WO2021073036A1 - 一种燃料电池公交车实时全局优化智能控制系统及方法 - Google Patents

一种燃料电池公交车实时全局优化智能控制系统及方法 Download PDF

Info

Publication number
WO2021073036A1
WO2021073036A1 PCT/CN2020/079244 CN2020079244W WO2021073036A1 WO 2021073036 A1 WO2021073036 A1 WO 2021073036A1 CN 2020079244 W CN2020079244 W CN 2020079244W WO 2021073036 A1 WO2021073036 A1 WO 2021073036A1
Authority
WO
WIPO (PCT)
Prior art keywords
fuel cell
soc
real
time
value
Prior art date
Application number
PCT/CN2020/079244
Other languages
English (en)
French (fr)
Inventor
胡东海
王晶
何洪文
衣丰艳
周稼铭
高建平
李中延
刘新磊
Original Assignee
江苏大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 江苏大学 filed Critical 江苏大学
Priority to CH01054/20A priority Critical patent/CH717533B1/de
Publication of WO2021073036A1 publication Critical patent/WO2021073036A1/zh

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60L—PROPULSION OF ELECTRICALLY-PROPELLED VEHICLES; SUPPLYING ELECTRIC POWER FOR AUXILIARY EQUIPMENT OF ELECTRICALLY-PROPELLED VEHICLES; ELECTRODYNAMIC BRAKE SYSTEMS FOR VEHICLES IN GENERAL; MAGNETIC SUSPENSION OR LEVITATION FOR VEHICLES; MONITORING OPERATING VARIABLES OF ELECTRICALLY-PROPELLED VEHICLES; ELECTRIC SAFETY DEVICES FOR ELECTRICALLY-PROPELLED VEHICLES
    • B60L50/00—Electric propulsion with power supplied within the vehicle
    • B60L50/50—Electric propulsion with power supplied within the vehicle using propulsion power supplied by batteries or fuel cells
    • B60L50/75—Electric propulsion with power supplied within the vehicle using propulsion power supplied by batteries or fuel cells using propulsion power supplied by both fuel cells and batteries
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/60—Other road transportation technologies with climate change mitigation effect
    • Y02T10/70—Energy storage systems for electromobility, e.g. batteries
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
    • Y02T90/40—Application of hydrogen technology to transportation, e.g. using fuel cells

Definitions

  • the invention belongs to the technical field of new energy vehicle energy management, and in particular relates to a real-time global optimization intelligent control system and method for fuel cell buses.
  • Machine learning is a relatively young branch of artificial intelligence research. It is a science of artificial intelligence. Its main research object is artificial intelligence, especially how to improve the performance of specific algorithms in empirical learning. Incremental learning is a dynamic and gradual update. Algorithm means that every time new data is added, it is not necessary to rebuild all the knowledge bases. It only trains the updates caused by the new data on the basis of the original knowledge base; this is also more in line with the principle of human thinking and can avoid Repeat learning in the case of massive data. In actual databases, the amount of data tends to increase gradually. Therefore, when facing new data, the learning method should be able to make certain changes to the trained system to learn the knowledge contained in the new data. The time cost of modifying a trained system is usually lower than that of renewing it. The cost of training a system.
  • the prior art relates to a plug-in hybrid vehicle energy management method based on deep reinforcement learning.
  • the disadvantages of this invention are as follows: 1) It does not involve modifying the rules involved due to data changes, and deep reinforcement learning only uses massive amounts of data. Perform dimensionality reduction and fusion processing. When new data is added, a system needs to be retrained; 2) Dynamic gradual update algorithm is not involved. The data in the database is dynamically changing. When faced with new data, the learning method should be able to The trained system makes certain changes to learn from the knowledge contained in the new data.
  • the prior art also relates to an energy management method for plug-in hybrid electric vehicles based on intelligent prediction.
  • This invention has the following shortcomings: 1) The use of deep learning for model prediction has a greater impact on the timeliness and accuracy of database search Therefore, the prediction range can only reach the short-term; 2) The model needs to be reconstructed when there is a big difference from the target driving route.
  • the present invention provides a real-time global optimization intelligent control system and method for fuel cell buses.
  • the vehicle controller combines the optimal SOC prediction model to control the fuel cell working status.
  • the incremental learning model in the cloud workstation is used for training again, and the model parameters are updated and downloaded to the vehicle controller VCU in time.
  • a fuel cell bus real-time global optimization intelligent control system including a vehicle driving communication unit, a vehicle driving information prediction and analysis unit, a fuel cell vehicle control unit, and a vehicle driving information acquisition unit.
  • the fuel cell vehicle control unit is driven by the vehicle
  • the communication unit is signally connected to the vehicle driving information prediction and analysis unit, and the fuel cell vehicle control unit is also signally connected to the vehicle driving information acquisition unit;
  • the vehicle driving information prediction and analysis unit obtains the prediction model parameters and downloads them to the fuel cell vehicle control unit
  • the unit updates the optimal SOC prediction model; the fuel cell vehicle control unit controls the working state of the fuel cell.
