Intelligent new energy locomotive power distribution and energy recovery control system and method

Abstract

The invention relates to the technical field of electric vehicle control, and provides an intelligent new energy locomotive power distribution and energy recovery control system and method. The system comprises a multi-source information sensing layer, an intelligent decision control layer and an execution feedback layer. The multi-source information sensing layer collects the state of a vehicle and external environment data. The intelligent decision control layer generates a power distribution and energy recovery instruction through data fusion and preprocessing, fuzzy logic assistance, reinforcement learning decision and cooperative work of a multi-objective optimization module; and the execution feedback layer executes the instruction, monitors feedback in real time, and optimizes related actions. According to the method, advanced technologies such as reinforcement learning and fuzzy logic are fused, complex and changeable driving scenes are accurately dealt with, self-adaptive optimization of power distribution and energy recovery strategies is achieved, the energy recovery efficiency is improved, power distribution is accurate and efficient, dynamic adjustment can be achieved according to driving intentions and working conditions, the driving safety and comfort are guaranteed, and meanwhile the driving efficiency is improved. And efficient utilization of energy is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of tram control, and specifically to an intelligent new energy locomotive power distribution and energy recovery control system and method. Background Art

[0002] Under the background of the global advocacy of green environmental protection and sustainable development, new energy locomotives (including pure electric and hybrid locomotives) are increasingly widely used in the transportation field due to their advantages such as low emissions and high efficiency. As the core component of a new energy locomotive, the electric control system, like the "brain" of the locomotive, plays a decisive role in the performance of the whole vehicle. It not only directly affects the power output and energy utilization efficiency of the locomotive, but also relates to multiple aspects such as driving safety, comfort, and the service life of the vehicle.

[0003] In the prior art, the electric control system has the following problems in energy recovery, power distribution, and driving control:

[0004] 1. Low energy recovery efficiency

[0005] The energy recovery strategy in the existing electric control system is relatively single and lacks flexibility, and cannot make precise adjustments according to the real-time changes of these complex road conditions. For example, in the working condition of frequent starts and stops, the recovery intensity cannot be reasonably controlled, resulting in a large amount of wasted braking energy; on a section with a large slope, the gravitational potential energy cannot be effectively utilized for energy recovery, making it difficult to achieve the optimal state of energy recovery efficiency.

[0006] Insufficient consideration of battery state: Factors such as the state of charge (SOC), state of health (SOH), and temperature of the battery have an important impact on energy recovery. However, the prior art often fails to fully consider the combined effect of these factors. When the battery SOC is relatively high, energy recovery is still carried out at a relatively high intensity, which may cause overcharging of the battery and damage the battery life; when the battery temperature is abnormal, the recovery strategy cannot be adjusted in time, further affecting the energy recovery effect and the safety of the battery.

[0007] 2. Single power distribution strategy

[0008] In different driving scenarios of a new energy locomotive, the power requirements vary greatly. For example, a large power output is required during acceleration and climbing; while a more energy-saving power distribution method is needed during constant-speed cruising and coasting. However, the existing power distribution strategies are usually based on fixed logic and preset parameters, and cannot sense and respond to the driver's operation intention and the changes of the actual driving conditions of the vehicle in real time. This results in insufficient power output when high power is required, affecting the driving experience and safety; while when low power is required, it may cause energy waste and reduce the energy utilization efficiency.

[0009] For hybrid locomotives, there are multiple power sources (such as lithium batteries, supercapacitors, etc.). However, the existing power distribution strategies fail to fully utilize the synergistic advantages of multiple power sources. They often simply distribute power according to a fixed ratio without dynamically optimizing based on the real-time status of each power source (such as battery charge, internal resistance, etc.) and the driving conditions of the vehicle, making it difficult for the power system to always operate efficiently.

[0010] 3. Lack of intelligent control algorithms

[0011] Most of the existing control algorithms for the electronic control system only focus on one aspect, such as simply pursuing energy recovery efficiency or power output performance, and it is difficult to take into account multiple important objectives such as driving performance and energy efficiency at the same time. For example, some algorithms, in order to improve energy recovery efficiency, overemphasize the role of electric braking, but ignore the smoothness and safety during braking, resulting in a jerky feeling when the vehicle brakes, affecting driving comfort and passenger experience; or when pursuing power performance, they ignore the reasonable utilization of energy, causing unnecessary energy consumption.

[0012] The actual driving environment is complex and changeable, and factors such as road conditions and weather are constantly changing. However, the existing control algorithms lack the ability to effectively sense and adaptively adjust to these environmental factors. When encountering bad weather (such as heavy rain, ice and snow), they cannot timely adjust the power distribution and braking strategies to adapt to the change of track adhesion, increasing the driving risk; in the case of shunting in the station, they also cannot intelligently optimize energy recovery and power output, reducing the overall performance of the vehicle in complex environments.

[0013] In summary, the existing electronic control systems of new energy locomotives have many defects in energy recovery, power distribution and driving control, and it is difficult to meet the current requirements for high performance, intelligence and high safety of new energy locomotives. With the continuous expansion of the application of new energy locomotives and the continuous progress of technology, there is an urgent need for a more advanced, intelligent and efficient electronic control system and its control method to break through the existing technical bottlenecks, promote the wider and more reliable application of new energy locomotives in the transportation field, and contribute to the construction and development of the global green transportation system. Summary of the Invention

[0014] The technical problem to be solved by the present invention is to provide a control method for power distribution and energy recovery of an intelligent new energy locomotive, which can accurately respond to complex and changeable driving scenarios and realize the adaptive optimization of power distribution and energy recovery strategies by integrating advanced technologies such as reinforcement learning and fuzzy logic.

[0015] The present invention proposes a control system for power distribution and energy recovery of an intelligent new energy locomotive, including:

[0016] The multi-source information perception layer includes vehicle state sensors and environmental perception sensors, and is used to collect vehicle self-state information and external environment data in real time; the vehicle state sensors include a vehicle speed sensor, an acceleration sensor, a battery state sensor, a motor torque sensor, and a wheel speed sensor; the environmental perception sensors include a lidar, a camera, a meteorological sensor, and an electronic map module;

[0017] The intelligent decision-making and control layer includes a data fusion and preprocessing module, a fuzzy logic assistance module, a reinforcement learning decision-making module, and a multi-objective optimization module; the data fusion and preprocessing module performs preprocessing on the collected multi-source data, including but not limited to fusion, noise reduction, and normalization; the fuzzy logic assistance module performs fuzzy logic processing on the multi-source data that is difficult to quantify after preprocessing, and fuzzifies the input variables and defuzzifies the output variables through a fuzzy rule base; the reinforcement learning decision-making module is based on the deep reinforcement learning algorithm, adopts the Actor-Critic network architecture, and generates action instructions for power distribution and energy recovery according to the multi-source data after preprocessing and fuzzy processing; the multi-objective optimization module guides the reinforcement learning algorithm to achieve multi-objective collaborative optimization by designing a reward function that synthesizes multiple objectives;

[0018] The execution and feedback layer includes a power execution module and a feedback monitoring module; the power execution module controls the motor controller and the braking system to perform power distribution and energy recovery actions according to the instructions generated by the intelligent decision-making and control layer; the feedback monitoring module monitors the actual operating state of the vehicle in real time, and transmits the feedback information back to the intelligent decision-making and control layer, and the multi-objective optimization module optimizes the energy recovery actions and power distribution actions.

