Hybrid electric vehicle PN emission prediction method based on PSO-GA-ACO hybrid optimization algorithm

Abstract

The invention discloses a hybrid electric vehicle PN emission prediction method based on a PSO-GA-ACO hybrid optimization algorithm, and relates to the technical field of traffic pollution control, and the method comprises the following steps: 1, connecting a vehicle-mounted data collection terminal of a hybrid electric vehicle with a sensor assembly, and collecting various data of the hybrid electric vehicle in the operation process in real time, comprise vehicle speed, engine speed, battery state, air flow, fuel consumption rate and catalyst temperature; according to the method, the global search capability of PSO, the local search capability of GA and the path optimization capability of ACO are comprehensively utilized, so that the prediction precision of the model on PN emission is improved, deviation and uncertainty in a prediction result are reduced, seamless switching and collaborative optimization between algorithms are realized, time consumption of a prediction process is shortened, a prediction task is completed in a relatively short time, and the prediction efficiency is improved. The real-time requirement is met, and by optimizing the structure and parameters of the model, the prediction model can better adapt to hybrid electric vehicles of different working conditions, different vehicle types or different emission standards.

Description

Technical Field

[0001] This invention relates to the field of traffic pollution control technology, specifically to a method for predicting PN emissions from hybrid electric vehicles based on a PSO-GA-ACO hybrid optimization algorithm. Background Technology

[0002] As global attention to environmental protection and sustainable development deepens, the automotive industry faces unprecedented pressure, especially in reducing greenhouse gas emissions and air pollution. Hybrid vehicles, which combine the advantages of internal combustion engines and electric motors, are gradually becoming an important direction for reducing carbon footprint and improving air quality. Compared with traditional fuel vehicles, hybrid vehicles can significantly reduce fuel consumption at low speeds by using batteries and engines working together, thereby reducing carbon dioxide emissions. However, despite their excellent performance in reducing carbon dioxide emissions, hybrid vehicles still face certain challenges in controlling particulate matter emissions.

[0003] Particulate matter refers to tiny solid particles suspended in the air by automobiles. It is mainly composed of unburned fuel, lubricating oil, and fine particles from the emission system. These particles can remain suspended in the air for a long time and have a negative impact on human health and the environment. Although the use of electric motors can effectively reduce exhaust emissions, particulate matter emissions are still a significant problem during the operation of internal combustion engines, especially during short-distance driving and low-load conditions. Therefore, how to accurately predict particulate matter emissions from hybrid vehicles and take effective measures to control them has become one of the hot topics of industry concern.

[0004] Existing technologies have the following shortcomings: due to unreasonable model parameter settings or insufficient optimization, existing PN emission prediction models result in significant deviations between predicted results and actual emission values, failing to meet high accuracy requirements. When facing high-dimensional and nonlinear problems, they involve large computational loads and long processing times, making real-time prediction difficult. Single optimization algorithms are prone to getting trapped in local optima during the search process, failing to find the globally optimal parameter combination. When facing different operating conditions, vehicle types, or emission standards, they often lack generalization ability and are unable to cope with PN emission prediction tasks under different conditions. When dealing with complex and variable emission data, they fail to fully explore the inherent patterns and characteristics of the data, resulting in limited prediction performance.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm, comprising the following steps: Step 1: Connect the sensor components to the on-board data acquisition terminal of the hybrid vehicle to collect various data during the operation of the hybrid vehicle in real time, including vehicle speed, engine speed, battery status, air flow, fuel consumption rate and catalyst temperature. Step 2: Clean and standardize the real-time collected data to remove noise and outliers; Step 3: Introduce the particle swarm optimization algorithm to quickly locate the solution space region by using multiple particles to perform a global search in the search space. Step 4: Introduce a genetic algorithm for local search and further optimize candidate solutions through selection, crossover, and mutation; Step 5: Introduce the ant colony optimization algorithm as input for further optimization of candidate solutions, find the optimal solution and output it; Step 6: Output the PN emission prediction of the hybrid vehicle through the neural network, and update and optimize the neural network by calling the particle swarm optimization algorithm, genetic algorithm and ant colony optimization algorithm according to the Monte Carlo tree.