  • the fuel cell vehicle control unit controls the working state of the fuel cell based on the power reference value and real-time power of the drive motor, the optimal SOC reference trajectory prediction value of the next operating condition segment of the vehicle, and the real-time SOC value of the battery pack Controlled.
  • the real-time SOC of the battery pack is detected in real-time by the battery management system BMS.
  • the power reference value of the drive motor is calculated by the MPC predictive control module in the fuel cell vehicle control unit according to the optimal SOC reference trajectory predicted value for the next operating condition segment.
  • the real-time power of the drive motor is calculated by the real-time power calculation module of the drive motor.
  • the prediction model parameters are the feature parameters calculated by the feature parameter calculation module and transmitted to the incremental learning model, and are obtained after training; the feature parameter calculation module receives the speed information receiving module and the dynamic programming module sent Vehicle driving condition information and optimal SOC reference trajectory.
  • the incremental learning model trains the feature parameters to obtain the regression model SVM1 and the support vector set SV1, integrates the new working condition information, and synthesizes the new sample library data with the support vector set SV1, and continues training
  • the new regression model SVM2 and the support vector set SV2 are obtained as the final model.
  • a real-time global optimization intelligent control method for fuel cell buses includes the following steps:
  • Step 1 Use the incremental learning model to train the obtained working condition information and the optimal SOC reference trajectory to obtain the optimal SOC prediction model parameters, and download them to the optimal SOC prediction model module to update the optimal SOC prediction model;
  • Step 2 Input the characteristic parameters of the current operating condition segment into the optimal SOC prediction model module, and output the optimal SOC reference trajectory prediction value of the next operating condition segment;
  • Step 3 Input the predicted value of the optimal SOC reference trajectory for the next operating condition segment into the MPC predictive control module to obtain the reference value of the drive motor power;
  • Step 4 The fuel cell control unit FCU judges the SOC value and power, and controls the working state of the fuel cell;
  • Step 5 The vehicle controller VCU judges whether the bus has completed a journey. If the journey is not over, it returns to step 2 for looping; if it ends, the cloud analysis workstation re-downloads the updated prediction model parameters to the vehicle controller VCU Update the optimal SOC prediction model.
  • the step 4 is specifically: judging whether the real-time SOC value of the battery pack is greater than the maximum value SOC max of the optimal SOC reference trajectory prediction value for the next operating condition segment, and if SOC> SOC max , judging whether the real-time power P is less than the power reference value P min, if P ⁇ P min, of the fuel cell will not turn off the charge control unit FCU; if P ⁇ P min, then determine if the real power P is greater than the reference power value P max, if P> P max, of the fuel cell The control unit FCU starts to work.
  • the fuel cell control unit FCU keeps the state unchanged; when the real-time SOC ⁇ SOC min or SOC min ⁇ SOC ⁇ SOC max and P> P max , the fuel cell control The unit FCU turns on and discharges the fuel cell stack, otherwise the state remains unchanged.
  • Figure 1 is a structural diagram of a real-time global optimization intelligent control system for a fuel cell bus according to the present invention
  • FIG. 2 is a schematic diagram of the structure of the fuel cell bus of the present invention.
  • FIG. 3 is a schematic diagram of the entire vehicle control of the fuel cell bus of the present invention.
  • Fig. 4 is a schematic diagram of generating the optimal SOC prediction model of the present invention.
  • Fig. 5 is a network structure diagram of the incremental learning model based on the support vector machine of the present invention.
  • Fig. 6 is a working flow chart of a real-time global optimization intelligent control of a fuel cell bus of the present invention
  • Fig. 7 is a case analysis diagram of a sudden working condition of the present invention.
  • 1-fuel cell bus 2-vehicle controller VCU, 3-wireless communication system, 4-satellite, 5-base station, 6-wired communication system, 7-cloud analysis workstation, 8-incremental learning model, 9-Optimal SOC prediction model module, 10-MPC predictive control module, 11-fuel cell control unit FCU, 12-fuel cell hydrogen storage tank, 13-fuel cell stack, 14-speed sensor, 15-acceleration sensor, 16 -Drive motor real-time power calculation module, 17-motor controller MCU, 18-drive motor, 19-battery management system BMS, 20-speed information receiving module, 21-dynamic planning module, 22-characteristic parameter calculation module.
  • a real-time global optimization intelligent control system for a fuel cell bus of the present invention includes a vehicle travel communication unit, a vehicle travel information prediction and analysis unit, a fuel cell vehicle control unit, and a vehicle travel information acquisition unit.
