A method and system for intelligent energy management of hybrid buses in an intelligent network environment

Through the vehicle-side cloud collaboration method, combined with cloud computing and edge computing, XGBoost and dynamic planning algorithms are used to realize the adaptive energy management of hybrid urban buses, solving the problems of working condition identification and vehicle speed planning, and improving fuel economy and driving safety.

CN116187161BActive Publication Date: 2024-11-01JIANGSU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211622738.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-12-16
Publication Date
2024-11-01
Estimated Expiration
2042-12-16

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively combine working conditions identification, vehicle speed planning and passenger number changes in hybrid urban buses to achieve a balance between fuel economy and driving safety, and there is a problem of separation between technical research on smart cars and new energy vehicles.

Method used

Using the method of vehicle-side cloud collaboration, data processing and working condition recognition model are trained through cloud computing, real-time working condition recognition is realized using XGBoost algorithm, and dynamic planning and particle swarm algorithm are combined with edge computing to design adaptive vehicle speed planning and Energy management strategy to maximize energy utilization and reduce fuel consumption.

Benefits of technology

It realizes optimal energy management under different working conditions and passenger capacity, improves the fuel economy of hybrid urban buses, provides adaptability and real-timeness, and enhances the economic and safety of vehicle speed planning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116187161B_ABST
    Figure CN116187161B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent energy management method and system for hybrid buses in an intelligent network environment, which integrates vehicle-side cloud collaboration, operating condition categories, vehicle speed planning, and changes in the number of passengers in urban buses. Urban buses upload data to the cloud during driving, and the cloud performs kinematic segment division, feature parameter extraction, PCA dimension reduction, and K-Means clustering on the data to generate typical driving condition data, and divides the training set and test set to train the XGBoost operating condition recognition model to achieve real-time discrimination of operating condition categories. Cloud computing delegates the recognition results to edge computing, which designs real-time vehicle speed planning based on a dynamic programming algorithm under different vehicle spacings, and uses a particle swarm algorithm to solve the optimal equivalent factor according to different operating condition categories and real-time passenger capacity. This strategy can switch the appropriate mode and the corresponding optimal energy management strategy according to different operating condition types, maximize the use of energy, reduce the fuel consumption of hybrid city buses, and improve fuel economy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of automobile intelligent assisted driving and energy management strategies, and in particular relates to an energy management strategy for a hybrid city bus taking into account operating condition identification, vehicle speed planning, and changes in the number of passengers. The strategy aims to improve the fuel economy of the entire vehicle while ensuring driving safety and riding comfort. Background Art

[0002] In recent years, problems such as traffic congestion, energy crisis, and environmental pollution have become increasingly serious. Under the background of smart cities, intelligent transportation, and intelligent automobiles, the integration of automobiles with electronics, communications, the Internet, and other fields has accelerated, and the information elements in road traffic have been shared. On this basis, the integration and utilization of intelligent transportation big data information has become a hot topic in current research. As a mode of travel that can effectively reduce traffic congestion and environmental pollution, urban public transportation has its own specific operating rules. It is very necessary to use existing data to mine its internal connections and formulate control strategies suitable for urban public transportation.

[0003] As vehicle cloud computing plays an increasingly important role in the fields of smart transportation and smart cars, it is possible to comprehensively consider the impact of road condition information on energy management. Cloud computing and edge computing have their own strengths. Cloud computing is good at analyzing and processing global, non-real-time, and long-term big data. Therefore, it can collect vehicle speed information, process GPS information, and perform dimensionality reduction clustering and other analyses in the cloud to achieve working condition identification. Edge computing (MEC) is more suitable for operations that require high frequency and real time. Therefore, vehicle speed planning and energy management strategy are quickly solved in edge computing. Cloud computing optimizes the output results through big data analysis and transfers them to the edge side. Edge computing quickly solves based on cloud results. Edge computing not only meets the real-time requirements, but also greatly reduces invalid data on the cloud, efficiently realizes vehicle speed planning, and returns the results to the vehicle side in real time.

[0004] In order to comprehensively solve the vehicle safety, energy conservation and environmental protection problems from the perspective of vehicle system control, and to break through the limitation that the new energy vehicle field and the smart car field are still conducting related technical research independently, a data-driven intelligent energy management method for hybrid city buses that integrates the advanced technologies of new energy vehicles and smart cars is proposed. Summary of the invention

[0005] Purpose of the invention: In view of the shortcomings in the prior art, the present invention provides a hybrid city bus intelligent energy management method, which integrates factors such as vehicle-side cloud collaboration, working condition category, vehicle speed planning, and changes in the number of passengers in a city bus to form a set of control strategies suitable for energy management of city buses. Under the condition of ensuring that the vehicle is driving normally, the advantages of the two power sources are fully utilized to achieve complementarity. The main purpose is that the city bus uploads relevant data to the cloud during driving, and cloud computing processes the quality problems of the data, and then performs kinematic segment division, feature parameter extraction, PCA dimension reduction, and K-Means clustering on the data. Finally, the typical driving condition data generated by clustering is used to divide the training set and the test set, and the working condition recognition model based on the XGBoost algorithm is trained to realize real-time identification of working condition categories. Cloud computing delegates the recognition results to edge computing, and the edge computing designs different vehicle spacings according to different working condition categories based on the results of cloud computing to realize real-time vehicle speed planning based on the dynamic programming algorithm. On the other hand, according to different working condition categories and real-time passenger capacity, the particle swarm algorithm is used to solve the optimal equivalent factor in the equivalent fuel consumption minimum energy management strategy, thereby realizing an adaptive equivalent fuel consumption minimum energy management strategy. This strategy can switch the appropriate mode according to different operating conditions and the core parameters of the corresponding optimal energy management strategy, maximize the use of energy, reduce the fuel consumption of hybrid city buses, and improve fuel economy.