[0008] Preferably, the on-board data acquisition terminal includes wheel speed sensors for real-time vehicle speed data acquisition, engine speed sensors for real-time engine speed data acquisition, a battery management component for real-time battery status data acquisition, an intake air flow meter for real-time airflow data acquisition, a fuel flow sensor for real-time fuel consumption rate data acquisition, and a temperature sensor for real-time catalyst temperature data acquisition. The wheel speed sensors are installed on each wheel driveshaft, the engine speed sensors are installed on the engine crankshaft, and the battery management component is integrated into the hybrid vehicle battery pack. The battery management component includes a voltage sensor for real-time battery pack voltage status acquisition and a current sensor for real-time battery pack current status acquisition. The intake air flow meter is installed before the engine intake manifold, the fuel flow sensor is installed at the fuel line inlet, and the temperature sensor is installed at the catalytic converter outlet. Furthermore, the frequencies for real-time vehicle speed data acquisition are set to 5Hz, engine speed data acquisition to 5Hz, battery status data acquisition to 2Hz, airflow data acquisition to 5Hz, fuel consumption rate data acquisition to 3Hz, and catalyst temperature data acquisition to 0.2Hz.

[0009] Preferably, a tree structure is constructed, where branch nodes represent outlier features of the real-time collected data, and leaf nodes represent outlier classification results. Multiple decision trees are constructed using randomly selected outlier features and real-time collected data, and optimal splitting is performed. Each decision tree performs independent outlier classification prediction. A voting mechanism is established to select the outlier classification results predicted by all decision trees. When more than 70% of the decision trees predict the current node as an outlier, the node is directly identified as an outlier and deleted. Duplicate values ​​in the real-time collected data are checked and deleted using a unique identifier. Missing values ​​in the real-time collected data are checked and replaced with the mean of the real-time collected data using a weighted mean imputation method. The specific formula is as follows:

[0010] in, This represents the weighted average. This represents the weight of the i-th real-time collected data. A descriptive symbol indicating whether or not to answer. To indicate "yes" Indicates "no". This indicates the amount of data collected in real time. Let represent the i-th real-time data point. Z-score standardization transforms the real-time data into a normal distribution with a mean of 0 and a standard deviation of 1. The specific formula is as follows:

[0011] in, Represents the standard scores of a normal distribution. This indicates data collected in real time. This represents the average value of the data collected in real time. This represents the standard deviation of the data collected in real time.

[0012] Preferably, the positions and velocities of the particles are randomly initialized, the individual optimal position of each particle is initialized to the current position, the optimized fitness value of each particle is calculated, the global optimal position is initialized according to the position of the particle with the best optimized fitness value among all particles, and the velocity and position of the particles are updated according to the update rules of particle swarm optimization. The specific formula is as follows:

[0013]

[0014] in, This represents the velocity of particle e in the d-th dimension and at the (t+d)-th iteration. This represents the velocity of particle e in the d-th dimension and at the t-th iteration. The inertial weights represent the influence of a particle's current velocity on its new velocity. These represent the learning factors that indicate the degree to which a particle learns from its own historical best position and the learning factors that indicate the degree to which a particle learns from the group's best position, respectively. They represent random numbers between [0, 1]. This represents the historical best position of particle e in the d-th dimension and at the t-th iteration. This represents the current position of particle e in the d-th dimension and at the t-th iteration. This represents the update position of particle e in the d-th dimension and at the t-th iteration. Let $\frac{d}{t}$ represent the global optimal position of all particles in the $d$-th dimension at the $t$-th iteration. If the current particle's optimized fitness value is better than its individual optimal fitness value, then the particle's individual optimal position and optimized fitness value are updated. If the current particle's optimized fitness value is better than the global optimal fitness value, then the global optimal position and optimized fitness value are updated. The relative error of the particle's optimized fitness value changing over multiple consecutive iterations is less than $\frac{d}{t}$. Then the global search is terminated, and the globally optimal candidate solution is output.