  • the vehicle driving communication unit includes a wireless communication system 3, a satellite 4, a base station 5, and a wired communication system 6, the vehicle driving information prediction analysis unit includes a cloud analysis workstation 7, and the fuel cell vehicle control unit includes a vehicle controller VCU 2, fuel cell control unit FCU 11, fuel cell hydrogen storage tank 12, fuel cell stack 13, motor controller MCU 17 and drive motor 18, and fuel cell hydrogen storage tank 12 provides fuel for fuel cell stack 13;
  • the vehicle driving information collection unit includes a speed sensor 14, an acceleration sensor 15, and a real-time power calculation module 16 of the drive motor. Both the speed sensor 14 and the acceleration sensor 15 are connected to the vehicle controller VCU 2 signal, and the real-time power calculation module 16 of the drive motor is connected to the fuel cell.
  • the control unit FCU 11 and the motor controller MCU 17 are signal connected.
  • vehicle controller VCU 2 fuel cell control unit FCU 11, drive motor 18, motor controller MCU 17, speed sensor 14, acceleration sensor 15, drive motor real-time power calculation module 16, fuel cell hydrogen storage
  • fuel cell control unit FCU 11 drive motor 18, motor controller MCU 17, speed sensor 14, acceleration sensor 15, drive motor real-time power calculation module 16, fuel cell hydrogen storage
  • fuel cell control unit FCU 11 drive motor 18, motor controller MCU 17, speed sensor 14, acceleration sensor 15, drive motor real-time power calculation module 16, fuel cell hydrogen storage
  • the tank 12 and the fuel cell stack 13 are both arranged on the top of the fuel cell bus 1.
  • the vehicle controller VCU 2 is connected to the satellite 4 through the wireless communication system 3, the satellite 4 is connected to the base station 5 through the wireless communication system 3, and the base station 5 is connected to the cloud analysis workstation 7 through the wired communication system 6;
  • the cloud analysis workstation 7 is provided with a speed information receiving module 20 and a dynamic planning module 21 that are connected to each other.
  • the speed information receiving module 20 and the dynamic planning module 21 are all connected to the characteristic parameter calculation module 22, and the characteristic parameter calculation module 22 is connected to the characteristic parameter calculation module 22.
  • Incremental learning model 8 is connected; the fuel cell bus 1 downloads the prediction model parameters to the vehicle controller VCU 2 through the vehicle driving communication system at the starting point of travel, and the vehicle controller VCU 2 has an interconnected optimal SOC prediction model 9 With MPC predictive control 10, the final fuel cell control unit FCU 11 integrates the power reference value and real-time power of the drive motor 18, and outputs the working status of the fuel cell.
  • the vehicle controller VCU 2 includes an optimal SOC prediction model module 9 and an MPC prediction control module 10 connected in sequence;
  • the MPC predictive control module 10 calculates the reference value of the drive motor power, and the drive motor power The reference value is transmitted to the fuel cell control unit FCU 11.
  • the battery management system BMS 19 detects the SOC value of the battery pack in real time, the real-time power calculation module 16 of the drive motor is connected to the motor controller MCU 17, and the speed and torque of the drive motor 18 are obtained through the motor controller MCU 17, thereby obtaining real-time power;
  • the fuel cell control unit FCU 11 receives the power reference value and real-time power of the drive motor 18, the next working condition segment optimal SOC reference trajectory predicted value, and the battery pack real-time SOC value, and controls the opening, closing, and maintenance of the fuel cell.
  • the speed information receiving module 20 transmits the vehicle driving condition information to the dynamic planning module 21 and obtains the optimal SOC reference trajectory.
  • the speed information receiving module 20 and the dynamic planning module 21 then The vehicle driving condition information and the optimal SOC reference trajectory are sent to the characteristic parameter calculation module 22, and the characteristic parameters calculated by the characteristic parameter calculation module 22 are transmitted to the incremental learning model 8.
  • the fuel cell bus 1 downloads the trained prediction model parameters to the optimal SOC prediction model module 9 in the vehicle controller VCU 2 through the vehicle driving communication system at the starting point of driving; the vehicle is driving
  • the vehicle controller VCU 2 divides the operating condition into equal segments of operating condition segments TS according to the time segments, and uploads them to the speed information receiving module 20.
  • bus 1 uses the operating condition data (including speed and acceleration) of the new trip and its optimal SOC reference trajectory to obtain feature parameters for incremental learning, generate new predictive model parameters and download them to the optimal SOC prediction
  • the model module 9 updates the optimal SOC prediction model therein.
  • the original training samples can be deleted due to the use of the incremental learning technology, avoiding the accumulation of a large amount of data, increasing the training burden and the production cost.
  • the incremental learning model 8 trains the feature parameters to obtain the regression model SVM1 and the support vector set SV1, integrates the new working condition information, and synthesizes the new sample library data with the support vector set SV1, continue Trained to obtain a new regression model SVM2 and support vector set SV2 as the final model.
