL (t), the operational efficiency generated by executing action A(t) under load (t). Historical data provides the model with empirical feedback under different operating conditions, so that the efficiency term can reflect the pros and cons of a specific state and action combination. The design of this efficiency function can help the reinforcement learning model tend to a better control combination in actual operation. For example, under high power consumption conditions, historical data may show that adjusting the cooling power will be more effective. The efficiency term will guide the model to give priority to cooling measures, while under low load conditions, the model may choose to adjust the contact resistance or current path to improve system efficiency. This efficiency term not only makes the reward function more in line with actual operating conditions, but also enhances the learning ability of the model, enabling it to continuously optimize strategies in long-term operation.

[0075] Example 8: In step 3, the Q value of the reinforcement learning model is continuously updated by the following formula until convergence:

[0076]

[0077]

[0078] Among them, Q t+1 (S RL (t+1), A(t+1)) is the updated time t+1, and the state space is S RL (t+1), the action space is the Q value at A(t+1); Q t (S RL(t), A(t)) is time t, the state space is S RL (t), the Q value when the action space is A(T); δ is the learning rate; γ is the discount factor; Q t (S RL When A(t)', A(t)') is time t, the state space is S RL (t)', the Q value when the action space is A(t)'; is the p-order derivative of the state vector X(t).

[0079] Specifically, during the Q value update process, the first part of the formula is the current Q value Q t (S RL (t), A(t)) is a combination of the reward R(t) at time step t+1 and the future optimal Q value. This design allows the Q value update to take into account both the immediate benefits brought by the current operation and the maximum benefits that may be obtained in the future. The immediate reward R(t) is the feedback of the current state and action. The larger its value, the more beneficial the current operation is to the system, so it will positively affect the update of the Q value. The future optimal Q value max A(t)' Q t (S RL (t)', A(t)') is based on the model's prediction of future states, measuring the highest cumulative benefit obtained in possible future actions. By considering the maximum Q value in the future, the model can not only focus on current benefits in operational decisions, but also balance long-term benefits to ensure the global nature of the optimization process. The discount factor γ is used here to adjust the degree of influence of future benefits on the current Q value. A larger γ value means that the model pays more attention to long-term benefits, while a smaller γ value pays more attention to immediate benefits. The learning rate δ in the update formula determines the speed of Q value update. A larger learning rate means that the model is more sensitive to new information in each update, while a smaller learning rate makes the model more dependent on past experience. In practical applications, it is very important to choose a suitable learning rate. A learning rate that is too large may cause the model to be unstable and difficult to converge; a learning rate that is too small will cause the model to converge slowly and it will be difficult to achieve dynamic optimization in a complex environment. Therefore, the learning rate setting of the present invention needs to be adjusted according to the specific operating conditions of the system to ensure that the Q value can converge stably within an appropriate time. The last part of the Q value update formula introduces an adaptive adjustment term based on the state vector gradient By taking an exponential decay on the L2 norm of the p-order derivative of the state vector X(t), the Q value update can be sensitively adjusted to the rate of change of the state. It reflects the changing trend of the system state. The higher the gradient, the more drastic the system state change. The model needs to be more cautious in operating adjustments to avoid system instability due to overly aggressive adjustments. Through the exponential decay processing of the gradient, the reinforcement learning model of the present invention can flexibly respond to the rapidly changing high-frequency environment, and will not make excessive actions due to the rapid change of the state, thereby ensuring the stability and safety of the system. The design of this gradient-sensitive adjustment item is particularly suitable for high-frequency wiring harness systems, because under high-frequency conditions, the state quantities such as temperature, stress and current density of the system may fluctuate rapidly. The high-order derivatives of the state vector reflect the severity of this fluctuation. When the state changes rapidly, the attenuation effect of the exponential term is more obvious, thereby reducing the update amplitude of the Q value and avoiding excessive adjustment of the model when it changes drastically. In this way, while ensuring the adaptive optimization of the model, the robustness of the system can be maintained, so that the Q value can still gradually converge in a changing environment. Overall, the Q value update formula realizes the reinforcement learning optimization of the dynamic connection of the high-frequency wiring harness through the adaptive adjustment of the immediate reward, the future optimal Q value, the learning rate and the state gradient. In each round of Q-value update, the model further improves its understanding of the system state and optimal action, so that the model can find the optimal connection strategy based on the current state and operation history. The Q-value update method of the present invention not only improves the flexibility and stability of the learning algorithm, but also enhances the robustness of the system in complex high-frequency environments, so that the model has strong security and adaptability while ensuring efficiency. This reinforcement learning Q-value update mechanism provides an intelligent and real-time solution for high-frequency wiring harness connection management of new energy equipment through continuous learning and optimization.

[0080] In terms of power loss, the present invention effectively reduces the overall energy consumption of the system through a multi-dimensional control strategy of reinforcement learning. The power loss in the high-frequency wiring harness system mainly comes from the power loss of the wiring harness resistance and the connection point. The reinforcement learning model of the present invention incorporates power loss data in the state vector, and encourages the model to choose operations that reduce loss through the design of the reward function. In this way, when the power loss is high, the model will give priority to paths that reduce power consumption or increase cooling measures, so that the system can operate under high-efficiency conditions. By combining power consumption information in the Q value update, the present invention can effectively reduce the ineffective power consumption of the system, so that the energy transmission efficiency of the high-frequency wiring harness system is improved. In addition, in a high-load environment, the present invention can dynamically adjust the current path to avoid the situation where high-frequency current is concentrated locally, thereby further reducing the overall power loss of the system. The reinforcement learning model of the present invention also shows excellent effects in terms of adaptability. New energy equipment operates in a complex high-frequency environment, and the frequent changes in load and external conditions make it difficult for traditional fixed control schemes to meet the needs. The present invention uses the Q value update mechanism of reinforcement learning to enable the model to dynamically optimize the operation strategy according to the immediate feedback value, thereby achieving adaptive control. With the help of the adaptive adjustment item of the state gradient, the model can flexibly adjust the update rate of the Q value when dealing with load fluctuations, ensuring that the model converges stably in a rapidly changing environment and does not cause system instability due to excessive adjustment. In this way, the reinforcement learning model of the present invention can intelligently select the best connection strategy in a changeable high-frequency environment, significantly improving the operational stability and efficiency of the system.