[0015] Preferably, a population of randomly generated candidate solutions is used, with each candidate solution treated as an individual and assigned a genetic fitness value. These solutions are then sorted according to their genetic fitness values, with priority given to those with higher values ​​for reproduction. The candidate solution with the highest genetic fitness value is selected as the parent solution. Parent solutions are repeatedly selected based on decreasing genetic fitness values. Single-point crossover is used, randomly selecting a location to cut off the parent solution as a crossover point. Gene exchange occurs between the two parent solutions at this crossover point, generating two new offspring solutions. Multi-point crossover is used, randomly selecting multiple locations to cut off the parent solutions as crossover points. Gene exchange occurs between multiple parent solutions at these crossover points, generating multiple new offspring solutions. Uniform crossover is used, with probabilistic selection for each gene locus. Crossover is then performed on selected parent solutions at the chosen gene locus, generating new offspring solutions. Candidate solutions are further optimized using various genetic crossover methods. The genetic termination condition is defined as reaching a maximum of 20 iterations. Offspring solutions that meet the termination condition are returned as candidate solutions for further optimization.

[0016] Preferably, candidate solutions are represented as vectors, and the distance between candidate solutions is calculated using Euclidean distance. The pheromone concentration value between all candidate solutions is initialized to 0.1 and stored in a pheromone matrix. A pheromone graph is constructed, with each candidate solution serving as a node or the starting point of a path in the pheromone graph. Ants calculate the probability of moving to the next node based on the pheromone concentration values ​​from the current node to other nodes and heuristic information using a proportional selection method. The specific formula is as follows:

[0017] in, Let represent the probability of an ant choosing to go from node j to node k. This represents the pheromone concentration value between node j and node k. This represents the heuristic information between node j and node k. A moderating factor indicating the degree of influence of pheromones. A moderating factor representing the degree to which heuristic information has an impact. This represents a node in the set of neighboring nodes of the current node j. Let j represent the set of neighboring nodes of the current node j. Ants select the next node based on probability until all ants have completed their movement search path. The pheromone concentration value is globally updated based on the volatile factor and the quality of the path built by the ants. The process of ant movement search and pheromone concentration value update is repeated until the optimal solution is found and output.

[0018] Preferably, the input real-time collected data is used to simultaneously invoke particle swarm optimization, genetic algorithm, and ant colony optimization algorithm to initialize the weights and biases of the neural network. The real-time collected data is then propagated forward through the neural network, and the specific formula for the forward propagation is as follows:

[0019] in This represents the input to the hidden layer of the neural network. This represents the weight matrix of the hidden layer of a neural network. A vector representation of the input data collected in real time. This represents the bias vector of the hidden layer of the neural network, which outputs the PN emission prediction for hybrid electric vehicles. The weights, biases, and parameters of the neural network are updated using gradients calculated through backpropagation. The specific formula for backpropagation calculation is as follows:

[0020] in Represents the optimization parameters of the neural network. Indicates the current neural network parameters. Indicates the learning rate. This represents the gradient obtained by differentiating the loss function with respect to parameter w. The algorithm iteratively propagates forward until the stopping criterion is met. Using a Monte Carlo tree, starting from the root node, it selects incompletely explored child nodes according to the simulation until a child node is reached. Based on the collaborative optimization mechanism of particle swarm optimization, genetic algorithm, and ant colony optimization, multiple incompletely explored child nodes are generated, and a complete simulation and random game are performed to obtain evaluation results. The results of the simulation and random game are fed back to the root node, and the visit count of each child node is updated to guide the next selection. In the simulation and random game, the particle swarm optimization, genetic algorithm, and ant colony optimization are invoked according to the Monte Carlo tree to update and optimize the neural network.