  • the feature parameters of the current operating condition segment are input to the incremental learning model, and the feature parameters of the optimal SOC reference trajectory of the next operating condition segment are used as the model output.
  • the mapping relationship with the characteristic parameters of the optimal SOC reference trajectory of the next operating condition segment realizes the calculation and prediction of the characteristic parameters of the optimal SOC reference trajectory of the next operating condition segment.
  • Y i+1 of the next operating condition segment it can be expressed as:
  • the parameters in the above formula respectively represent several characteristics of the optimal SOC reference trajectory of the working condition segment: maximum SOC value, minimum SOC value, SOC standard deviation, SOC maximum rate of change, and average SOC.
  • K max max: K j (7)
  • n is the number of data points in the operating condition segment
  • ⁇ t is the time interval of data points
  • K is the rate of change of SOC
  • j 1, 2,...,n,.
  • the characteristic parameters of the current operating condition segment are recorded as:
  • the parameters in the above formula respectively represent the working condition information of the current working condition segment: maximum speed, minimum speed, maximum acceleration, minimum acceleration and average speed.
  • V max max: V j (9)
  • V is the speed of the fuel cell bus 1
  • a is the acceleration of the fuel cell bus 1.
  • Dynamic programming mainly includes the forward method and the reverse method.
  • the forward method uses the state transition equation from the first stage to recurse from front to back. The principle is as follows:
  • stage index function r k is the energy consumption of the k-th stage, calculated as follows:
  • f(P fc ) is the energy consumed by the fuel cell when outputting power P fc; It is the equivalent energy consumption of the power battery; Z includes the dynamic working efficiency of the fuel cell, DC-DC converter and power battery, which can be calculated through experiments or equivalent circuit models.
  • the state transition equation from stage k to stage k+1 is:
  • P b_k is the power battery output power
  • U b is the bus voltage
  • C b is the power battery capacity.
  • the control parameter is the fuel cell output power P fc_k at the kth stage, and the constraint conditions of the state variables and control variables are:
  • P fc_max is the maximum output power of the fuel cell
  • P b_min and P b_max are the maximum charge and discharge power of the power battery
  • the optimization goal is to find the optimal control variable P fc_k in the entire driving cycle to minimize the energy consumption J:
  • the state variable SOC is divided into N nodes according to a certain step length within the range of SOC min and SOC max, and each node saves the optimal trajectory to the node.
  • the calculation process of node i in step k is: first find all nodes that can be transferred to node i in step k-1 under constraints, calculate the cumulative energy consumption f k for these state transitions, and make the smallest state transition that is Is the optimal strategy through node i in the k-th step. Recursively to the end of the loop, find the node with the smallest f k , and find the optimal SOC reference trajectory through the trajectory information saved by the node.
  • the input data is normalized, and the value range of the normalized data is [0,1].
  • the deviation standardization method is used for normalization, the conversion formula is
  • max is the maximum value of the sample data
  • min is the minimum sample data
  • X is the original training data (including a large segment of the current operating conditions characteristic parameter X i)
  • X * is the normalized data.
  • the optimal SOC reference trajectory for the next operating condition segment is:
  • W ⁇ W 1 ,W 2 ,,W i ,...W N ⁇
  • ⁇ the maximum allowable error of the regression
  • C the weight parameter used to balance the two, called the regularization parameter.
  • ⁇ and ⁇ * are dual parameters.
  • the predicted value is denormalized and restored to the predicted value of the optimal SOC reference trajectory for the next operating condition segment:
  • a real-time global optimization intelligent control method for a fuel cell bus of the present invention specifically includes the following steps:
  • Step 1 The speed sensor 14 and the acceleration sensor 15 pre-collect the working condition information of the fuel cell bus 1 and send it to the vehicle controller VCU2, the vehicle controller VCU2 sends it to the speed information receiving module 20, and the speed information receiving module 20 sends it Give the dynamic planning module 21 to obtain the optimal SOC reference trajectory for the next working condition.
  • the characteristic parameter calculation module 22 receives the operating condition information and the optimal SOC reference trajectory for the next operating condition sent by the speed information receiving module 20 and the dynamic planning module 21, and calculates its characteristic parameters respectively.
  • Step 2 Use the incremental learning model 8 to train the obtained working condition information and the optimal SOC reference trajectory, and then obtain the optimal SOC prediction model parameters, and download them to the optimal SOC prediction model module 9.
  • Step 3 Input the characteristic parameters of the current operating condition segment into the optimal SOC prediction model module 9, and then output the optimal SOC reference trajectory prediction value for the next operating condition segment.
  • Step 4 Input the optimal SOC reference trajectory prediction value of the next operating condition segment into the MPC prediction control module 10, output the corresponding speed and torque, and then obtain the power P reference value.
  • the power P reference value includes the maximum value P max and The minimum value P min .
  • Step 5 The battery management system BMS 19 monitors the real-time SOC value of the battery pack, and compares it with the predicted value of the next working condition segment TS i+1 output from the step 3 and the optimal SOC reference trajectory.