[0019] Preferably, the state space of the reinforcement learning decision-making module is:

[0020] S = {v, a, SOC, SOH, T m , λ, u, θ, weather, traffic}, the action space A = {k bat , k sc , τ rec}, where v is the vehicle speed, a is the acceleration, SOC is the state of charge of the battery, SOH is the state of health of the battery, T m is the motor torque, λ is the wheel slip ratio, u is the track adhesion coefficient, θ is the track gradient, weather is the weather condition, traffic is the traffic flow condition, k bat is the power ratio of the lithium battery, k sc is the power ratio of the super capacitor, τ rec is the electric braking torque;

[0021] The reinforcement learning decision module dynamically adjusts the power ratio k of the lithium battery based on the collected multi-source data and the fuzzy information provided by the fuzzy logic auxiliary module, combined with the driving conditions, vehicle operating parameters and the reward function of the multi-objective optimization module. bat 、Supercapacitor power ratio k sc and electric braking torque τ rec The action space parameters

[0022] Preferably, the fuzzy logic auxiliary module performs fuzzy logic processing, fuzzifies input variables and defuzzifies output variables through a fuzzy rule base, specifically:

[0023] Real-time collection of vehicle operating parameters, including but not limited to the opening of the acceleration and speed control handle, deceleration and braking intensity, vehicle speed change rate, lithium battery charge state and supercapacitor charge state, is divided into corresponding fuzzy sets respectively, and membership functions are defined for each fuzzy set;

[0024] Construct a fuzzy rule base for fuzzy reasoning, calculate the activation degree of each rule, and generate a fuzzy output set;

[0025] The center of gravity method is used for defuzzification, and the power distribution ratio and energy recovery intensity coefficient are output to the reinforcement learning decision module.

[0026] Preferably, the output power distribution ratio and energy recovery intensity coefficient specifically include:

[0027] (1) Power distribution ratio output control method:

[0028] The locomotive control system calculates the total required power P according to the driving demand and the current operating status of the vehicle based on the vehicle dynamics model and the driver's operating instructions. total ;

[0029] Power distribution ratio obtained based on fuzzy control algorithm According to the formula Calculate the output power P of the lithium battery bat , where η bat is the conversion efficiency of the lithium battery;

[0030] The calculated output power P bat The battery management system monitors the status of the lithium battery and calculates the output power P bat Regulating the output current of lithium batteries The required current size is achieved by controlling the switching devices in the battery charging and discharging circuit, where U bat Output voltage for lithium battery;

[0031] The battery management system monitors the voltage, current and temperature of the lithium battery in real time. If they exceed the safety threshold, it automatically adjusts the output power. At the same time, it feeds back the actual output power and status information of the battery to the locomotive control system.

[0032] (2) Energy recovery intensity coefficient output control method:

[0033] The system determines whether the locomotive enters the energy recovery working condition by monitoring the vehicle speed change rate v` and the opening signal β of the braking speed control handle.

[0034] The energy recovery intensity obtained according to the fuzzy control algorithm Combined with the kinetic energy of the vehicle And the braking demand, according to Calculate the target power P of energy recovery rec , where k adj Is the adjustment coefficient;

[0035] The locomotive control system sends the target power P of energy recovery rec To the motor controller. The motor controller switches the motor from the driving mode to the generating mode under the energy recovery working condition, and generates a reverse torque by changing the control signal of the motor, converting the kinetic energy of the vehicle into electrical energy.

[0036] The electrical energy generated by the motor is transmitted to the energy storage device through the inverter and the charging circuit. The system distributes the energy according to the state of charge of the lithium battery and the super capacitor, and preferentially stores it in the device with a lower state of charge, and monitors the state of the energy storage device.

[0037] During the energy recovery process, the system monitors the vehicle operation state and the energy storage device state in real time, and dynamically adjusts the energy recovery intensity and the energy distribution strategy according to the actual situation.

[0038] Preferably, the reward function of the multi-objective optimization module is:

[0039] R = αR r + βR e + γR s + δR c + εR b

[0040] α, β, γ, δ, ε are the weight coefficients of the energy recovery reward R r , efficiency reward R e , safety reward R s , ride comfort reward R c , battery health reward R b , used to guide the reinforcement learning algorithm to achieve multi-objective collaborative optimization;

[0041] Among them, the energy recovery reward R r= E c -E l , E c represents the recovered energy, and E l is the energy loss during the recovery process;

[0042] Efficiency Reward Among them is the actual output power of the battery, is the optimal power of the battery, is the maximum power of the battery;

[0043] R s is the safety reward, and a corresponding penalty is given when the vehicle is detected to be in a dangerous state;

[0044] Δv is the change in speed, and Δt is the change in time

[0045] Battery Health Reward R b = -λ×(ΔSOC) 2 -u×(ΔT) 2 , where: λ and u are coefficients adjusted according to the battery characteristics; ΔSOC is the change in the battery SOC; ΔT is the change in the battery temperature, which is used to protect the battery life and reduce the damage to the battery caused by the drastic changes in SOC and temperature.

[0046] Preferably, the optimization method of the energy recovery action of the multi-objective optimization module specifically includes a slope road condition dynamic adjustment strategy, a battery SOC adaptive strategy, and a slip rate control strategy, where:

[0047] The slope road condition dynamic adjustment strategy uses a high-precision map and an environmental perception sensor to obtain the road condition information ahead in advance, and dynamically adjusts the energy recovery intensity according to different road conditions. The slope road condition dynamic adjustment strategy satisfies the slope threshold piecewise function. Let θ be the track slope and the recovery intensity adjustment coefficient be K θ , and the recovery intensity adaptive function is f(θ), then it satisfies the formula:

[0048]

[0049] Among them, f(θ) is positively correlated with the slope, f(θ) = α·θ, and α is the slope sensitivity coefficient, with a value range of 0.01 - 0.02;

[0050] For the battery SOC adaptive strategy, let the state of charge of the battery be SOC and the recovery intensity adjustment coefficient be k SOC , and the battery SOC adaptive strategy satisfies the formula;

[0051]

[0052] The slip ratio control strategy sets the wheel slip ratio as λ and the electric braking torque as T rec , and the maximum electric braking torque as T max , satisfying the formula:

[0053]

[0054] where ΔT = k·(T max - T rec ), and k is an adjustment coefficient.

[0055] Preferably, the power distribution optimization method of the multi-objective optimization module includes:

[0056] Using sensors and monitoring devices to collect in real time the power distribution analysis control parameters including but not limited to the position and change rate of the acceleration and speed regulation handle, vehicle speed, driving slope, load pressure, state of charge of the lithium battery, state of charge of the super capacitor, and internal resistance;

[0057] Inputting the collected multi-dimensional driving-related data into a deep learning model to identify the driving intention: If the change rate of the acceleration and speed regulation handle position is greater than 10% / s and the duration is greater than 0.5 s, it is determined as an acceleration intention; if the speed regulation handle changes smoothly and the vehicle speed is stable, it is determined as an energy-saving intention; if the detected slope is greater than 3% and the position of the speed regulation handle is high, it is determined as a climbing intention;

[0058] Dividing the working conditions according to the vehicle driving state, including: high-speed cruise: the vehicle speed is stable at greater than 80 km / h and the operation of the speed regulation handle is less; frequent start-stop: multiple stops and starts within a certain period of time; heavy-load transportation: the load exceeds 80% of the rated load;

[0059] Combining the driving intention and the real-time working conditions to formulate a power distribution plan: For high-speed cruise, on the basis of the original distribution, the power ratio of the lithium battery increases by 10% - 15%, and at the same time, the motor speed is optimized to the efficient range; for frequent start-stop, the lithium battery and the super capacitor supply power together, and the fast discharge characteristic of the super capacitor is used to assist in starting; energy-saving intention: optimize the motor speed to the efficient range and reduce the power output of the lithium battery; climbing intention: if heavy load is detected, increase the upper limit of the lithium battery output power by 5% - 10%, and at the same time optimize the motor torque control strategy; if the lithium battery has low power but the super capacitor is sufficient, increase the power output ratio of the super capacitor;

[0060] Sending the power distribution strategy to the execution layer, adjusting the power output of the lithium battery, super capacitor and the motor control parameters, and continuously monitoring the battery temperature and the motor working state during execution.