[0006] An intelligent energy management method for a hybrid city bus in an intelligent network environment comprises the following steps:

[0007] Step 1: Upload road traffic information to the cloud, including real-time collection of the road traffic information of the vehicle and the driving status information of the vehicle. The road traffic information includes: longitude, latitude, slope, distance to the front vehicle, and speed of the front vehicle; the driving status information of the vehicle includes: vehicle speed, GPS information, SOC, and passenger capacity change of the vehicle;

[0008] Step 2: Cloud computing optimizes the output results through big data analysis and transfers them to the edge side: This includes processing the quality problems of the data in the cloud, and then performing kinematic segmentation, feature parameter extraction, PCA dimension reduction, and K-Means clustering on the data. Finally, the typical driving condition data generated by clustering is used to divide the training set and test set to train the working condition recognition model based on the XGBoost algorithm, thereby realizing real-time recognition of working condition categories.

[0009] Step 3: Edge computing quickly solves the cloud results and returns the results to the vehicle in real time: This includes cloud computing transferring the recognition results to edge computing. Edge computing designs different vehicle spacings according to different working conditions based on the results of cloud computing and implements real-time vehicle speed planning based on dynamic programming algorithms. On the other hand, according to different working conditions and real-time passenger capacity, the particle swarm algorithm is used to solve the core control parameters in the energy management strategy to implement an adaptive energy management strategy.

[0010] Furthermore, in the step 2, cloud computing optimizes the output results through big data analysis and transfers them to the edge side, including the following processes: first, filtering is performed on duplicate data, invalid data, lost data, and impulse noise and high-frequency noise caused by improper operation of the driver, and map matching and station identification are performed using GPS information to achieve synchronous matching of vehicle speed and passenger information. Then the data is divided according to the definition of motion segments, wherein the definition of motion segments is the operating interval from one idle to the next idle. Since the driving condition of urban buses is composed of a large number of motion segments, each motion segment needs to be calculated according to the selected characteristic parameters when performing the operating condition analysis. Too many characteristic parameters will lead to a significant increase in the amount of calculation, thereby increasing the difficulty of calculation. Too few characteristic parameters may not accurately reflect the running state of the car. In summary, 12 characteristic parameters are selected to comprehensively reflect the characteristics of urban bus motion segments, specifically including acceleration time ratio, deceleration time ratio, idle time ratio, uniform speed time ratio, maximum vehicle speed, average vehicle speed, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration, and average deceleration. Since the routes of urban buses have certain regularity in actual driving, under the influence of the same road conditions, traffic conditions, driving time periods and other factors, their kinematic segments will show certain similarities. Therefore, PCA and K-Means algorithms are used to classify kinematic segments with similar characteristics into one category, and the subsequent working condition identification and energy management strategy design are completed based on the clustering results. PCA uses interpretable variance as a measure of information volume, and the dimension reduction results are further used as the input of the clustering algorithm. The purpose is to divide the data into meaningful clusters. The process of the clustering algorithm is mainly to first randomly select K samples as the initial centroid, and then start the cycle: assign each sample point to the nearest centroid to generate K clusters. For each cluster, the average value of all sample points assigned to the cluster is calculated as the new centroid. When the position of the centroid no longer changes, the iteration stops. After clustering is completed, the silhouette coefficient is introduced as an evaluation indicator for the clustering results of the K-means algorithm. Finally, cluster analysis was performed based on actual bus route data to obtain typical driving conditions of different categories. Based on the original data, samples were divided into 120s as the sample duration, and the divided samples were divided into training sets and test sets. The XGBoost model was then trained, and the Scikit-learn library was used to select model hyperparameters with the goal of selecting the optimal loss function. The effectiveness of the model was evaluated by three indicators: precision, recall rate, and F1score. Finally, the recognition cycle was set to 120s and the update cycle to 20s to perform online recognition of the actual vehicle speed.

[0011] Furthermore, in step 3, the edge computing is used to quickly solve the cloud results and return the results to the vehicle in real time, including the following process: cloud computing transfers the real-time working condition category identification results to the edge computing, and the edge computing designs the corresponding safe time distance between the city bus and the preceding vehicle according to different working condition categories based on the results of cloud computing, and realizes the speed planning of the safe time distance under different working condition types based on the dynamic programming algorithm. In addition, since the city bus operation route is long and the road conditions involved in driving are complex, the working condition category and the number of passengers in the city bus have a great influence on the design of the energy management strategy of the city bus. Therefore, it is necessary to solve the core control parameters of the equivalent fuel consumption minimization strategy (ECMS) corresponding to different categories of typical working conditions and different passenger capacities. Among them, the equivalent factor has a great influence on the control effect of the ECMS energy management strategy. Therefore, the particle swarm optimization algorithm is used to design the corresponding fitness function to solve the optimal equivalent factor for different working conditions and passenger capacities.

[0012] Beneficial effects of the present invention:

[0013] 1. The present invention solves the optimal control parameters and cruising speed corresponding to different operating conditions and changes in passenger capacity, and applies them to the energy management strategy of hybrid city buses.

[0014] 2. The present invention is designed to collect vehicle speed information, process GPS information, and perform dimensionality reduction clustering and other analyses in the cloud to achieve working condition identification. Edge computing MEC (Mobile Edge Computing) is more suitable for operations that require high frequency and real-time. Therefore, vehicle speed planning and energy management strategies are quickly solved in edge computing.