[0021] The technical effects and advantages provided by the present invention in the above technical solution are as follows: By comprehensively utilizing the global search capability of PSO, the local search capability of GA, and the path optimization capability of ACO, the parameters of the prediction model are fully optimized, improving the model's prediction accuracy for PN emissions. Through optimization algorithm combination and switching strategies, biases and uncertainties in the prediction results are reduced, improving reliability. A reasonable algorithm flow and parameter settings are designed to reduce unnecessary computation and improve computational efficiency. Leveraging the advantages of hybrid optimization algorithms reduces reliance on high-performance computing resources, lowering computational costs. Seamless switching and collaborative optimization between algorithms are achieved, shortening the prediction process time. The optimized prediction model can complete the prediction task in a shorter time, meeting real-time requirements. By optimizing the model's structure and parameters, the prediction model can better adapt to different operating conditions, vehicle types, or hybrid vehicles with different emission standards, improving its portability and adaptability. This allows the prediction model to maintain stable prediction performance under different conditions. By improving prediction accuracy and efficiency and reducing prediction costs, it can provide strong support for environmental policy formulation, vehicle design optimization, and emission control. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0023] Figure 1 This is a flowchart of the method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm of the present invention. Detailed Implementation

[0024] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0025] Example 1 This invention provides, for example Figure 1 The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm, as shown, includes the following steps: Step 1: Connect the sensor components to the on-board data acquisition terminal of the hybrid vehicle to collect various data during the operation of the hybrid vehicle in real time, including vehicle speed, engine speed, battery status, air flow, fuel consumption rate and catalyst temperature. The on-board data acquisition terminal includes wheel speed sensors for real-time vehicle speed data acquisition, engine speed sensors for real-time engine speed data acquisition, a battery management component for real-time battery status data acquisition, an intake air flow meter for real-time airflow data acquisition, a fuel flow sensor for real-time fuel consumption rate data acquisition, and a temperature sensor for real-time catalyst temperature data acquisition. The wheel speed sensors are installed on each wheel driveshaft, the engine speed sensors are installed on the engine crankshaft, and the battery management component is integrated into the hybrid vehicle battery pack. The battery management component includes a voltage sensor for real-time battery pack voltage status acquisition and a current sensor for real-time battery pack current status acquisition. The intake air flow meter is installed before the engine intake manifold, the fuel flow sensor is installed at the fuel line inlet, and the temperature sensor is installed at the catalytic converter outlet. Furthermore, the frequencies for real-time vehicle speed data acquisition are set to 5Hz, engine speed data acquisition to 5Hz, battery status data acquisition to 2Hz, airflow data acquisition to 5Hz, fuel consumption rate data acquisition to 3Hz, and catalyst temperature data acquisition to 0.2Hz.

[0026] In this embodiment, the first step specifically needs to be explained. By installing various sensors, data such as vehicle speed, engine speed, battery status, air flow, fuel consumption rate, and catalyst temperature can be acquired in real time, thereby providing basic data support for PN emissions. Setting different acquisition frequencies makes the monitoring of important parameters more precise, enabling rapid response to changes and improving vehicle safety and efficiency.

[0027] Step 2: Clean and standardize the real-time collected data to remove noise and outliers; A tree structure is constructed, where branch nodes represent outlier features of the real-time collected data, and leaf nodes represent outlier classification results. Multiple decision trees are built using randomly selected outlier features and real-time collected data, and optimal splits are performed. Each decision tree independently predicts outlier classification. A voting mechanism is established to select the classification results of all decision trees predicting outliers. If more than 70% of the decision trees predict the current node as an outlier, that node is directly identified as an outlier and deleted. Duplicate values ​​in the real-time collected data are checked and deleted using unique identifiers. Missing values ​​in the real-time collected data are checked and replaced with the mean of the real-time collected data using a weighted mean imputation method. The specific formula is as follows:

[0028] in, This represents the weighted average. This represents the weight of the i-th real-time collected data. A descriptive symbol indicating whether or not to answer. To indicate "yes" Indicates "no". This indicates the amount of data collected in real time. Let represent the i-th real-time data point. Z-score standardization transforms the real-time data into a normal distribution with a mean of 0 and a standard deviation of 1. The specific formula is as follows:

[0029] in, Represents the standard scores of a normal distribution. This indicates data collected in real time. This represents the average value of the data collected in real time. This represents the standard deviation of the data collected in real time.

[0030] In this embodiment, step two specifically needs to be explained. By comprehensively using data cleaning, anomaly detection and classification, missing value handling, and standardization techniques, the quality and availability of real-time data from hybrid electric vehicles are greatly improved. The entire process not only improves the accuracy of the system but also enhances decision-making efficiency, contributing to better optimization of PN emissions. Step 3: Introduce the particle swarm optimization algorithm to quickly locate the solution space region by using multiple particles to perform a global search in the search space. The positions and velocities of particles are randomly initialized. The individual optimal position of each particle is initialized to its current position. The optimal fitness value of each particle is calculated. The global optimal position is initialized based on the position of the particle with the best optimal fitness value among all particles. The velocity and position of the particles are updated according to the particle swarm optimization update rules. The specific formula is as follows:

[0031]

[0032] in, This represents the velocity of particle e in the d-th dimension and at the (t+d)-th iteration. This represents the velocity of particle e in the d-th dimension and at the t-th iteration. The inertial weights represent the influence of a particle's current velocity on its new velocity. These represent the learning factors that indicate the degree to which a particle learns from its own historical best position and the learning factors that indicate the degree to which a particle learns from the group's best position, respectively. They represent random numbers between [0, 1]. This represents the historical best position of particle e in the d-th dimension and at the t-th iteration. This represents the current position of particle e in the d-th dimension and at the t-th iteration. This represents the update position of particle e in the d-th dimension and at the t-th iteration. Let $\frac{d}{t}$ represent the global optimal position of all particles in the $d$-th dimension at the $t$-th iteration. If the current particle's optimized fitness value is better than its individual optimal fitness value, then the particle's individual optimal position and optimized fitness value are updated. If the current particle's optimized fitness value is better than the global optimal fitness value, then the global optimal position and optimized fitness value are updated. The relative error of the particle's optimized fitness value changing over multiple consecutive iterations is less than $\frac{d}{t}$. Then the global search is terminated, and the globally optimal candidate solution is output.

[0033] In this embodiment, step three needs to be specifically explained. The particle swarm optimization algorithm achieves global search by exploring the solution space with multiple particles, avoiding the problem of local optima. It can quickly locate the optimal solution region in a complex search space. By calculating the optimization fitness value of the particles, a fitness evaluation mechanism is formed, which enables the algorithm to dynamically adjust and improve the search process, ensuring that the particles continue to move towards better solutions. The use of update formulas based on inertia weight, individual optimality, and swarm optimality makes the adjustment of particle speed and position more flexible. The search termination condition is determined by the relative error of the optimization fitness value in multiple consecutive iterations, which ensures the effectiveness of the algorithm's convergence and can output the optimal solution in a timely manner.