[0061] Preferably, inputting the collected multi-dimensional driving-related data into a deep learning model to identify the driving intention specifically means:

[0062] Real-time collect multi-dimensional driving-related data through sensors, including but not limited to the position of the acceleration and speed control handle, the rate of change of position, the duration of change, vehicle speed, driving slope, and motor speed. Calculate the change amplitude of the speed control handle position and the characteristic quantity of the speed change rate within a short period of time. At the same time, combine the slope data to judge the driving road conditions of the vehicle and construct a feature vector that can reflect the driving intention;

[0063] Label the collected data according to different driving intentions, divide the training set, validation set, and test set in proportion, and normalize the data to make each feature data have the same scale range. At the same time, use oversampling or undersampling methods to balance the quantity of different category data;

[0064] Use the long short-term memory network (LSTM) to build a driving intention recognition model. Input the preprocessed training set data into the model and continuously adjust the model parameters through the backpropagation algorithm to minimize the error between the predicted driving intention and the labeled intention;

[0065] Input the newly collected real-time data into the trained model. The model outputs the probability distribution of each driving intention category, and select the category with the highest probability as the predicted driving intention. If the maximum probability is lower than the set confidence threshold, no intention decision is made or other auxiliary methods are used for judgment. According to the predicted driving intention, dynamically adjust the power distribution strategy;

[0066] Use the test set to evaluate the final performance of the model. Adjust the model architecture or parameters according to the evaluation results, and regularly add new data to the training set to perform incremental training on the model.

[0067] Preferably, in the combination of driving intention and real-time working conditions to formulate a power distribution plan, optimize the motor speed to the efficient interval, specifically:

[0068] Use sensors to collect the motor speed n and torque T in real time;

[0069] Preset the high-efficiency interval of the motor, with the speed range of n1 ≤ n ≤ n2 and the torque range of T1 ≤ T ≤ T2, where n1 and n2 are the high-efficiency speed boundary values, and T1 and T2 are the high-efficiency torque boundary values;

[0070] According to the real-time motor speed n and torque T, determine the power distribution ratio through the following function:

[0071] When n1 ≤ n ≤ n2 and T1 ≤ T ≤ T2, it indicates that the motor is in the high-efficiency interval. At this time, k = k base + Δk, where k base is the basic power distribution ratio, and Δk is the high-efficiency interval gain, with a value range of 0.1 ≤ Δk ≤ 0.2;

[0072] When n < n1 or n > n2, or T < T1 or T > T2, it indicates that the motor is in the low-efficiency range, and at this time, k = k base -Δk, Δk are the gains in the low-efficiency range, and the value range is 0.1 ≤ Δk ≤ 0.15;

[0073] Continuously and cyclically monitor the rotational speed n and torque T of the motor, and substitute the real-time data into the above power distribution ratio function to dynamically adjust the power distribution.

[0074] The present invention also proposes an intelligent new energy locomotive power distribution and energy recovery control method, including the following steps:

[0075] Through vehicle state sensors and environmental perception sensors, multi-source data of the vehicle's own state information and external environmental data are collected in real time;

[0076] Perform preprocessing of fusion, noise reduction, and normalization on the collected multi-source data to form standardized input data;

[0077] Perform fuzzy processing on the input variables that are difficult to accurately quantify after preprocessing, divide the fuzzy sets and define the membership functions, perform logical reasoning based on the fuzzy rule base, and obtain the fuzzy outputs of the power distribution ratio and energy recovery intensity through defuzzification calculation;

[0078] Adopt the Actor-Critic network architecture, combine the preprocessed multi-source data and the fuzzy processing results of the fuzzy logic output to generate action instructions for power distribution and energy recovery;

[0079] Design a comprehensive reward function including energy recovery reward, efficiency reward, safety reward, ride comfort reward, and battery health reward to guide the reinforcement learning algorithm to perform multi-objective collaborative optimization on the power distribution and energy recovery strategies;

[0080] According to the generated action instructions, control the motor controller to adjust the power output of the lithium battery / supercapacitor, and control the braking system to perform the energy recovery action of the coordinated electric braking and mechanical braking;

[0081] Monitor the vehicle running state in real time, transmit the feedback information to the intelligent decision-making control layer, and update the parameters of the reinforcement learning algorithm based on the feedback data to dynamically optimize the power distribution and energy recovery strategies.

[0082] The beneficial effects of the above technical solutions of the present invention are as follows:

[0083] 1. The present invention introduces a recovery strategy based on road condition prediction, battery state adaptive recovery, and intelligent collaborative braking control, significantly improving the energy recovery efficiency under complex road conditions. This not only increases the cruising range of the locomotive, reduces the dependence on external charging facilities, and lowers the operating cost, but also avoids overcharging and over-discharging of the battery, extends the service life of the battery, and reduces the frequency and cost of battery replacement.

[0084] 2. Based on driving intention recognition and power response, real-time working condition adaptive power distribution, and multi-power source collaborative optimization, the present invention can dynamically adjust in real time and accurately according to various factors. Under different working conditions, the motor can always operate in the high-efficiency range, and the power output is more stable and efficient. In addition, the collaborative optimization of multi-power sources gives full play to the advantages of each power source, improves the energy utilization efficiency, and reduces energy waste.

[0085] 3. The present invention applies intelligent control algorithms such as the fusion of fuzzy logic and reinforcement learning, dynamic target adjustment based on scenario recognition, and real-time environment perception and adaptive control, enabling the locomotive to automatically make optimal decisions according to complex and changing environments and driving requirements. The driving performance and energy efficiency are better balanced, achieving efficient utilization of energy while ensuring driving safety and comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 is the architecture diagram of the power distribution and energy recovery control system of the intelligent new energy locomotive of the present invention;

[0087] Figure 2 is the flowchart of the power distribution and energy recovery control method of the intelligent new energy locomotive of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0088] To make the technical problems, technical solutions, and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0089] As Figure 1 shown, the present invention proposes a power distribution and energy recovery control system for an intelligent new energy locomotive, including:

[0090] A multi-source information perception layer, including vehicle state sensors and environment perception sensors, for real-time collection of vehicle self-state information and external environment data; the vehicle state sensors include a vehicle speed sensor, an acceleration sensor, a battery state sensor, a motor torque sensor, and a wheel speed sensor; the environment perception sensors include a lidar, a camera, a meteorological sensor, and an electronic map module, where:

[0091] Vehicle speed sensor: A high-precision Hall effect vehicle speed sensor is selected and installed near the wheel to accurately measure the driving speed of the vehicle by sensing the rotation of the wheel. The pulse signal output by the sensor is processed by the signal conditioning circuit and then transmitted to the data acquisition module of the multi-source information perception layer.