[0081] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

Claims

1. A dynamic connection optimization method for high-frequency wiring harnesses of new energy equipment based on reinforcement learning, characterized in that: The method comprises: Step 1: Collect the operating data of each battery module of the new energy equipment, combine the high-frequency wire harness characteristic data of each high-frequency wire harness in the new energy equipment, the connection point power loss data of each connection point, the temperature distribution vector and the current density vector, and establish a state vector; Step 2: Based on the state vector and the temperature distribution vector, calculate the predicted temperature distribution vector; based on the elastic modulus tensor and mechanical strain tensor of the high-frequency wire harness material, calculate the high-frequency wire harness stress; based on the state vector, the high-frequency wire harness stress and the predicted temperature distribution vector, construct the state space and action space of the reinforcement learning model; Step 3: Define the reward function of the reinforcement learning model; continuously update the Q value of the reinforcement learning model through iterative optimization until convergence, output the corresponding action space at this time, and complete the high-frequency wire harness dynamic connection optimization based on the action space.

2. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 1, characterized in that: The operating data of each battery module includes: battery module voltage, battery module current and battery module state of charge; the high-frequency wiring harness characteristic data includes: high-frequency wiring harness length, high-frequency wiring harness resistance and high-frequency wiring harness maximum allowable current; the connection point power loss data includes: connection point power loss and connection point rated power; the environmental data includes: temperature distribution vector and current density vector.

3. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 2, characterized in that: The state vector is expressed using the following formula: Where X(t) represents the state vector at time t; i, j, and k are all integer subscript indices; SOC i (t) represents the charge state of the ith battery module at time t; V i (t) represents the voltage of the i-th battery module at time t; I i (t) represents the current of the t-th battery module at time t; N b is the number of battery modules; N w is the number of high-frequency wiring harnesses; R j is the high-frequency wiring harness resistance of the jth high-frequency wiring harness; L j is the high-frequency wire bundle length of the j-th high-frequency wire bundle; N is the maximum allowable current of the high-frequency wiring harness of the jth high-frequency wiring harness; c is the number of connection points; is the connection point power loss of the kth connection point; is the rated power of the kth connection point; T(τ) is the temperature distribution vector; J(τ) is the current density vector; τ is the integration time variable.

4. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 3, characterized in that: In step 2, the predicted temperature distribution vector is calculated based on the state vector and the temperature distribution vector using the following formula: Wherein, T(t+Δt) is the predicted temperature distribution vector at time t+Δt obtained by calculation; T(t) is the temperature distribution vector at time t; Δt is the time step; C p is the specific heat capacity of the high-frequency wire harness material; ρ is the density of the high-frequency wire harness material; k is the thermal conductivity of the high-frequency wire harness material; is the gradient of T(t); is the gradient operator; is the current of the jth high-frequency wire bundle at time t; A j is the cross-sectional area of ​​the jth high-frequency beam; X2(t) represents the second component of X(t); h is the convection heat transfer coefficient; T amb is the ambient temperature.

5. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 3, characterized in that: In step 2, the high-frequency harness stress is calculated based on the elastic modulus tensor and mechanical strain tensor of the high-frequency harness material using the following formula: Where E is the elastic modulus tensor; α is the thermal expansion coefficient; :: represents the double tensor product operation; μ0 is the vacuum permeability; μ r is the relative magnetic permeability; X1(t) represents the first component of X(t); ∈ mech is the mechanical strain tensor; a(t) is the high-frequency wire stress at time t.

6. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 5, characterized in that: In step 2, the state space is expressed using the following formula: Among them, S RL (t) is the state space at time t; T max is the maximum allowable temperature of the high-frequency wiring harness; σ yield is the yield strength of the high-frequency harness material; the action space is expressed using the following formula: Where A(t) is the action space at time t; ΔI(t) is the current adjustment at time t; ΔR contact (t) is the contact resistance adjustment at time t; ΔP cooling (t) is the cooling power adjustment at time t; represents the second-order L2 norm operation; f is the frequency of the high-frequency line beam; f0 is the preset reference frequency.

7. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 6, characterized in that: In step 3, the reward function R(t) of the reinforcement learning model is defined as: Where ||T(t)|| ∞ represents the infinite norm of the temperature distribution vector at time t; η(S RL (t), A(t)) means that in the state space S RL (t), the efficiency when the action space is A(t) is obtained through historical data.

8. The method for dynamic connection optimization of high-frequency wiring harnesses for new energy equipment based on reinforcement learning as claimed in claim 6, characterized in that: In step 3, the Q value of the reinforcement learning model is continuously updated by the following formula until convergence: Among them, Q t+1 (S RL (t+1), A(t+1)) is the updated time t+1, and the state space is S RL (t+1), the action space is the Q value at A(t+1); Q t (S RL (t), A(t)) is the time t, and the state space is S RL (t), the Q value when the action space is A(t); δ is the learning rate; γ is the discount factor; Q t (S RL (t)′, A(t)′) is the time t, and the state space is S RL (t)′, the Q value when the action space is A(t)′; is the p-order derivative of the state vector X(t).

Images