  • Step 6 The real-time power calculation module 16 of the drive motor measures the real-time power P from the speed and torque of the drive motor 18, and compares it with the reference value of the drive motor power obtained in step 4.
  • Step 7 The fuel cell control unit FCU 11 comprehensively judges the SOC value and power according to the power-following energy control strategy, specifically:
  • Step 8 The vehicle controller VCU 2 judges whether the fuel cell bus 1 has reached the end to complete a journey. If the journey is not over, it returns to the step three cycle; if it is over, the cloud analysis workstation 7 uses the incremental learning model 8 to compare the new journey. Increase the training data to complete incremental learning and re-download the updated prediction model parameters to the optimal SOC prediction model module 9 in the vehicle controller VCU 2 to update the optimal SOC prediction model.
  • the vehicle controller VCU 2 samples the operating condition information and divides the segments according to time. At the beginning of each operating condition segment The control starts at the beginning time point; the characteristic parameters of the current operating condition segment are input into the optimal SOC prediction model module 9, and the optimal SOC reference trajectory prediction value of the next operating condition segment is obtained as an input and sent to the MPC predictive control module 10, MPC predictive control
  • the module 10 obtains the reference value of the drive motor power
  • the battery management system BMS module 19 monitors the SOC value of the fuel cell vehicle battery pack in real time
  • the drive motor real-time power calculation module 16 uses the motor controller MCU 17 to calculate the motor speed and torque when the vehicle is running.
  • the MPC prediction control module 10 predict the next segment of the optimal SOC reference trajectory prediction value .
  • the reference value of the drive motor power is input to the fuel cell control unit FCU 11, and the working status of the fuel cell is judged according to the power-following energy control strategy, and then the control of the next working condition segment is completed.
  • the fuel cell bus 1 continuously uploads the working condition information of each segment through the vehicle driving communication system.
  • the cloud analysis workstation 7 performs incremental learning and training through the working condition information uploaded in real time.