[0092] Acceleration sensor: An MEMS acceleration sensor is adopted and installed near the center of gravity of the vehicle, which can accurately measure the acceleration changes of the vehicle on three axes (X, Y, and Z axes). The output signal of the sensor is amplified and filtered and then input to the data acquisition module.

[0093] Battery state sensor: Integrated in an advanced battery management system (BMS), it can accurately monitor the SOC of the battery and can also measure parameters such as SOH and temperature of the battery. The battery state sensor communicates with the data acquisition module of the multi-source information perception layer through the CAN bus to transmit the battery state information in real time.

[0094] Motor torque sensor: A high-precision strain gauge torque sensor is selected and installed on the shaft between the motor and the transmission system to measure the output torque of the motor in real time. The signal of the sensor is converted and amplified and then transmitted to the data acquisition module.

[0095] Wheel speed sensor: A magnetoelectric wheel speed sensor is adopted and installed on the hub of each wheel to obtain the rotation information of the wheel by sensing the wheel speed. The sine wave signal output by the sensor is converted into a square wave signal by the shaping circuit and then transmitted to the data acquisition module.

[0096] LiDAR: A high-performance solid-state LiDAR is selected and installed at the front and side of the vehicle to scan the surrounding environment of the vehicle in real time and obtain information such as the distance, position, and shape of obstacles. The LiDAR performs high-speed data transmission with the data acquisition module of the multi-source information perception layer through the Ethernet interface.

[0097] Camera: An in-vehicle camera with high resolution and wide viewing angle is equipped, which has good image capture ability and dynamic range. It is installed at the front, rear, and side of the vehicle respectively to identify information such as the road conditions ahead, traffic signs, and other vehicles.

[0098] Weather sensor: An integrated weather sensor is adopted, which can detect meteorological parameters such as track humidity, rainfall, and temperature in real time. It is installed on the roof of the vehicle or other well-ventilated positions and transmits the collected data to the data acquisition module through serial communication.

[0099] Electronic Map Module: It conducts real-time data interaction with the in-vehicle positioning system (such as GPS / Beidou) and the map server to obtain high-precision map information. The electronic map module communicates with the data acquisition module of the multi-source information perception layer through a network interface, providing real-time information such as road slope, curvature, and traffic signs.

[0100] Intelligent Decision and Control Layer, including a data fusion and preprocessing module, a fuzzy logic assistance module, a reinforcement learning decision module, and a multi-objective optimization module; the data fusion and preprocessing module performs preprocessing on the collected multi-source data, including but not limited to fusion, noise reduction, and normalization; the fuzzy logic assistance module performs fuzzy logic processing on the multi-source data that is difficult to quantify after preprocessing, fuzzifying the input variables and defuzzifying the output variables through a fuzzy rule base; the reinforcement learning decision module is based on the deep reinforcement learning algorithm, adopts the Actor-Critic network architecture, and generates action instructions for power distribution and energy recovery according to the multi-source data after preprocessing and fuzzy processing; the multi-objective optimization module guides the reinforcement learning algorithm to achieve multi-objective collaborative optimization by designing a reward function that synthesizes multiple objectives;

[0101] Execution and Feedback Layer, including a power execution module and a feedback monitoring module; the power execution module controls the motor controller and the braking system to perform power distribution and energy recovery actions according to the instructions generated by the intelligent decision and control layer; the feedback monitoring module monitors the actual operating state of the vehicle in real time and transmits the feedback information back to the intelligent decision and control layer, and the multi-objective optimization module optimizes the energy recovery actions and power distribution actions. The power execution module specifically includes a motor controller and a brake controller, where:

[0102] The motor controller adopts an advanced permanent magnet synchronous motor controller, supports the direct torque control (DTC) mode. The motor controller communicates with the power execution module of the intelligent decision and control layer through the CAN bus, receives the power distribution instruction, and precisely controls the output power and torque of the motor. At the same time, the motor controller monitors the operating state of the motor in real time, such as current, voltage, temperature, etc., and feeds back this information to the power execution module. The braking system includes an electric brake controller and a mechanical brake controller. The electric brake controller works in coordination with the motor controller to control the reverse torque of the motor according to the energy recovery instruction to achieve electric braking. The braking system communicates with the power execution module through the CAN bus and receives the braking instruction.

[0103] In this embodiment, the state space of the reinforcement learning decision module is:

[0104] S = {v, a, SOC, SOH, T m , λ, u, θ, weather, traffic}, and the action space A = {k bat , k sc , τ rec, where v is the vehicle speed, a is the acceleration, SOC is the state of charge of the battery, SOH is the state of health of the battery, T m is the motor torque, λ is the wheel slip ratio, u is the track adhesion coefficient, θ is the track gradient, weather is the weather condition, traffic is the traffic flow condition, k bat is the proportion of lithium battery power, k sc is the proportion of supercapacitor power, τ rec is the electric braking torque. These state variables are encoded and converted into a vector form suitable for neural network processing. For example, after normalizing continuous variables such as vehicle speed and acceleration, they are concatenated with other discrete variables to form a multi-dimensional state vector. When generating actions, the value ranges of k bat and τ rec are restricted to ensure they are within a reasonable and safe range.

[0105] Based on the multi-source data collected and the fuzzified information provided by the fuzzy logic assistance module, the reinforcement learning decision-making module dynamically adjusts the action space parameters including, but not limited to, the proportion of lithium battery power k bat , the proportion of supercapacitor power k sc and the electric braking torque τ rec in combination with the driving conditions, vehicle operating parameters and the reward function of the multi-objective optimization module. For factors that are difficult to accurately quantify (such as the opening of the acceleration and speed regulation handle, braking intensity, track slipperiness), they are processed through fuzzy logic and integrated with the reinforcement learning algorithm to optimize the power distribution and energy recovery strategy. For example, through fuzzy logic processing, the track slipperiness is divided into fuzzy sets such as "low slipperiness", "medium slipperiness", "high slipperiness", and reasoning is carried out through the fuzzy rule base. For example, when it is judged that the track is "highly slippery", the fuzzy logic module outputs the corresponding control signal and transmits it to the reinforcement learning decision-making module, prompting it to automatically reduce the upper limit of the electric braking torque (such as reducing by 20%-30%), and at the same time increasing the proportion of mechanical braking to ensure braking safety.

[0106] Based on the supplementary information provided by the fuzzy logic module and combined with other sensor data, the reinforcement learning algorithm further optimizes the power distribution and energy recovery strategy. For example, in the case of a "highly slippery" track, in addition to reducing the electric braking torque, the reinforcement learning algorithm will also comprehensively consider factors such as the current speed of the vehicle, battery state and driving conditions. If the vehicle is in a high-speed driving state at this time, the algorithm will, while reducing the electric braking torque, appropriately adjust the power distribution and reduce the output power of the motor to prevent the vehicle from losing control due to excessive power; if the vehicle is climbing a slope, the algorithm will, on the premise of ensuring safety, try to balance the power output and braking requirements to ensure that the vehicle can drive stably and recover energy reasonably. Through this integration method, the system can make more intelligent and reasonable decisions when facing complex and difficult-to-accurately-quantify situations.