[0015] 3. The present invention uses a dynamic programming optimization algorithm to design different safety time limits according to the dynamics of the preceding vehicle and the working conditions of the road section to obtain the final planned vehicle speed. This strategy can deeply mine the operating rules of public transportation data to design a speed planning method suitable for urban public transportation, while improving the economy of the planned vehicle speed and the adaptability of speed planning for different road sections. In addition, the speed planning method provides a new idea, which can provide the driver with the most reasonable driving method reference in a manner similar to an assisted driving system, and can also provide a driving decision basis for higher-level autonomous driving technology in an intelligent network environment, so that it can be applied in more occasions. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the overall strategy of the present invention;

[0017] Figure 2 The ECMS energy management strategy process based on the operating conditions and passenger volume changes of the present invention;

[0018] Figure 3 The vehicle speed planning strategy process of the present invention;

[0019] Figure 4 The rolling optimization strategy of the present invention;

[0020] Figure 5 Dynamically plan the process for the present invention; DETAILED DESCRIPTION

[0021] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0022] like Figure 1 As shown in the figure, an intelligent energy management system for hybrid buses in an intelligent network environment includes cloud devices and edge devices. Through the vehicle-mounted global positioning system (GPS), geographic information system (GIS) and intelligent transportation system (ITS) as well as vehicle speed sensors and laser radars, road traffic information is obtained, including slope, distance to the front vehicle, front vehicle speed and driving status information of the vehicle, including: vehicle speed, vehicle SOC, and passenger volume change of the vehicle. Then, cluster analysis is performed through data preprocessing, feature parameter calculation, PCA dimension reduction, and K-Means algorithm. The clustered typical working condition 0, typical working condition 1, typical working condition 2 and other working condition categories that can represent urban working conditions, suburban working conditions, and high-speed working conditions are used for the training of the working condition recognition model, and finally the working condition online recognition on the cloud is realized. The cloud device transfers the recognition results to the edge device. The edge device designs different vehicle spacings according to different working condition categories based on the results calculated by the cloud device, and realizes real-time vehicle speed planning based on the dynamic programming algorithm to complete the calculation of strategy one. Strategy 2 selects the optimal equivalent factor based on typical operating condition 0, typical operating condition 1, typical operating condition 2 and other operating condition segments that can represent different operating condition categories such as urban areas, suburbs, and highways, as well as the load levels of light load, medium load, and heavy load corresponding to different passenger capacities. Finally, the optimal equivalent factor and vehicle speed at the next moment are returned to the vehicle end to implement an adaptive energy efficiency and minimum fuel consumption energy management strategy.

[0023] The present invention relates to an adaptive cruise energy management method for a hybrid city bus in an intelligent network environment, which mainly comprises the following steps:

[0024] Step 1: Upload road traffic information to the cloud, including real-time collection of the road traffic information of the vehicle and the driving status information of the vehicle. The road traffic information includes: longitude, latitude, slope, distance to the front vehicle, and speed of the front vehicle; the driving status information of the vehicle includes: vehicle speed, GPS information, SOC, and passenger capacity change of the vehicle;

[0025] Step 2: Cloud computing optimizes the output results through big data analysis and transfers them to the edge side: This includes processing the quality problems of the data in the cloud, and then performing kinematic segmentation, feature parameter extraction, PCA dimension reduction, and K-Means clustering on the data. Finally, the typical driving condition data generated by clustering is used to divide the training set and test set to train the working condition recognition model based on the XGBoost algorithm, thereby realizing real-time recognition of working condition categories.

[0026] Step 3: Edge computing quickly solves the cloud results and returns the results to the vehicle in real time: This includes cloud computing transferring the recognition results to edge computing. Edge computing designs different vehicle spacings according to different working conditions based on the results of cloud computing and implements real-time vehicle speed planning based on dynamic programming algorithms. On the other hand, according to different working conditions and real-time passenger capacity, the particle swarm algorithm is used to solve the core control parameters in the energy management strategy to implement an adaptive energy management strategy.

[0027] Furthermore, in the step 2, cloud computing optimizes the output results through big data analysis and transfers them to the edge side, including the following processes: first, filtering the duplicate data, invalid data, lost data, and impulse noise and high-frequency noise caused by improper operation of the driver, and using GPS information for map matching to identify the site to achieve synchronous matching of vehicle speed and passenger information. Then the data is divided according to the definition of motion segments, wherein the definition of motion segments is the operating interval from one idle to the next idle. Since the driving condition of urban buses is composed of a large number of motion segments, it is necessary to calculate each motion segment according to the selected characteristic parameters when performing the operating condition analysis. Too many characteristic parameters will lead to a significant increase in the amount of calculation, thereby increasing the difficulty of calculation. Too few characteristic parameters may not accurately reflect the running state of the car. In summary, 12 characteristic parameters are selected to comprehensively reflect the characteristics of urban bus motion segments, including acceleration time ratio, deceleration time ratio, idle time ratio, uniform speed time ratio, maximum vehicle speed, average vehicle speed, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration, and average deceleration. Since the routes of urban buses have certain regularity in actual driving, under the influence of the same road conditions, traffic conditions and driving time periods, their kinematic segments will show certain similarities. Therefore, PCA and K-Means algorithms are used to classify kinematic segments with similar characteristics into one category. The urban working conditions generally have the characteristics of high acceleration / deceleration time ratio, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration and average deceleration, and low uniform speed time ratio, maximum vehicle speed and average vehicle speed. The suburban working conditions generally have the characteristics of high acceleration / deceleration time ratio, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration and average deceleration, and low uniform speed time ratio, maximum vehicle speed and average vehicle speed. The deceleration time ratio, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration, and average deceleration are relatively low, while the uniform speed time ratio, maximum vehicle speed, and average vehicle speed are relatively high. High-speed working conditions generally have the characteristics of acceleration / deceleration time ratio, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration, and average deceleration are low, while the uniform speed time ratio, maximum vehicle speed, and average vehicle speed are high. Finally, the appropriate clustering category is selected according to the actual road conditions. After obtaining the clustering results, the appropriate control strategy can be selected for different working condition categories through working condition identification.The ultimate goal of PCA in the dimensionality reduction process is to reduce the number of features while retaining most of the effective information, creating a new feature matrix that can represent most of the information of the original feature matrix. PCA uses interpretable variance as a measure of the amount of information, and the dimensionality reduction result is further used as the input of the clustering algorithm. The purpose is to divide the data into meaningful clusters. The process of the clustering algorithm is mainly to first randomly select K samples as the initial centroid, and then start a cycle: assign each sample point to the centroid closest to it to generate K clusters. For each cluster, calculate the average value of all sample points assigned to the cluster as the new centroid. When the position of the centroid no longer changes, the iteration stops clustering and the silhouette coefficient is introduced as an evaluation indicator for the clustering results of the K-means algorithm. Finally, cluster analysis was performed based on the actual bus route data to obtain typical driving conditions. Based on the original data, the samples were divided into 120s as the sample duration, and the divided samples were divided into training set and test set, and then the XGBoost model was trained. XGBoost (eXtreme Gradient Boosting) is a machine learning method based on the ensemble learning principle Boosting. The specific implementation of XGBoost is: construct multiple weak learners, where the role of the first base learner is to fit the sample data, and the fitting error is used as the target of the second base learner to continue fitting. All base learners are accumulated as the final result through a residual approximation method similar to statistics.