  • the cloud analysis workstation 7 uses the incremental learning model 8 to perform incremental learning training on the newly added data, then updates the model parameters, and downloads the trained model parameters to the fuel cell vehicle controller VCU2.
  • the vehicle runs again under the same emergency conditions, it can adapt to changes in the new environment.

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Strategic Management (AREA)
  • Physics & Mathematics (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • General Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Marketing (AREA)
  • Software Systems (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Primary Health Care (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Development Economics (AREA)
  • General Engineering & Computer Science (AREA)
  • Game Theory and Decision Science (AREA)
  • Computing Systems (AREA)
  • Medical Informatics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Sustainable Energy (AREA)
  • Sustainable Development (AREA)
  • Mechanical Engineering (AREA)
  • Transportation (AREA)
  • Power Engineering (AREA)
  • Electric Propulsion And Braking For Vehicles (AREA)
  • Fuel Cell (AREA)

Abstract

一种燃料电池公交车实时全局优化智能控制系统及方法，燃料电池公交车（1）在行驶起点时，车辆行驶通信单元下载预测模型参数至燃料电池整车控制器（VCU 2）；电池管理系统、驱动电机实时功率计算模块（16）分别得到电池组实时SOC和实时功率（P），最优SOC预测模型模块（9）得到下一工况片段最优SOC参考轨迹预测值，MPC预测控制模块（10）得到功率参考值，上述参数均输入到燃料电池控制单元（FCU 11），判断燃料电池的工作状态。在行驶过程中，公交车（1）不断通过车辆行驶通信单元上传片段化工况信息，每完成一次行程之后云端分析工作站（7）通过实时上传的工况信息进行增量学习训练，更新最优SOC预测模型（9）。所述智能控制系统能够准确、灵活得实时控制燃料电池公交车（1），降低燃料消耗。

Description

一种燃料电池公交车实时全局优化智能控制系统及方法 技术领域
本发明属于新能源车辆能量管理技术领域，尤其涉及一种燃料电池公交车实时全局优化智能控制系统及方法。
背景技术
随着我国汽车保有量不断增加，汽车产业的能源、环境压力也随之不断增加。由于石油等不可再生能源对外依赖度逐年上升，实施能源替代迫在眉睫，故氢能以其热值较高、储量丰富以及具有极好的环境友好度进入公众视野；从氢能的应用角度看，燃料电池汽车成为重点研究方向之一，统计显示，目前中国市场已经有超过40家车企参与了氢燃料电池汽车生产制造。另一方面，公共出行交通工具的使用也对能源、环境压力起到极大的减缓作用，故推行燃料电池电动公交车驻扎市场势在必行。
机器学习是人工智能研究较为年轻的分支，是一门人工智能的科学，其主要研究对象是人工智能，特别是如何在经验学习中改善具体算法的性能；而增量学习是一种动态逐步更新算法，是指每当新增数据时不需要重建所有的知识库，仅仅在原有知识库的基础上，对新增数据所引起的更新进行训练；这也更加符合人的思维原理，能够避免在海量数据的情况下重复学习。在实际的数据库中，数据量往往是逐渐增加的。因此，在面临新的数据时，学习方法应能对训练好的系统进行某些改动，以对新数据中蕴涵的知识进行学习，而对一个训练好的系统进行修改的时间代价通常低于重新训练一个系统所需的代价。
现有技术涉及一种基于深度强化学习的插电式混合动力车辆能量管理方法，该发明存在的不足有：1)未涉及修改因数据变化而涉及的规则，深度强化学习只是利用海量数据的情况进行降维、融合处理，当有新增数据加入时需要重新训练一个系统；2)未涉及动态逐步更新算法，数据库中的数据是动态变化的，在面临新的数据时，学习方法应能对训练好的系统进行某些改动，以对新数据中蕴涵的知识进行学习。现有技术还涉及一种基于智能预测的插电式混合动力车辆的能量管理方法，该发明存在的不足有：1)采用深度学习进行模型预测对数据库搜索的时效性、准确度有较大影响，因此预测范围只能到达短期；2)当出现与所述目标行驶路线差异较大时需重构模型。
因此，如何方便而有效的反映数据的变化成为较为迫切的研究话题。设计一套高效、准确、灵活的燃料电池公交车实时全局优化智能控制系统及方法具有极高的现实意义。
发明内容
本发明提供了一种燃料电池公交车实时全局优化智能控制系统及方法，燃料电池公交车 在行驶过程中，整车控制器结合最优SOC预测模型对燃料电池工作状态的控制，在新增工况信息时，利用云端工作站中的增量学习模型再次进行训练，更新模型参数并及时下载至整车控制器VCU。
本发明的技术方案为：