[0107] In this embodiment, the fuzzy logic assistance module performs fuzzy logic processing, fuzzifying the input variables and defuzzifying the output variables through the fuzzy rule base, specifically as follows:

[0108] Vehicle operating parameters are collected in real time, including but not limited to the opening of the acceleration and speed control handle, the deceleration and braking intensity, the vehicle speed change rate, the state of charge of the lithium battery, and the state of charge of the supercapacitor. These parameters are divided into corresponding fuzzy sets, and membership functions are defined for each fuzzy set. Taking the track wetness level as an example, the "low wetness" fuzzy set uses a triangular membership function. The base of the function covers the range of track humidity from dry to slightly wet, and the vertex corresponds to a lower humidity value, indicating that the membership degree of "low wetness" is 1 at this humidity value. As the humidity value deviates from the vertex, the membership degree gradually decreases. The "medium wetness" fuzzy set can use a trapezoidal membership function, where the membership degree remains 1 within a certain humidity interval and gradually decreases at both ends, representing that the track is in a medium wetness state within this interval. For driving intention, taking "acceleration intention" as an example, if it is defined based on the change rate and duration of the acceleration and speed control handle position, an appropriate membership function can be used to quantify the degree to which it belongs to the "acceleration intention".

[0109] A fuzzy rule base is constructed for fuzzy inference. The activation degree is calculated for each rule to generate a fuzzy output set. Based on the experience of domain experts and actual driving conditions, a fuzzy rule base is constructed. For example, for the relationship between the track wetness level and the braking strategy, there may be the following rules: If the track wetness level is "high wetness" and the vehicle speed is relatively fast, then the electric braking torque should be reduced and the mechanical braking ratio should be increased; If the track wetness level is "low wetness" and the vehicle is in a normal driving state, then the current braking strategy should be maintained. For the relationship between driving intention and power distribution, the rule may be: If the driving intention is "acceleration intention" and the battery power is sufficient, then the proportion of the lithium battery power should be increased. The rules in the fuzzy rule base are presented in the form of "If..., then...", and logical reasoning is used to guide the system decision-making. When sensor data is obtained, first, the precise input data (such as the track humidity value, the change rate of the acceleration and speed control handle, etc.) is fuzzified through the membership function to obtain its membership degree in each fuzzy set. Then, fuzzy inference is performed according to the fuzzy rule base, using an appropriate inference method (such as the Mamdani inference method), and based on the input fuzzy sets and the rules in the rule base, the output fuzzy set is calculated.

[0110] The centroid method is used for defuzzification, and the power distribution ratio and the energy recovery intensity coefficient are output to the reinforcement learning decision module. Finally, through the defuzzification operation, the output fuzzy set is converted into an exact control quantity. For example, for the adjustment of the electric braking torque in the braking strategy, the centroid method is used for defuzzification to obtain a specific value, which is used to guide the control of the braking system.

[0111] In this embodiment, the output power distribution ratio and the energy recovery intensity coefficient specifically include:

[0112] (1) Method for controlling the output of the power distribution ratio:

[0113] Based on the driving demand and the current operating state of the vehicle, the locomotive control system calculates the total required power P based on the vehicle dynamics model and the driver's operation instructions. total ;

[0114] The power distribution ratio obtained according to the fuzzy control algorithm According to the formula Calculate the output power P of the lithium battery bat , where η bat is the conversion efficiency of the lithium battery;

[0115] Send the calculated output power P bat to the battery management system, monitor the state of the lithium battery, and adjust the output current of the lithium battery according to the calculated output power P bat Realize the required current magnitude by controlling the switching device in the battery charge and discharge circuit, where U is the output voltage of the lithium battery; bat is the output voltage of the lithium battery;

[0116] The battery management system monitors the voltage, current and temperature of the lithium battery in real time. If it exceeds the safety threshold, it automatically adjusts the output power; at the same time, it feeds back the actual output power and status information of the battery to the locomotive control system;

[0117] (2) Method for controlling the output of the energy recovery intensity coefficient:

[0118] The system determines whether the locomotive enters the energy recovery working condition by monitoring the vehicle speed change rate v` and the opening signal β of the braking speed control handle;

[0119] The energy recovery intensity obtained according to the fuzzy control algorithm Combined with the kinetic energy of the vehicle and the braking demand, according to Calculate the target power P of energy recovery rec , where k adj is the adjustment coefficient;

[0120] The locomotive control system sends the target power P of energy recovery rec to the motor controller. The motor controller switches the motor from the driving mode to the generating mode under the energy recovery working condition, and generates a reverse torque by changing the control signal of the motor to convert the kinetic energy of the vehicle into electrical energy;

[0121] The electrical energy generated by the motor is transmitted to the energy storage device through an inverter and a charging circuit. The system distributes energy according to the state of charge of the lithium battery and the supercapacitor, preferentially storing it in the device with a lower state of charge, and monitors the state of the energy storage device.

[0122] During the energy recovery process, the system monitors the vehicle operating state and the energy storage device state in real time, and dynamically adjusts the energy recovery intensity and the energy distribution strategy according to the actual situation.

[0123] In this embodiment, the reward function of the multi-objective optimization module is:

[0124] R = αR r + βR e + γR s + δR c + εR b

[0125] α, β, γ, δ, ε are the weight coefficients of the energy recovery reward R r , the efficiency reward R e , the safety reward R s , the ride comfort reward R c , and the battery health reward R b respectively, and are used to guide the reinforcement learning algorithm to achieve multi-objective collaborative optimization;

[0126] Among them, the energy recovery reward R r = E c - E l , E c represents the recovered energy, and E l is the energy loss during the recovery process;

[0127] The efficiency reward where is the actual output power of the battery, is the optimal power of the battery, is the maximum power of the battery;

[0128] R s is the safety reward, and a corresponding penalty is given when the vehicle is detected to be in a dangerous state;

[0129] Δv is the change in speed, and Δt is the change in time

[0130] The battery health reward R b = -λ×(ΔSOC) 2 - u×(ΔT) 2 , where: λ, u are coefficients adjusted according to the battery characteristics; ΔSOC is the change in the battery SOC; ΔT is the change in the battery temperature, which is used to protect the battery life and reduce the damage to the battery caused by the drastic changes in SOC and temperature.

[0131] In this embodiment, the optimization method for the energy recovery action of the multi-objective optimization module specifically includes a slope road condition dynamic adjustment strategy, a battery SOC adaptive strategy, and a slip ratio control strategy, where:

[0132] The slope road condition dynamic adjustment strategy uses a high-precision map and an environment perception sensor to obtain the road condition information ahead in advance, and dynamically adjusts the energy recovery intensity according to different road conditions. The slope road condition dynamic adjustment strategy satisfies the slope threshold piecewise function. Let θ be the track slope and the recovery intensity adjustment coefficient be K θ , and the recovery intensity adaptive function be f(θ), then it satisfies the formula:

[0133]

[0134] Among them, f(θ) is positively correlated with the slope, f(θ) = α·θ, α is the slope sensitivity coefficient, and the value range is 0.01 - 0.02;

[0135] For the battery SOC adaptive strategy, let the state of charge of the battery be SOC and the recovery intensity adjustment coefficient be k SOC , and the battery SOC adaptive strategy satisfies the formula;

[0136]

[0137] For the slip ratio control strategy, let the wheel slip ratio be λ and the electric braking torque be T rec , and the maximum electric braking torque be T max , and it satisfies the formula:

[0138]

[0139] Among them, ΔT = k·(T max -T rec ), and k is the adjustment coefficient.