[0028]

[0029] Among them, the feature of the i-th sample is x i , and the prediction value of the kth tree is f k (x i ), is the predicted value of the model. The optimization process starts from the first tree and optimizes them sequentially until all trees are optimized.

[0030] XGBoost objective function obj k As shown in the following formula, it consists of a loss function and a regularization term, where the role of the loss function is to balance the training error, and the purpose of the regularization term is to control the complexity.

[0031]

[0032] In the formula is the loss function, n represents the sample overview, Ω(f k ) represents the complexity of the kth tree.

[0033] In summary, XGBoost is modeled based on the objective function, the measurement indicator is the minimization of the objective function, and the algorithm strategy is the greedy strategy. The advantage of XGBoost is that the regular term is added to the objective function, which can ensure that the model complexity is hedged at each iteration, which can reduce the occurrence of overfitting, and use the Scikit-learn library to select the model hyperparameters with the goal of selecting the optimal loss function. Finally, the effectiveness of the model is evaluated by three indicators: precision, recall rate, and F1score. Finally, the recognition cycle is set to 120s and the update cycle is set to 20s to perform online recognition of the actual vehicle speed.

[0034] Furthermore, the edge computing in step three quickly solves the cloud results and returns the results to the vehicle in real time, including the following process: cloud computing transfers the real-time working condition category identification results to the edge computing. For strategy one, since the urban bus operation route is long and the driving involves complex road conditions, the working condition category and the number of passengers in the urban bus have a great influence on the design of the energy management strategy of the urban bus. Therefore, it is necessary to solve the core control parameters of the equivalent fuel consumption minimization strategy (ECMS) corresponding to different categories of typical working conditions and different passenger capacities, wherein the vehicle load is set to three levels of light load, medium load and heavy load according to 0-30%, 30%-70% and 70%-100% of the maximum passenger capacity, respectively. Since the equivalent factor has a great influence on the control effect of the ECMS energy management strategy, the particle swarm optimization algorithm is used to design the corresponding fitness function to solve the optimal equivalent factor for different working condition categories such as urban areas, suburbs, highways and different load levels such as light load, medium load and heavy load corresponding to typical working conditions 0, typical working conditions 1 and typical working conditions 2. The specific process is as follows Figure 2 As shown in the figure, the equivalent fuel consumption minimum strategy can be regarded as a simplification of the energy management strategy based on Pontryagin's minimum principle (PMP) in the instantaneous optimization process. The basic idea of ​​the equivalent fuel consumption theory is that the electric energy out of the battery power balance point will come from the fuel. Therefore, when performing optimization calculations, the electric energy currently planned to be consumed (or recovered) can be converted into the fuel consumed (or saved) in the future to make up for (or consume) this electric energy. The conversion coefficient is called the equivalent factor, that is, at a certain moment, according to the current driving conditions of the vehicle, traversing different engine and motor torque combinations, there will always be a point with the minimum total energy consumption. This point is the optimal solution under the current conditions, corresponding to the optimal distribution relationship between fuel and electric energy at this moment:

[0035]

[0036]

[0037]

[0038] in is the instantaneous total equivalent fuel consumption, f pen is the penalty coefficient related to power consumption, is the instantaneous equivalent fuel consumption related to the motor, is the instantaneous fuel consumption, p el is the motor power, η el is the motor efficiency, sign(P el ) is a sign function related to the positive and negative power of the motor, where when the motor power is positive, sign(P el ) takes 1; when the motor power is zero, sign(P el ) takes 0; when the motor power is negative, sign(P el ) takes -1; γ(P el ) is the correlation coefficient with the positive and negative power of the motor, which is H LHV is the lower heating value of the fuel, and s(t) is the equivalent factor. Therefore, the value of different equivalent factors will affect the equivalent fuel consumption, which will ultimately affect the SOC and the working state of the engine and motor. Therefore, the particle swarm optimization algorithm is used to design the fitness function to solve the optimal equivalent factor for different working conditions and passenger capacity. In the process of finding the optimal solution by the particle swarm optimization algorithm (Particle Swarm Optimization PSO), each optimal solution can be regarded as a bird, also called a "particle", and each particle has a certain fitness value. When there are n particles in the d-dimensional space; the position of the particle: X i =(x i1 ,x i2 ,…x id ), X i Substitute the fitness function f(X i ); velocity of particle i: V i =(V i1 ,V i2 ,…V id ); the best position experienced by particle i: pbest i =(p i1 ,p i2 ,…p id ); the best position that the population has ever experienced: gbest = (g1, g2, ... g d ); the update method of each particle position and velocity is as follows:

[0039]

[0040] Where ω is the inertia weight factor; c1c2 is the learning factor; rand is the random function; x i(t) is the position of the particle at the current moment; v i (t) is the velocity of the particle at the current moment; p gd is the optimal position of the particle group, p id is the optimal position of the individual particle. The fitness function corresponding to the ECMS control strategy is shown in the formula

[0041]

[0042] Among them J e is the fuel consumption, where SOC (State of Charge) is the battery charge state, also known as the remaining power, SOC ref For reference remaining power value, SOC end is the final remaining power value after the trip, and α is the weight coefficient considering the change of remaining power. The working condition identification algorithm is combined with the control strategy that adapts to different working conditions and changes in the number of people, and an energy management strategy based on XGBoost working condition identification is established. Under the premise of ensuring the power requirements of hybrid buses, the fluctuation of SOC can be reduced, and the fuel saving potential of urban buses can be further improved.

[0043] For strategy 2, edge computing designs the corresponding safe time distance between city buses and the preceding vehicle based on the results of cloud computing according to different working conditions, and implements the speed planning of safe time distance under different working conditions based on dynamic programming algorithm. The specific process is as follows: Figure 3 As shown in the figure, in the intelligent network environment, based on the provided urban bus geographic information and traffic information, a research framework for intelligent urban bus speed planning considering safe headway under different working conditions is proposed. First, the road scene is constructed. The present invention mainly studies the speed planning problem of urban buses in the vehicle networking environment. The limiting conditions for constructing the road scene are simplified and assumed as follows:

[0044] (1) Every vehicle on the road is equipped with communication equipment that enables vehicle-to-vehicle communication, vehicle-to-traffic facilities, and the cloud.

[0045] (2) Only the impact of the status change of the preceding vehicle and the speed limit information on the vehicle is considered, and other factors are not considered for the time being.

[0046] (3) The interference from other factors such as surrounding pedestrians and non-motor vehicles is not considered.

[0047] (4) Situations such as lane changing and overtaking are not considered.

[0048] (5) This paper assumes that each bus stop section is covered by a MEC server, and the speed planning is calculated by it and fed back to the vehicle end of this section in real time.

[0049] The architecture mainly includes cloud computing to transfer the recognition results to edge computing. Edge computing designs different vehicle spacings according to different working conditions based on the results of cloud computing. Real-time vehicle speed planning is achieved based on dynamic programming algorithms. The specific process of rolling optimization is as follows: Figure 4 As shown in the figure, the solution of the planned vehicle speed includes first dividing the global problem into stages in the distance domain according to a time interval of 20 seconds, the update time is 5 seconds, the total distance is D, the number of intervals is n, and the distance of each interval is D i , the subscript i represents the i-th interval (i = 1, 2, 3 ... n), and the speed is optimized by dynamic programming in each stage, and the discrete distance interval of each stage is defined as s k , then the i-th stage problem can be solved according to s k Discretized into N i Step (k=1,2,3……N i ), the specific process of dynamic planning is as follows Figure 5 As shown, first the optimization problem is discretized in the distance domain, according to s k Discretized into N steps k∈(1,N), where the control variable is acceleration a k , the state variable is v k Represents the corresponding vehicle speed, energy consumption is E, v lim,k is the maximum speed limit of the corresponding road section, t k represents the duration from the start, l time is the penalty coefficient related to time. The time and speed of the next moment can be calculated by the time and speed of the current moment:

[0050]

[0051]

[0052] The acceleration is a k , v k represents the corresponding vehicle speed, t k Represents the duration from the start, s k is the discrete distance.

[0053] According to the longitudinal dynamics model,

[0054] F wheel =(ma+F f +F w +F i )

[0055] Among them, F wheel represents the longitudinal force, m represents the mass of the vehicle; a represents the acceleration; F f Rolling resistance, F w Air resistance, F i Ramp resistance, as shown below

[0056]

[0057] Where g represents the acceleration of gravity; f represents the rolling resistance coefficient; θ represents the road slope; ρ represents the air density; C D represents the air resistance coefficient; A represents the frontal area; v represents the longitudinal speed.

[0058] Instantaneous cost E c (k) is the energy consumed during driving, and the calculation formula is:

[0059] E c (k) = F wheel (k)(s k+1 -s k )η tr

[0060] Among them, F wheel represents the longitudinal force, η tr Indicates efficiency, s k For discrete distance intervals, the time limit calculation process is:

[0061]

[0062] t k,max =t k,min +t end -t N,mmin

[0063] where t k,max is the upper limit of the time for the kth step, t k,min is the lower limit of the time for the kth step, v k,max is the maximum speed that can be achieved without exceeding the maximum acceleration limit, t N,min is the shortest time from the beginning to the Nth step, t end is the maximum time required to complete the route based on historical data and actual conditions. The cost function of each step is defined as the sum of the current energy consumption and the previous energy consumption. In order to minimize the cost of the entire driving process, the cost of each discrete stage needs to be superimposed to minimize the total cost. Therefore, the objective function and related constraints can be expressed as:

[0064]

[0065] Constraint equations

[0066]

[0067] The control variable is acceleration a k , v k Represents the corresponding vehicle speed, v lim,k is the current speed limit, sk is the discrete distance interval, t k represents the time from the beginning to the kth step, E(v k , a k ) is the instantaneous energy consumption, which can be calculated based on E c (k) is calculated, f(v k , a k ,s k )represent Part, used to calculate the specific value of the next long speed, f(v k , t k ,s k )represent Part, used to calculate the moment of the next step, l time It is a penalty factor to limit the completion of the station interval distance within a certain time.