一种燃料电池公交车实时全局优化智能控制系统，包括车辆行驶通信单元、车辆行驶信息预测分析单元、燃料电池整车控制单元以及车辆行驶信息采集单元，所述燃料电池整车控制单元通过车辆行驶通信单元与车辆行驶信息预测分析单元信号连接，所述燃料电池整车控制单元还与车辆行驶信息采集单元信号连接；所述车辆行驶信息预测分析单元获取预测模型参数，下载至燃料电池整车控制单元进行最优SOC预测模型更新；所述燃料电池整车控制单元控制燃料电池的工作状态。
上述技术方案中，所述燃料电池整车控制单元控制燃料电池的工作状态是根据驱动电机的功率参考值和实时功率、车辆下一工况片段最优SOC参考轨迹预测值和电池组实时SOC值进行控制的。
上述技术方案中，所述电池组实时SOC是由电池管理系统BMS实时检测。
上述技术方案中，所述下一工况片段最优SOC参考轨迹预测值是由燃料电池整车控制单元中的最优SOC预测模型模块接收当前工况片段特征参数，由Y *＝min+f(X)(max-min)得到，其中f(X)是回归函数，max为样本数据的最大值，min为样本数据的最小值，Y *为下一工况片段最优SOC参考轨迹预测值。
上述技术方案中，所述驱动电机的功率参考值是燃料电池整车控制单元中的MPC预测控制模块根据下一工况片段最优SOC参考轨迹预测值计算得到。
上述技术方案中，所述驱动电机的实时功率是由驱动电机实时功率计算模块计算得到。
上述技术方案中，所述预测模型参数是由特征参数计算模块计算出的特征参数传送至增量学习模型，经过训练后得到；所述特征参数计算模块接收速度信息接收模块和动态规划模块发送的车辆行驶工况信息和最优SOC参考轨迹。
上述技术方案中，所述增量学习模型对特征参数进行训练，得到回归模型SVM1和支持向量集SV1，对新增工况信息进行整合，并与支持向量集SV1合成新样本库数据，继续训练得到全新的回归模型SVM2和支持向量集SV2作为最终模型。
一种燃料电池公交车实时全局优化智能控制方法，包括以下步骤：
步骤一，利用增量学习模型对得到的工况信息和最优SOC参考轨迹进行训练，得到最优SOC预测模型参数，下载至最优SOC预测模型模块进行最优SOC预测模型更新；
步骤二，将当前工况片段特征参数输入到最优SOC预测模型模块，输出下一工况片段最优SOC参考轨迹预测值；
步骤三，将下一工况片段最优SOC参考轨迹预测值输入到MPC预测控制模块中，得到驱动电机功率参考值；
步骤四，燃料电池控制单元FCU对SOC值与功率判断，控制燃料电池的工作状态；
步骤五，整车控制器VCU判断公交车是否完成一次行程，若行程未结束，则返回步骤二循环进行；若结束，则云端分析工作站将更新的预测模型参数重新下载至整车控制器VCU中进行最优SOC预测模型更新。
进一步，所述步骤四具体为：判断电池组实时SOC值是否大于下一工况片段最优SOC参考轨迹预测值的最大值SOC max，若SOC＞SOC max，则判断实时功率P是否小于功率参考值P min，若P＜P min，则燃料电池控制单元FCU关闭不再充电；如果P≥P min，则继续判断实时功率P是否大于功率参考值P max，若P＞P max，则燃料电池控制单元FCU开始工作，当P min≤P≤P max，则燃料电池控制单元FCU保持状态不变；当实时SOC＜SOC min或者SOC min≤SOC≤SOC max且P＞P max，则燃料电池控制单元FCU使燃料电池电堆开启并放电，否则保持状态不变。
本发明有益效果：
1)避免在海量数据的情况下重复学习。当新增数据时，不需要重建所有的知识库，而是在原有模型的基础上，仅对新增数据所引起的更新进行训练。
2)节约云端存储空间，降低成本。由于采用增量学习，原有的训练数据在训练结束后即可删除，避免造成数据堆积，节约成本。
3)工况适应性不断提高。车辆在一次行程结束后，云端即对新增数据进行增量训练，并将训练好的模型下载至车辆，使得车辆在同一通勤路线上的工况适应性不断提高。
附图说明
图1为本发明一种燃料电池公交车实时全局优化智能控制系统结构图；
图2为本发明燃料电池公交车结构示意图；
图3为本发明燃料电池公交车整车控制原理图；
图4为本发明最优SOC预测模型生成原理图；
图5为本发明基于支持向量机的增量学习模型网络结构图；
图6为本发明一种燃料电池公交车实时全局优化智能控制工作流程图；
图7为本发明突发工况案例分析图；
其中：1-燃料电池公交车，2-整车控制器VCU，3-无线通信系统，4-卫星，5-基站，6-有线通信系统，7-云端分析工作站，8-增量学习模型，9-最优SOC预测模型模块，10-MPC预测控制模块，11-燃料电池控制单元FCU，12-燃料电池储氢罐，13-燃料电池电堆，14-速度传感器，15-加速度传感器，16-驱动电机实时功率计算模块，17-电机控制器MCU，18-驱动电机，19-电池管理系统BMS，20-速度信息接收模块，21-动态规划模块，22-特征参数计算模块。
具体实施方式
下面结合附图来说明本发明一种燃料电池公交车实时全局优化智能控制系统及方法的结构和工作原理。
如图1和图2所示，本发明一种燃料电池公交车实时全局优化智能控制系统，包括车辆行驶通信单元、车辆行驶信息预测分析单元、燃料电池整车控制单元以及车辆行驶信息采集单元。所述车辆行驶通信单元包括无线通信系统3、卫星4、基站5和有线通信系统6，所述车辆行驶信息预测分析单元包括云端分析工作站7，所述燃料电池整车控制单元包括整车控制器VCU 2、燃料电池控制单元FCU 11、燃料电池储氢罐12、燃料电池电堆13、电机控制器MCU 17和驱动电机18，燃料电池储氢罐12给燃料电池电堆13提供燃料；所述车辆行驶信息采集单元包括速度传感器14、加速度传感器15和驱动电机实时功率计算模块16，速度传感器14和加速度传感器15均与整车控制器VCU 2信号连接，驱动电机实时功率计算模块16与燃料电池控制单元FCU 11及电机控制器MCU 17信号连接。
如图2所示，整车控制器VCU 2、燃料电池控制单元FCU 11、驱动电机18、电机控制器MCU 17、速度传感器14、加速度传感器15、驱动电机实时功率计算模块16、燃料电池储氢罐12、燃料电池电堆13均设置在燃料电池公交车1顶部。