[0140] In this embodiment, the power distribution optimization method of the multi-objective optimization module includes:

[0141] Using sensors and monitoring devices to collect power distribution analysis and control parameters in real time, including but not limited to the position and change rate of the acceleration and speed control handle, vehicle speed, driving slope, load pressure, state of charge of the lithium battery, state of charge of the super capacitor, and internal resistance;

[0142] Inputting the collected multi-dimensional driving-related data into a deep learning model to identify the driving intention: if the change rate of the acceleration and speed control handle position is greater than 10% / s and the duration is greater than 0.5 s, it is determined as an acceleration intention; if the speed control handle changes gently and the vehicle speed is stable, it is determined as an energy-saving intention; if it is detected that the slope is greater than 3% and the position of the speed control handle is relatively high, it is determined as a climbing intention;

[0143] Divide the operating conditions according to the vehicle driving state, including: High-speed cruise: The vehicle speed is stable at greater than 80 km / h and the speed control handle operation is less; Frequent start-stop: Multiple stops and starts within a certain period of time; Heavy-load transportation: The load exceeds 80% of the rated load;

[0144] Combine the driving intention and the real-time operating condition to formulate a power distribution plan: If it is high-speed cruise, on the basis of the original distribution, the power ratio of the lithium battery increases by 10%-15%, and at the same time, optimize the motor speed to the efficient range; If it is frequent start-stop, the lithium battery and the supercapacitor supply power in coordination, and use the fast discharge characteristic of the supercapacitor to assist in starting; Energy-saving intention: Optimize the motor speed to the efficient range and reduce the power output of the lithium battery; Climbing intention: If heavy load is detected, increase the upper limit of the lithium battery output power by 5%-10%, and at the same time optimize the motor torque control strategy; If the lithium battery has low power but the supercapacitor is sufficient, increase the power output ratio of the supercapacitor;

[0145] Send the power distribution strategy to the execution layer, adjust the power output of the lithium battery and the supercapacitor and the motor control parameters, and continuously monitor the battery temperature and the motor working state during execution.

[0146] In this embodiment, input the collected multi-dimensional data related to driving into the deep learning model to identify the driving intention, specifically:

[0147] Collect multi-dimensional data related to driving in real time through sensors, including but not limited to the position of the acceleration speed control handle, the position change rate, the change duration, the vehicle speed, the driving slope, and the motor speed, calculate the characteristic quantities of the change amplitude of the speed control handle position and the speed change rate in a short time, and at the same time combine the slope data to judge the vehicle driving road condition, and construct a characteristic vector that can reflect the driving intention;

[0148] Label the collected data according to different driving intentions, divide the training set, validation set and test set according to a certain proportion, normalize the data to make each characteristic data have the same scale range, and at the same time use oversampling or undersampling methods to balance the number of different category data;

[0149] Use the long short-term memory network LSTM to build a driving intention recognition model, input the preprocessed training set data into the model, and continuously adjust the model parameters through the backpropagation algorithm to minimize the error between the predicted driving intention and the labeled intention;

[0150] Input the newly collected real-time data into the trained model, the model outputs the probability distribution of each driving intention category, select the category with the highest probability as the predicted driving intention, if the highest probability is lower than the set confidence threshold, do not make an intention decision or use other auxiliary methods to judge, and dynamically adjust the power distribution strategy according to the predicted driving intention;

[0151] Use the test set to conduct a final performance evaluation of the model. Adjust the model architecture or parameters according to the evaluation results, and regularly add new data to the training set to perform incremental training on the model.

[0152] In this embodiment, when formulating a power distribution plan by combining driving intention and real-time working conditions, the motor speed is optimized to the high-efficiency range, specifically:

[0153] Use sensors to collect the speed n and torque T of the motor in real time;

[0154] Preset the high-efficiency range of the motor, where the speed range is n1 ≤ n ≤ n2, and the torque range is T1 ≤ T ≤ T2, where n1 and n2 are the high-efficiency speed boundary values, and T1 and T2 are the high-efficiency torque boundary values;

[0155] According to the real-time speed n and torque T of the motor, determine the power distribution ratio through the following function:

[0156] When n1 ≤ n ≤ n2 and T1 ≤ T ≤ T2, it indicates that the motor is in the high-efficiency range. At this time, k = k base + Δk, where k base is the basic power distribution ratio, and Δk is the high-efficiency range gain, with a value range of 0.1 ≤ Δk ≤ 0.2;

[0157] When n < n1 or n > n2 or T < T1 or T > T2, it indicates that the motor is in the low-efficiency range. At this time, k = k base - Δk, Δk is the low-efficiency range gain, with a value range of 0.1 ≤ Δk ≤ 0.15;

[0158] Continuously monitor the speed n and torque T of the motor in a loop, and substitute the real-time data into the above power distribution ratio function to dynamically adjust the power distribution.

[0159] As Figure 2 shown, the present invention also proposes an intelligent new energy locomotive power distribution and energy recovery control method, including the following steps:

[0160] S1. Through vehicle state sensors and environmental perception sensors, collect multi-source data of the vehicle's own state information and external environment data in real time;

[0161] S2. Perform preprocessing of fusion, noise reduction, and normalization on the collected multi-source data to form standardized input data;

[0162] S3. Perform fuzzification processing on the input variables that are difficult to accurately quantify after preprocessing, divide the fuzzy sets and define the membership functions, perform logical reasoning based on the fuzzy rule base, and obtain the fuzzy outputs of the power distribution ratio and energy recovery intensity through defuzzification calculation;

[0163] S4. Adopt the Actor-Critic network architecture, combine the preprocessed multi-source data and the fuzzy processing results output by fuzzy logic to generate action instructions for power distribution and energy recovery;

[0164] S5. Design a comprehensive reward function including energy recovery reward, efficiency reward, safety reward, ride comfort reward, and battery health reward to guide the reinforcement learning algorithm to perform multi-objective collaborative optimization on the power distribution and energy recovery strategies;

[0165] S6. According to the generated action instructions, control the motor controller to adjust the power output of the lithium battery / supercapacitor, and control the braking system to perform energy recovery actions that coordinate electric braking and mechanical braking;

[0166] S7. Monitor the vehicle operation state in real time, transmit the feedback information to the intelligent decision-making control layer, and update the parameters of the reinforcement learning algorithm based on the feedback data to dynamically optimize the power distribution and energy recovery strategies.

[0167] In summary, for the intelligent new energy locomotive power distribution and energy recovery control system and method of the present invention, through the coordinated operation of the multi-source information perception, intelligent decision-making control, and execution feedback layers, efficient power distribution and energy recovery are achieved, and the overall performance of the locomotive is improved.

[0168] The multi-source information perception layer is the foundation of the entire system, and the vehicle state sensors and environmental perception sensors play a key role. The vehicle speed sensor, acceleration sensor, battery state sensor, etc. collect the vehicle's own state information in real time, providing key data such as vehicle speed, acceleration, and battery charge state for the system; the lidar, camera, meteorological sensor, and electronic map module collect external environmental data, such as road conditions, weather conditions, etc.

[0169] After receiving the data from the perception layer, the intelligent decision-making control layer's data fusion and preprocessing module first performs preprocessing on the multi-source data, such as fusion, noise reduction, and normalization, to make the data more suitable for subsequent processing. The fuzzy logic assistance module performs fuzzy logic processing on the data that is difficult to quantify, fuzzifies the input variables and defuzzifies the output variables through the fuzzy rule base, and provides supplementary information for the reinforcement learning decision-making module. The reinforcement learning decision-making module, based on the deep reinforcement learning algorithm and the Actor-Critic network architecture, combines the preprocessed and fuzzy-processed data to generate action instructions for power distribution and energy recovery. The multi-objective optimization module guides the reinforcement learning algorithm to achieve multi-objective collaborative optimization by designing a comprehensive multi-objective reward function, ensuring that the system can make optimal decisions under different working conditions.