[0068] The two following distance speed planning methods are ACC max_min and ACC min , according to the best safety time limit for different working conditions, when the working condition type is congestion type, it is ACC max_min When the vehicle is moving in a straight line, MEC will use the dynamic planning rolling optimization to obtain the optimized speed, and then calculate the current vehicle spacing based on the feedback of the vehicle in front. If the current vehicle spacing is between the maximum and minimum vehicle spacing, the speed optimized by the dynamic planning of the current road section will be used as the planned speed. If it exceeds this range, the vehicle speed needs to be replanned, and the vehicle spacing with the vehicle in front will be adjusted to the specified range within a certain period of time. When the working condition type belongs to a smooth road section or a suburban type, it is ACC. min When the distance between vehicles is greater than the minimum distance between vehicles, MEC obtains the optimized speed based on dynamic planning rolling optimization, and then calculates the current distance between vehicles based on the feedback of the front vehicle dynamics. If the current distance between vehicles is greater than the maximum distance between vehicles, the speed optimized by dynamic planning of the current road section is used as the planned speed. If it is less than the minimum distance between vehicles, the speed needs to be replanned, and the distance between the vehicle and the front vehicle is adjusted to the specified range within a certain period of time. The distance between the vehicle and the front vehicle at time k is defined as s. diff,k The specific calculation method can be to convert the known time-speed curve of the preceding vehicle into a time-mileage curve through coordinate transformation, and then calculate the travel time t corresponding to the kth step. k Interpolate to get the mileage s of the current front vehicle leader , define the distance between the vehicle and the preceding vehicle at time k as:

[0069]

[0070] The minimum distance between vehicles is:

[0071]

[0072] The maximum distance between vehicles is:

[0073]

[0074] Where S0 is the safety distance, v k is the current vehicle speed, s k is the discrete distance, v leader is the speed of the front vehicle, T is the reaction time, a min is the minimum acceleration, a max is the maximum acceleration. That is, edge computing MEC (Mobile Edge Computing) applies dynamic programming to divide the speed planning problem into a multi-stage optimization problem through discrete distance. The final speed of the previous stage is the initial speed of this stage. Then, according to the initial speed of each stage, all possible final speeds are exhausted. The optimal speed is obtained by solving the minimum energy consumption of the objective function. After obtaining the dynamic planning of the i-th stage, the initial optimized speed is obtained. Then, MEC judges whether the current safety time distance limit is met through the real-time speed and position information of the current front vehicle and the vehicle and the working condition type obtained by cloud analysis and processing. That is, if the section is congested, the maximum and minimum safety time distances must be met at the same time to adjust the initial optimized speed. If it is a smooth urban road section or suburb, only the minimum vehicle distance is considered to adjust the initial speed. When the optimized speed makes the distance to the front vehicle not within the corresponding constraint range, the speed is adjusted to obtain the final planned speed, where f s_max is the difference between the current inter-vehicle distance and the maximum inter-vehicle distance constrained, f s_min The difference between the distance between the front vehicle and the minimum distance between vehicles

[0075] When diff,k >S max,k

[0076] f s_max =s diff,k -S max,k

[0077] When diff,k <S min,k

[0078] f s_min =s diff,k -S min,k

[0079] Get the new speed value at time k so that the following distance is within the constraint range. The specific method is to use the fzero function provided in MATLAB to solve the control variable a at the previous moment. k-1 and the state variable v k-1 and the current acceleration a k , speed vk 、The current distance between vehicles s diff,k The difference f from the constraint following distance s The root of the nonlinear equation of a single variable is obtained to obtain a new vehicle speed that meets the following distance constraint, so as to maximize the dynamic programming optimization effect while ensuring safety and road traffic efficiency.