[0034] Step 4: Introduce a genetic algorithm for local search and further optimize candidate solutions through selection, crossover, and mutation; A population of randomly generated candidate solutions is formed, with each candidate solution treated as an individual and assigned a genetic fitness value. Solutions are sorted according to their genetic fitness values, prioritizing those with higher values ​​for reproduction. The candidate solution with the highest genetic fitness value is selected as the parent solution. Parent solutions are repeatedly selected based on decreasing genetic fitness values. Single-point crossover is used, randomly selecting a location on the chromosome to cut off the parent solution as a crossover point. Gene exchange occurs between two parent solutions at the crossover point, generating two new offspring solutions. Multi-point crossover is used, randomly selecting multiple locations to cut off the parent solutions as crossover points. Gene exchange occurs between multiple parent solutions at these crossover points, generating multiple new offspring solutions. Uniform crossover is used, with probabilistic selection for each gene locus. Crossover is then performed on selected parent solutions at the chosen gene locus, generating new offspring solutions. Candidate solutions are further optimized using various genetic crossover methods. The genetic termination condition is defined as reaching a maximum of 20 iterations. Offspring solutions meeting the termination condition are returned as candidate solutions for further optimization.

[0035] In this embodiment, step four needs to be specifically explained. Through the biomimetic evolutionary process, local search can be performed more effectively in the neighborhood of the solution, improving the quality of the solution. Through fitness-based priority selection, it can be ensured that better candidate solutions are more likely to be passed on to the next generation, increasing the proportion of excellent features in the population and improving the overall fitness of the population. Multiple crossover methods are used to maintain the diversity of the population in the solution space and avoid premature convergence to local optima.

[0036] Step 5: Introduce the ant colony optimization algorithm as input for further optimization of candidate solutions, find the optimal solution and output it; Candidate solutions are represented as vectors. The distance between candidate solutions is calculated using Euclidean distance. The pheromone concentration of all candidate solutions is initialized to 0.1 and stored in a pheromone matrix. A pheromone graph is constructed, with each candidate solution serving as a node or the starting point of a path in the pheromone graph. The connection strength between nodes is determined by the pheromone concentration, which dynamically increases with the number of visits by the ants acting as virtual agents and gradually decreases over time. Ants calculate the probability of moving to the next node using a proportional selection method based on the pheromone concentration from the current node to other nodes and heuristic information (such as the distance between nodes). The specific formula is as follows:

[0037] in, Let represent the probability of an ant choosing to go from node j to node k. This represents the pheromone concentration value between node j and node k. This represents the heuristic information between node j and node k. A moderating factor indicating the degree of influence of pheromones. A moderating factor representing the degree to which heuristic information has an impact. This represents a node in the set of neighboring nodes of the current node j. Let represent the set of neighboring nodes of the current node j. Ants select the next node based on probability until all ants have completed their movement search path. The pheromone concentration value is globally updated based on the volatile factor and the quality of the path built by the ants. The pheromone concentration value establishes a volatile mechanism based on the volatile factor to simulate the phenomenon of pheromone evaporation over time in nature. An accumulation mechanism is established based on the inverse proportionality between the increase in pheromone concentration value and the quality of the path built by the ants to adjust the pheromone concentration value on the path. The process of ant movement search and pheromone concentration value update is repeated until the optimal solution is found and output.

[0038] In this embodiment, step five specifically needs to be explained. By utilizing the dynamic update mechanism of pheromones and simulating ant foraging behavior, a global search is performed in the solution space, which can effectively find the optimal solution. By increasing and decreasing the concentration of pheromones, the priority of path selection can be adaptively adjusted, enhancing the attractiveness of high-quality paths, thereby guiding ants to find the optimal solution faster. Combining heuristic information when selecting paths improves search efficiency and makes the algorithm more adaptable to complex problems.

[0039] Step 6: Output the PN emission prediction of hybrid electric vehicles through the neural network, and update and optimize the neural network by calling particle swarm optimization algorithm, genetic algorithm and ant colony optimization algorithm according to the Monte Carlo tree. The system inputs real-time acquired data and simultaneously invokes particle swarm optimization, genetic algorithm, and ant colony optimization algorithm to initialize the weights and biases of the neural network. The real-time acquired data is then propagated forward through the neural network, with the specific formula for the forward propagation being:

[0040] in This represents the input to the hidden layer of the neural network. This represents the weight matrix of the hidden layer of a neural network. A vector representation of the input data collected in real time. This represents the bias vector of the hidden layer of the neural network, which outputs the PN emission prediction for hybrid electric vehicles. The weights, biases, and parameters of the neural network are updated using gradients calculated through backpropagation. The specific formula for backpropagation calculation is as follows:

[0041] in Represents the optimization parameters of the neural network. Indicates the current neural network parameters. Indicates the learning rate. This represents the gradient obtained by differentiating the loss function with respect to parameter w. The algorithm iteratively propagates forward until the stopping criterion is met. Using a Monte Carlo tree, starting from the root node, it selects incompletely explored child nodes according to the simulation until a child node is reached. Based on the collaborative optimization mechanism of particle swarm optimization, genetic algorithm, and ant colony optimization, multiple incompletely explored child nodes are generated, and a complete simulation and random game are performed to obtain evaluation results. The results of the simulation and random game are fed back to the root node, and the visit count of each child node is updated to guide the next selection. In the simulation and random game, the particle swarm optimization, genetic algorithm, and ant colony optimization are invoked according to the Monte Carlo tree to update and optimize the neural network.

[0042] In this embodiment, step six specifically needs to be explained. Using a neural network, accurate nonlinear modeling of the PN emissions of hybrid electric vehicles can be performed to predict emission levels and support the decision-making process. Through the synergistic optimization of PSO, GA, and ACO, the weights and biases of the neural network can be effectively updated, improving the speed and accuracy of network training. Combined with real-time data acquisition, the neural network can achieve online learning, quickly adapt to environmental changes and operating conditions, and improve the real-time performance and accuracy of predictions. Through MCTS, the state space is explored and optimized to generate multiple exploration paths, ensuring that the optimal decision path is found, which helps in the exploration and discovery of local optima.

[0043] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm, characterized in that, Includes the following steps: Step 1: Connect the sensor components to the on-board data acquisition terminal of the hybrid vehicle to collect various data during the operation of the hybrid vehicle in real time, including vehicle speed, engine speed, battery status, air flow, fuel consumption rate and catalyst temperature. Step 2: Clean and standardize the real-time collected data to remove noise and outliers; Step 3: Introduce the particle swarm optimization algorithm to quickly locate the solution space region by using multiple particles to perform a global search in the search space. Step 4: Introduce a genetic algorithm for local search and further optimize candidate solutions through selection, crossover, and mutation; Step 5: Introduce the ant colony optimization algorithm as input for further optimization of candidate solutions, find the optimal solution and output it; Step 6: Output the PN emission prediction of the hybrid vehicle through the neural network, and update and optimize the neural network by calling the particle swarm optimization algorithm, genetic algorithm and ant colony optimization algorithm according to the Monte Carlo tree.

2. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: In step one, the vehicle data acquisition terminal includes a wheel speed sensor for real-time acquisition of vehicle speed data, an engine speed sensor for real-time acquisition of engine speed data, a battery management component for real-time acquisition of battery status data, an intake air flow meter for real-time acquisition of air flow data, a fuel flow sensor for real-time acquisition of fuel consumption rate data, and a temperature sensor for real-time acquisition of catalyst temperature data.

3. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: Step two, cleaning the real-time collected data, specifically includes identifying and deleting outliers, checking for duplicate values ​​and deleting them, and checking for missing values ​​and replacing them. The specific steps are as follows: Construct a tree structure, where branch nodes represent outlier features of the real-time collected data, and leaf nodes represent outlier classification results. Randomly select outlier features and real-time collected data to construct multiple decision trees and perform optimal splitting. Perform independent outlier classification prediction on each decision tree. Establish a voting mechanism to select the outlier classification results predicted by all decision trees. When more than 70% of the decision trees predict the current node as an outlier, directly determine that node as an outlier and delete it. Check and delete duplicate values ​​in the real-time collected data using unique identifiers. Check for missing values ​​in the real-time collected data. Replace missing values ​​with the mean of the real-time collected data using weighted mean imputation. Standardize the data using Z-score standardization.

4. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 3, characterized in that: The specific formula for the weighted mean interpolation method is as follows: ； in, This represents the weighted average. This represents the weight of the i-th real-time collected data. A descriptive symbol indicating whether or not to answer. Indicates "yes" Indicates "no". This indicates the amount of data collected in real time. The Z-score standardization formula is as follows: (This represents the i-th real-time data acquisition.) ； in, Represents the standard scores of a normal distribution. This indicates data collected in real time. This represents the average value of the data collected in real time. This represents the standard deviation of the data collected in real time.

5. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: In step three, the update rule of the particle swarm optimization algorithm includes updating the velocity and position of the particles, and its specific formula is as follows: ； in, This represents the velocity of particle e in the d-th dimension and at the (t+d)-th iteration. This represents the velocity of particle e in the d-th dimension and at the t-th iteration. The inertial weights represent the influence of a particle's current velocity on its new velocity. These represent the learning factors that indicate the degree to which a particle learns from its own historical best position and the learning factors that indicate the degree to which a particle learns from the group's best position, respectively. They represent random numbers between [0, 1]. This represents the historical best position of particle e in the d-th dimension and at the t-th iteration. This represents the current position of particle e in the d-th dimension and at the t-th iteration. This represents the update position of particle e in the d-th dimension and at the t-th iteration. This represents the globally optimal position of all particles in the d-th dimension at the t-th iteration.

6. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: In step four, the specific steps of the genetic algorithm are as follows: A population of candidate solutions is randomly generated. Each candidate solution is treated as an individual and assigned a genetic fitness value. Solutions are sorted according to their genetic fitness values, with priority given to those with higher values ​​for reproduction. The candidate solution with the highest genetic fitness value is selected as the parent solution. Parent solutions are repeatedly selected based on decreasing genetic fitness values. Single-point crossover is used to randomly select a location to cut off the parent solution as a crossover point. Genes are exchanged between the two parent solutions at the crossover point, generating two new offspring solutions. Multi-point crossover is used to randomly select multiple locations to cut off the parent solutions as crossover points. Genes are exchanged between the multiple parent solutions at the crossover points, generating multiple new offspring solutions. Uniform crossover is used to probabilistically select and determine the crossover operation for each gene locus. Crossover is performed on the parent solutions at the selected gene locus to generate new offspring solutions. Candidate solutions are further optimized using various genetic crossover methods. The genetic termination condition is defined as reaching a maximum of 20 iterations. Offspring solutions that meet the termination condition are returned as candidate solutions for further optimization.

7. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: In step five, the ant colony optimization algorithm calculates the probability of moving to the next node based on the pheromone concentration values ​​from the current node to other nodes and heuristic information using a proportional selection method. The specific formula is as follows: ； in, Let represent the probability of an ant choosing to go from node j to node k. This represents the pheromone concentration value between node j and node k. This represents the heuristic information between node j and node k. A moderating factor indicating the degree of influence of pheromones. A moderating factor representing the degree to which heuristic information has an impact. This represents a node in the set of neighboring nodes of the current node j. This represents the set of neighboring nodes of the current node j.

8. The method for predicting PN emissions of hybrid electric vehicles based on the PSO-GA-ACO hybrid optimization algorithm according to claim 1, characterized in that: In step six, the neural network performs forward propagation, and the specific formula is as follows: ； in This represents the input to the hidden layer of the neural network. This represents the weight matrix of the hidden layer of a neural network. A vector representation of the input data collected in real time. The specific formula for calculating the bias vector of a hidden layer in a neural network via backpropagation is as follows: ； in Represents the optimization parameters of the neural network. Indicates the current neural network parameters. Indicates the learning rate. This represents the gradient obtained by differentiating the loss function with respect to parameter w.

Images