[0170] The execution feedback layer is responsible for converting decision instructions into actual actions. The power execution module controls the motor controller and the braking system to perform power distribution and energy recovery actions according to the instructions of the intelligent decision-making control layer. The feedback monitoring module monitors the actual operating state of the vehicle in real time and transmits the feedback information back to the intelligent decision-making control layer. The multi-objective optimization module optimizes the energy recovery actions and power distribution actions based on the feedback information to form a closed-loop control and continuously improve the system performance.

[0171] All layers of the present invention cooperate closely to achieve precise control of the power distribution and energy recovery of new energy locomotives. Under complex road conditions, the system can intelligently adjust the power distribution and energy recovery strategies according to the real-time road conditions and vehicle status, improve the energy recovery efficiency, and reduce energy waste. By optimizing the motor speed and the coordinated control of multiple power sources, the smoothness and efficiency of the power output are improved, taking into account both driving performance and energy efficiency.

[0172] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art of this technology, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent new energy locomotive power distribution and energy recovery control system, characterized in that, Including: A multi-source information perception layer, including vehicle state sensors and environmental perception sensors, for collecting vehicle own state information and external environment data in real time; the vehicle state sensors include a vehicle speed sensor, an acceleration sensor, a battery state sensor, a motor torque sensor, and a wheel speed sensor; the environmental perception sensors include a lidar, a camera, a meteorological sensor, and an electronic map module; An intelligent decision-making and control layer, including a data fusion and preprocessing module, a fuzzy logic assistance module, a reinforcement learning decision-making module, and a multi-objective optimization module; the data fusion and preprocessing module performs preprocessing on the collected multi-source data, including but not limited to fusion, noise reduction, and normalization; The fuzzy logic assistance module performs fuzzy logic processing on the multi-source data that is difficult to quantify after preprocessing, and fuzzifies the input variables and defuzzifies the output variables through a fuzzy rule base; The reinforcement learning decision-making module is based on a deep reinforcement learning algorithm, adopts an Actor-Critic network architecture, and generates action instructions for power distribution and energy recovery according to the multi-source data after preprocessing and fuzzy processing; the multi-objective optimization module guides the reinforcement learning algorithm to achieve multi-objective collaborative optimization by designing a reward function that synthesizes multiple objectives; An execution and feedback layer, including a power execution module and a feedback monitoring module; the power execution module controls the motor controller and the braking system to perform power distribution and energy recovery actions according to the instructions generated by the intelligent decision-making and control layer; the feedback monitoring module monitors the actual operating state of the vehicle in real time and transmits the feedback information back to the intelligent decision-making and control layer, and the multi-objective optimization module optimizes the energy recovery action and the power distribution action.

2. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 1, characterized in that, The state space of the reinforcement learning decision-making module is: S = {v, a, SOC, SOH, T m , λ, u, θ, weather, traffic}, the action space A = {k bat , k sc , τ rec}, where v is the vehicle speed, a is the acceleration, SOC is the state of charge of the battery, SOH is the state of health of the battery, T m is the motor torque, λ is the wheel slip ratio, u is the track adhesion coefficient, θ is the track gradient, weather is the weather condition, traffic is the traffic flow condition, k bat is the power ratio of the lithium battery, k sc is the power ratio of the supercapacitor, τ rec is the electric braking torque; The reinforcement learning decision-making module dynamically adjusts the action space parameters including, but not limited to, the lithium battery power ratio k, bat the supercapacitor power ratio k, sc and the electric braking torque τ rec according to the collected multi-source data and the fuzzified information provided by the fuzzy logic assistance module, in combination with the driving conditions, vehicle operating parameters, and the reward function of the multi-objective optimization module.

3. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 1, characterized in that The fuzzy logic assistance module performs fuzzy logic processing, and fuzzifies the input variables and defuzzifies the output variables through a fuzzy rule base. Specifically: Collect vehicle operation parameters in real time, including but not limited to the opening of the acceleration and speed control handle, the deceleration and braking intensity, the vehicle speed change rate, the state of charge of the lithium battery, and the state of charge of the super capacitor, and divide them into corresponding fuzzy sets, and define membership functions for each fuzzy set; Construct a fuzzy rule base for fuzzy inference, calculate the activation degree for each rule, and generate a fuzzy output set; Use the centroid method for defuzzification, and output the power distribution ratio and the energy recovery intensity coefficient to the reinforcement learning decision-making module.

4. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 3, characterized in that, The output power distribution ratio and energy recovery intensity coefficient specifically include: (1) Power distribution ratio output control method: The locomotive control system calculates the total required power P based on the driving demand and the current operating state of the vehicle, using the vehicle dynamics model and the driver's operation instructions. total ; Power distribution ratio obtained according to the fuzzy control algorithm According to the formula Calculate the output power P of the lithium battery bat , where η bat is the conversion efficiency of the lithium battery; Send the calculated output power P bat to the battery management system to monitor the state of the lithium battery and adjust the output current of the lithium battery according to the calculated output power P bat The required current magnitude is achieved by controlling the switching device in the battery charge and discharge circuit, where U bat is the output voltage of the lithium battery;​ The battery management system monitors the voltage, current, and temperature of the lithium battery in real time. If it exceeds the safety threshold, it automatically adjusts the output power; at the same time, it feeds back the actual output power and state information of the battery to the locomotive control system; (2) Energy recovery intensity coefficient output control method: The system determines whether the locomotive enters the energy recovery working condition by monitoring the vehicle speed change rate v 、 and the opening signal β of the braking speed control handle The energy recovery intensity obtained according to the fuzzy control algorithm Combined with the kinetic energy of the vehicle and the braking demand, according to Calculate the target power P of energy recovery rec , where k adj is the adjustment coefficient; The locomotive control system sends the target power P for energy recovery rec to the motor controller. Under the energy recovery condition, the motor controller switches the motor from the driving mode to the generating mode, generates a reverse torque by changing the control signal of the motor, and converts the kinetic energy of the vehicle into electrical energy; The electric energy generated by the motor is transmitted to the energy storage device through an inverter and a charging circuit. The system distributes energy according to the state of charge of the lithium battery and the super capacitor, and preferentially stores it in the device with a lower state of charge, and monitors the state of the energy storage device; During the energy recovery process, the system monitors the vehicle operating status and the status of the energy storage device in real time, and dynamically adjusts the energy recovery intensity and the energy distribution strategy according to the actual situation.

5. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 1, characterized in that, The reward function of the multi-objective optimization module is as follows: R = αR r + βR e + γR s + δR c + εR b α, β, γ, δ, ε are the weight coefficients of the energy recovery reward R r , the efficiency reward R e , the safety reward R s , the ride comfort reward R c , the battery health reward R b , used to guide the reinforcement learning algorithm to achieve multi-objective collaborative optimization; Among them, the energy recovery reward R r = E c - E l where E c represents the recovered energy, and E l is the energy loss during the recovery process; Efficiency Reward wherein is the actual output power of the battery, is the optimal power of the battery, is the maximum power of the battery; R s For safety rewards, corresponding penalties are given when the vehicle is detected to be in a dangerous state; Δv is the change in velocity, and Δt is the change in time Battery Health Reward R b = -λ × (ΔSOC) 2 - u × (ΔT) 2 , where: λ and u are coefficients adjusted according to battery characteristics; ΔSOC is the change in battery SOC; ΔT is the change in battery temperature, which is used to protect battery life and reduce the damage to the battery caused by drastic changes in SOC and temperature.

6. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 1, characterized in that, The optimization method for the energy recovery action of the multi-objective optimization module specifically includes a dynamic adjustment strategy for slope road conditions, a battery SOC adaptive strategy, and a slip ratio control strategy, where: The dynamic adjustment strategy for slope road conditions utilizes high-precision maps and environmental perception sensors to obtain road condition information ahead in advance, and dynamically adjusts the energy recovery intensity according to different road conditions. The dynamic adjustment strategy for slope road conditions satisfies the slope threshold piecewise function. Let θ be the track slope and the recovery intensity adjustment coefficient be K θ , and the recovery intensity adaptive function is f(θ), then it satisfies the formula: Among them, f(θ) is positively correlated with the slope, f(θ) = α·θ, α is the slope sensitivity coefficient, and the value range is 0.01 - 0.02; The battery SOC adaptive strategy sets the state of charge of the battery as SOC and the recovery intensity adjustment coefficient as k SOC , and the battery SOC adaptive strategy satisfies the formula; The slip ratio control strategy sets the wheel slip ratio as λ and the electric braking torque as T rec , and the maximum electric braking torque as T max , satisfying the formula: where ΔT = k·(T max - T rec ), and k is an adjustment coefficient.

7. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 1, characterized in that, The power distribution optimization method of the multi-objective optimization module includes: Using sensors and monitoring devices to collect in real time the power distribution analysis and control parameters including but not limited to the position and change rate of the acceleration and speed regulation handle, vehicle speed, driving slope, load pressure, state of charge of the lithium battery, state of charge of the super capacitor, and internal resistance; Input the collected multi-dimensional driving-related data into the deep learning model to identify the driving intention: if the change rate of the acceleration and speed regulation handle position is greater than 10% / s and the duration is greater than 0.5 s, it is determined as an acceleration intention; if the speed regulation handle changes gently and the vehicle speed is stable, it is determined as an energy-saving intention; if it is detected that the slope is greater than 3% and the position of the speed regulation handle is high, it is determined as a climbing intention; Divide the working conditions according to the vehicle driving status, including: high-speed cruise: the vehicle speed is stable at greater than 80 km / h and the operation of the speed regulation handle is less; frequent start-stop: multiple stops and starts within a certain period of time; heavy-load transportation: the load exceeds 80% of the rated load; Combined with the driving intention and the real-time working condition, formulate a power distribution plan: for high-speed cruise, on the basis of the original distribution, the power ratio of the lithium battery increases by 10% - 15%, and at the same time, the motor speed is optimized to the efficient range; for frequent start-stop, the lithium battery and the super capacitor supply power in cooperation, and use the fast discharge characteristic of the super capacitor to assist in starting; energy-saving intention: optimize the motor speed to the efficient range and reduce the power output of the lithium battery; climbing intention: if it is detected that the load is heavy, increase the upper limit of the lithium battery output power by 5% - 10%, and at the same time optimize the motor torque control strategy; if the lithium battery has a low power level but the super capacitor is sufficient, increase the power output ratio of the super capacitor; Send the power distribution strategy to the execution layer, adjust the power output of the lithium battery and the super capacitor and the motor control parameters, and continuously monitor the battery temperature and the motor working status during the execution.

8. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 7, characterized in that, The specific method of inputting the collected multi-dimensional driving-related data into the deep learning model to identify the driving intention is as follows: Collect in real time through sensors the multi-dimensional driving-related data including but not limited to the position of the acceleration and speed regulation handle, the position change rate, the change duration, the vehicle speed, the driving slope, and the motor speed, calculate the characteristic quantities of the change amplitude of the position of the speed regulation handle within a short time and the speed change rate, and at the same time combine the slope data to judge the vehicle driving road conditions, and construct a characteristic vector that can reflect the driving intention; Label the collected data according to different driving intentions, divide the training set, validation set, and test set according to a certain proportion, normalize the data to make each characteristic data have the same scale range, and at the same time use oversampling or undersampling methods to balance the number of different category data; An LSTM (Long Short-Term Memory) network is used to build a driving intention recognition model. The preprocessed training set data is input into the model, and the model parameters are continuously adjusted through the backpropagation algorithm to minimize the error between the predicted driving intention and the annotated intention. The newly collected real-time data is input into the trained model. The model outputs the probability distribution of each driving intention category, and the category with the highest probability is selected as the predicted driving intention. If the maximum probability is lower than the set confidence threshold, no intention decision is made or other auxiliary methods are used for judgment. According to the predicted driving intention, the power distribution strategy is dynamically adjusted. The test set is used to evaluate the final performance of the model. The model architecture or parameters are adjusted according to the evaluation results, and new data is regularly added to the training set for incremental training of the model.

9. The intelligent new energy locomotive power distribution and energy recovery control system according to claim 7, characterized in that In the formulation of the power distribution plan by combining the driving intention with the real-time working conditions, the motor speed is optimized to the efficient range, specifically: The sensor is used to collect the motor speed n and torque T in real time. The high-efficiency range of the motor is preset, with the speed range of n1 ≤ n ≤ n2 and the torque range of T1 ≤ T ≤ T2, where n1 and n2 are the high-efficiency speed boundary values, and T1 and T2 are the high-efficiency torque boundary values. According to the real-time motor speed n and torque T, the power distribution ratio is determined through the following function: When n1 ≤ n ≤ n2 and T1 ≤ T ≤ T2, it indicates that the motor is in the high-efficiency range. At this time, k = k base + Δk, where k base is the basic power distribution ratio, and Δk is the high-efficiency range gain, with a value range of 0.1 ≤ Δk ≤ 0.2; When n < n1 or n > n2, or T < T1 or T > T2, it indicates that the motor is in the low-efficiency range. At this time, k = k base -Δk, Δk are the gains in the low-efficiency range, and the value range is 0.1 ≤ Δk ≤ 0.15; The motor speed n and torque T are continuously and cyclically monitored, and the real-time data is substituted into the above power distribution ratio function to dynamically adjust the power distribution.

10. A control method for power distribution and energy recovery of an intelligent new energy locomotive, characterized in that, It includes the following steps: Through the vehicle state sensor and the environment perception sensor, multi-source data of the vehicle's own state information and external environment data are collected in real time. The collected multi-source data is preprocessed by fusion, noise reduction, and normalization to form standardized input data. The input variables that are difficult to accurately quantify after preprocessing are fuzzified. The fuzzy sets are divided and the membership functions are defined. Logical reasoning is carried out based on the fuzzy rule base, and the fuzzy outputs of the power distribution ratio and the energy recovery intensity are obtained through defuzzification calculation. The Actor-Critic network architecture is adopted, and combined with the preprocessed multi-source data and the fuzzy processing results of the fuzzy logic output, action instructions for power distribution and energy recovery are generated. A comprehensive reward function including energy recovery reward, efficiency reward, safety reward, ride comfort reward, and battery health reward is designed to guide the reinforcement learning algorithm to perform multi-objective collaborative optimization on the power distribution and energy recovery strategies. According to the generated action instructions, the motor controller is controlled to adjust the power output of the lithium battery / supercapacitor, and the braking system is controlled to perform the energy recovery action of the coordinated electric braking and mechanical braking. The vehicle running state is monitored in real time, and the feedback information is transmitted to the intelligent decision-making control layer. Based on the feedback data, the parameters of the reinforcement learning algorithm are updated to dynamically optimize the power distribution and energy recovery strategies.

Images