[0080] The series of detailed descriptions listed above are only specific descriptions of feasible implementation methods of the present invention. They are not intended to limit the scope of protection of the present invention. All equivalent methods or changes that do not deviate from the technical creation of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent energy management method for hybrid buses in an intelligent network environment, characterized in that: The steps include: S1. Upload road traffic information to the cloud, including real-time collection of the vehicle's road traffic information and the vehicle's driving status information; S2. Cloud computing: Optimize the output results through big data analysis and transfer them to the edge side: including processing the quality problems of the data in the cloud, and then dividing the data into kinematic segments, extracting feature parameters, PCA dimension reduction and K-Means clustering calculations. Finally, the typical driving condition data generated by clustering is used to divide the training set and test set, and the working condition recognition model based on the XGBoost algorithm is trained to realize real-time recognition of working condition categories. S3, the edge side quickly solves the results based on the cloud and returns the results to the vehicle side in real time: including cloud computing to send the recognition results to the edge computing, edge computing based on the cloud computing results, designing different vehicle spacing according to different working conditions, and realizing real-time vehicle speed planning based on the dynamic programming algorithm. On the other hand, according to different working conditions and real-time passenger capacity, the particle swarm algorithm is used to solve the core control parameters in the energy management strategy to realize adaptive energy management strategy; The specific design process of the particle swarm optimization algorithm of S3 includes: The equivalent fuel consumption minimum strategy ECMS is regarded as a simplification of the energy management strategy based on Pontryagin's minimum principle (PMP) in the instantaneous optimization process. The basic idea of ​​the equivalent fuel consumption theory is that the electric energy out of the battery power balance point will come from the fuel. Therefore, when performing the optimization calculation, the electric energy currently planned to be consumed is converted into the fuel consumed in the future to make up for the electric energy. The conversion coefficient is called the equivalent factor, that is, at a certain moment, according to the current driving conditions of the vehicle, traversing different combinations of engine and motor torque, there will always be a point with the minimum total energy consumption. This point is the optimal solution under the current conditions, corresponding to the optimal distribution relationship between fuel and electric energy at this moment: in is the instantaneous total equivalent fuel consumption, f pen is the penalty coefficient related to power consumption, is the instantaneous equivalent fuel consumption related to the motor, p el is the motor power, η el is the motor efficiency, H LHV is the lower heating value of the fuel, s(t) is the equivalent factor; γ is the correlation coefficient with the positive and negative power of the motor; is the instantaneous fuel consumption; The fitness function is designed by using the particle swarm optimization algorithm to solve the optimal equivalent factor for different working conditions and passenger capacity. In the process of finding the optimal solution by the particle swarm optimization algorithm (PSO), each optimal solution is regarded as a bird, also called a "particle", and each particle has a certain fitness value; when there are n particles in the d-dimensional space; the position of the particle: X i =(X i1 ,X i2 ,…X id ), X i Substitute the fitness function f(X i ); velocity of particle i: V i =(V i1 ,V i2 ,…V id ); the best position experienced by particle i: pbest i =(p i1 ,p i2 ,…p id ); the best position that the population has ever experienced: gbest = (g1, g2, ... g d ); the update method of each particle position and velocity is as follows: Where ω is the inertia weight factor; c1 c2 is the learning factor; rand is the random function, p gd is the optimal position of the particle group, x i (t) is the position of the particle at the current moment; v i (t) is the velocity of the particle at the current moment; the fitness function corresponding to the ECMS control strategy is shown in the formula Among them J e is the fuel consumption, SOC ref For reference SOC, SOC end is the final SOC of the trip, γ is the weight coefficient considering the SOC change; In S3, the specific process of edge computing designing different vehicle spacings according to different working conditions based on the results of cloud computing and implementing real-time vehicle speed planning based on the dynamic programming algorithm includes: First, the global problem is divided into stages in the distance domain according to a time interval of 20 seconds. The update time is 5 seconds, the total distance is D, the number of intervals is n, and the distance of each interval is D i , the subscript i represents the i-th interval (i = 1, 2, 3 ... n), and the speed is optimized by dynamic programming in each stage, and the discrete distance interval of each stage is defined as s k , then the i-th stage problem can be solved according to s k Discretized into N i Step (k=1,2,3……N i ), the dynamic programming process is as follows: First, the optimization problem is discretized in the distance domain, according to s k Discretized into N steps k∈(1,N), where the control variable is acceleration a k , the state variable is v k Represents the corresponding vehicle speed, energy consumption is E, v lim,k is the maximum speed limit of the corresponding road section, t k represents the duration from the start, l time is the penalty coefficient related to time. The time and speed of the next moment can be calculated by the time and speed of the current moment: The acceleration is a k , shape k represents the corresponding vehicle speed, t k Represents the duration from the start, s k is the discrete distance; According to the longitudinal dynamics model, F wheel =(ma+F f +F w +F i ) Where m represents the mass of the vehicle; a represents the acceleration; F f Represents rolling resistance, F w Represents air resistance, F i The representative ramp resistance is as follows Where g represents the acceleration of gravity; f represents the rolling resistance coefficient; θ represents the road slope; ρ represents the air density; C D represents the air resistance coefficient; A represents the frontal area; v represents the longitudinal speed; Instantaneous cost E c (k) is the energy consumed during driving, and the calculation formula is: E c (k)=F wheel (k)(s) k+1 -s k )or tr Among them, F wheel represents the longitudinal force, η tr Indicates efficiency, s k For discrete distance intervals, the time limit calculation process is: t k,max =t k,min +t end -t N,min where t k,max is the time limit, t k,min is the lower limit of time, v k,max is the maximum speed that can be achieved without exceeding the maximum acceleration limit, t end To determine the maximum time required to complete the route based on historical data and actual conditions, the cost function of each step is defined as the sum of the current energy consumption and the previous energy consumption. In order to minimize the cost of the entire driving process, the cost of each discrete stage needs to be superimposed to minimize the total cost. Therefore, the objective function and related constraints can be expressed as: Constraint equations The control variable is acceleration a k , v k Represents the corresponding vehicle speed, v lim,k is the current speed limit, t k Represents the duration from the start, and the energy consumption is E, l time The penalty coefficient for completing the station interval within a certain time; The two following distance speed planning methods are ACC max_min and ACC min , according to the best safety time limit for different working conditions, when the working condition type is congestion type, it is ACC max_min When the vehicle is in the forward direction, the edge computing MEC obtains the optimized speed based on the dynamic planning rolling optimization, and then calculates the current vehicle spacing based on the feedback of the preceding vehicle dynamics. If the current vehicle spacing is between the maximum vehicle spacing and the minimum vehicle spacing, the speed optimized by the dynamic planning of the current road section is used as the planned speed. If it exceeds this range, the vehicle speed needs to be replanned, and the vehicle spacing with the preceding vehicle is adjusted to the specified range within a certain period of time. When the working condition type belongs to a smooth road section or a suburban type, it is ACC. min When the distance between vehicles is greater than the minimum distance between vehicles, the edge computing MEC obtains the optimized speed based on the dynamic planning rolling optimization, and then calculates the current distance between vehicles based on the feedback of the front vehicle dynamics. If the current distance between vehicles is greater than the maximum distance between vehicles, the speed optimized by the dynamic planning of the current road section is used as the planned speed. If it is less than the minimum distance between vehicles, the speed needs to be re-planned, and the distance between the vehicle and the front vehicle is adjusted to the specified range within a certain period of time. The distance between the vehicle and the front vehicle at time k is defined as s diff,k The specific calculation method can be to convert the known time-speed curve of the preceding vehicle into a time-mileage curve through coordinate transformation, and then calculate the travel time t corresponding to the kth step. k Interpolate to get the mileage s of the current front vehicle leader , define the distance between the vehicle and the preceding vehicle at time k as: The minimum distance between vehicles is: The maximum distance between vehicles is: Where S0 is the safety distance, v k is the current vehicle speed, v leader is the speed of the front vehicle, T is the reaction time, a min is the minimum acceleration, a max is the maximum value of acceleration, that is, the edge computing MEC applies dynamic programming to divide the speed planning problem into a multi-stage optimization problem through discrete distance. The final speed of the previous stage is the initial speed of this stage. Then, according to the initial speed of each stage, all possible final speeds are exhausted. The optimal speed is obtained by solving the minimum energy consumption of the objective function. After obtaining the dynamic planning of the i-th stage, the initial optimized speed is obtained. Then, the edge computing MEC judges whether the current safety time distance limit is met through the real-time speed and position information of the current front vehicle and the vehicle and the working condition type obtained by cloud analysis and processing. That is, if the section is a congested road condition, the maximum and minimum safety time distances must be met at the same time to adjust the initial optimized speed. If it is a smooth urban road section or suburb, only the minimum vehicle spacing is considered to adjust the initial speed. When the optimized speed makes the distance to the front vehicle not within the corresponding constraint range, the speed is adjusted to obtain the final planned speed, where f s_max is the difference between the current inter-vehicle distance and the maximum inter-vehicle distance constrained, f s_min is the difference between the front vehicle spacing and the constrained minimum vehicle spacing, When diff,k >S max,k f s_max =s diff,k -S max,k When diff,k min,k ​ f s_min =s diff,k -S min,k 。 2. According to the method for intelligent energy management of hybrid buses in an intelligent network environment of claim 1, it is characterized in that: In S1, the road traffic information includes: longitude, latitude, slope, distance to the front vehicle, and speed of the front vehicle; the driving status information of the own vehicle includes: speed of the own vehicle, GPS information of the own vehicle, SOC of the own vehicle, and change in passenger capacity of the own vehicle.

3. The intelligent energy management method for hybrid buses in an intelligent network environment according to claim 1 is characterized in that: The specific process of S2 includes: S2.1, filtering out duplicate data, invalid data, lost data, impulse noise and high-frequency noise caused by improper driver operation, and using GPS information for map matching and station identification to achieve synchronous matching of vehicle speed and passenger information; S2.2, divide the data according to the definition of motion segments, where the definition of motion segments is the operating condition interval from one idle to the next idle. The urban bus driving condition is composed of a large number of motion segments. Therefore, when performing the working condition analysis, it is necessary to calculate each motion segment according to the selected characteristic parameters. 12 characteristic parameters are selected to comprehensively reflect the characteristics of the urban bus motion segments. The specific parameters are: acceleration time ratio, deceleration time ratio, idling time ratio, uniform speed time ratio, maximum vehicle speed, average vehicle speed, vehicle speed standard deviation, maximum acceleration, maximum deceleration, acceleration standard deviation, average acceleration, and average deceleration. PCA and K-Means algorithms are used to classify kinematic segments with similar characteristics into one category, and the subsequent working condition identification and energy management strategy design are completed based on the clustering results; In S2.2, PCA uses interpretable variance as a measure of information volume, and the dimension reduction result is used as the input of the clustering algorithm, the purpose of which is to divide the data into meaningful clusters. The process of the clustering algorithm is mainly to first randomly select K samples as the initial centroids, and then start a cycle: assign each sample point to the centroid closest to it, generate K clusters, for each cluster, calculate the average value of all sample points assigned to the cluster as the new centroid, when the position of the centroid no longer changes, the iteration stops, and after clustering is completed, the silhouette coefficient is introduced as an evaluation indicator of the clustering result of the K-means algorithm; S2.3, cluster analysis is performed based on actual bus route data to obtain typical driving conditions of different categories, and samples are divided into 120s sample duration based on the original data, and the divided samples are divided into training set and test set, and then the XGBoost model is trained; In S2.3, the XGBoost model is modeled based on the objective function, the measurement indicator is the minimization of the objective function, the algorithm strategy is the greedy strategy, XGBoost adds a regular term to the objective function, and uses the Scikit-learn library to select the model hyperparameters with the goal of selecting the optimal loss function, and evaluates the effectiveness of the model through the three indicators of precision, recall rate and F1score; S2.4, set the recognition period to 120s and the update period to 20s to perform online recognition of the actual vehicle speed.

4. The intelligent energy management method for hybrid buses in an intelligent network environment according to claim 3 is characterized in that: The specific implementation of the XGBoost model: Construct multiple weak learners, where the first base learner is used to fit the sample data, and the fitting error is used as the target of the second base learner to continue fitting. All base learners are accumulated as the final result through a statistical residual approximation method; Among them, the feature of the i-th sample is x i , and the prediction value of the kth tree is f k (x i ), is the predicted value of the model; the optimization process starts from the first tree and performs optimization in sequence until all trees are optimized.

5. The intelligent energy management method for hybrid buses in an intelligent network environment according to claim 3 is characterized in that: The objective function of XGBoost is shown in the following formula, which consists of a loss function and a regularization term. The loss function is used to balance the training error, and the regularization term is used to control the complexity. In the formula is the loss function, n represents the sample overview, Ω(f k ) represents the complexity of the kth tree.

6. An intelligent energy management system for a hybrid bus in an intelligent networked environment, characterized in that it includes a cloud device and an edge device, wherein the cloud device can execute the content of S2 in the intelligent energy management method for a hybrid bus in an intelligent networked environment described in any one of claims 1-5; the edge device can execute the content of S3 in the intelligent energy management method for a hybrid bus in an intelligent networked environment described in any one of claims 1-5.