在车辆行驶过程中，整车控制器VCU 2通过无线通信系统3与卫星4连接，卫星4通过无线通信系统3与基站5连接，基站5通过有线通信系统6与云端分析工作站7连接；如图4所示，云端分析工作站7内部设有相互连接的速度信息接收模块20和动态规划模块21，速度信息接收模块20、动态规划模块21均与特征参数计算模块22连接，特征参数计算模块22与增量学习模型8连接；燃料电池公交车1在行驶起点通过车辆行驶通信系统下载预测模型参数至整车控制器VCU 2，整车控制器VCU 2中设有相互连接的最优SOC预测模型9和MPC预测控制10，最终燃料电池控制单元FCU 11集成驱动电机18的功率参考值和实时功率，输出燃料电池的工作状态。
如附图3所示，整车控制器VCU 2包括依次连接的最优SOC预测模型模块9和MPC预测控制模块10；最优SOC预测模型模块9接收当前工况片段特征参数，由 Y *＝min+f(X)(max-min)得到下一工况片段最优SOC参考轨迹预测值，传输给MPC预测控制模块10，由MPC预测控制模块10计算得到驱动电机功率参考值，驱动电机功率参考值传输给燃料电池控制单元FCU 11。电池管理系统BMS 19实时检测电池组的SOC值，驱动电机实时功率计算模块16与电机控制器MCU 17信号连接，通过电机控制器MCU 17获取驱动电机18的转速、转矩，进而得到实时功率；燃料电池控制单元FCU 11接收驱动电机18的功率参考值和实时功率、下一工况片段最优SOC参考轨迹预测值和电池组实时SOC值，控制燃料电池的开启、关闭和保持。
如附图4所示，云端分析工作站7中，速度信息接收模块20将车辆行驶工况信息输送给动态规划模块21并得到最优SOC参考轨迹，速度信息接收模块20、动态规划模块21再把车辆行驶工况信息、最优SOC参考轨迹发送给特征参数计算模块22，特征参数计算模块22计算出的特征参数传送至增量学习模型8，经过训练后得到预测模型参数(包括对偶参数α、α *，RBF核函数，偏差b)；燃料电池公交车1在行驶起点通过车辆行驶通信系统下载训练好的预测模型参数至整车控制器VCU 2中的最优SOC预测模型模块9；车辆行驶过程中，整车控制器VCU 2按照时间片段将工况划分成等段的工况片段TS，并上传至速度信息接收模块20。公交车1每完成一次行程后，利用新行程的工况数据(包括速度和加速度)及其最优SOC参考轨迹获取特征参数进行增量学习，生成新的预测模型参数并下载至最优SOC预测模型模块9更新其中的最优SOC预测模型。
所述云端分析工作站7每次经过增量学习模型8的训练后，由于使用了增量学习技术，原有的训练样本可被删除，避免造成大量数据堆积，增加训练负担和生产成本。
如附图5所示，增量学习模型8对特征参数进行训练，得到回归模型SVM1和支持向量集SV1，对新增工况信息进行整合，并与支持向量集SV1合成新样本库数据，继续训练得到全新的回归模型SVM2和支持向量集SV2作为最终模型。将当前工况片段特征参数为增量学习模型输入，将下一工况片段最优SOC参考轨迹的特征参数作为模型输出，利用大量上午输入输出训练增量学习模型，建立当前工况片段特征参数与下一工况片段最优SOC参考轨迹的特征参数之间的映射关系，实现对下一工况片段最优SOC参考轨迹的特征参数的推算和预测。对于下一工况片段的最优SOC参考轨迹Y i+1，可以表示为：
Figure PCTCN2020079244-appb-000001
上式中的参数分别表示工况片段最优SOC参考轨迹的几项特征：最大SOC值、最小SOC值、SOC标准差、SOC最大变化率、平均SOC。
SOC max＝max:SOC j             (2)
SOC min＝min:SOC j              (3)
Figure PCTCN2020079244-appb-000002
Figure PCTCN2020079244-appb-000003
Figure PCTCN2020079244-appb-000004
K max＝max：K j          (7)
其中，n为该工况片段中的数据点个数，Δt为数据点时间间隔，K为SOC变化率，j＝1,2,…,n，。
为实现下一工况片段最优SOC参考轨迹预测，将当前工况片段特征参数记为：
Figure PCTCN2020079244-appb-000005
上式中的参数分别表示当前工况片段的工况信息：最大速度、最小速度，最大加速度、最小加速度及平均速度。
V max＝max：V j               (9)
Figure PCTCN2020079244-appb-000006
a max＝max：a j             (11)
a min＝min：a j             (12)
Figure PCTCN2020079244-appb-000007
其中V为燃料电池公交车1的车速，a为燃料电池公交车1的加速度。
在进行增量学习的训练前，需要利用动态规划模块21对车辆行驶的路谱信息进行计算，以得到最优SOC参考轨迹。动态规划主要包括顺推法和逆推法，其中顺推法是从第一阶段开始利用状态转移方程由前向后递推，其原理如下：
Figure PCTCN2020079244-appb-000008
其中，k为阶段编号；s k为状态变量；u k为控制变量；r k为阶段指标函数；f k为最优指标函数；T k为状态转移函数。
取动力电池SOC为状态变量，整个步长被划分为步长为1s的m个阶段。阶段指标函数r k为第k个阶段能量消耗，计算如下：
Figure PCTCN2020079244-appb-000009
其中：f(P fc)为燃料电池在输出功率P fc时消耗能量；
Figure PCTCN2020079244-appb-000010
为动力电池等效能耗；Z包含了燃料电池、DC-DC变换器以及动力电池的动态工作效率，可通过实验或等效电路模型计算得到。第k阶段到k+1阶段的状态转移方程为：
Figure PCTCN2020079244-appb-000011
其中：P b_k为动力电池输出功率；U b为总线电压；C b为动力电池容量。控制参数为第k阶段燃料电池输出功率P fc_k，状态变量和控制变量约束条件为：
Figure PCTCN2020079244-appb-000012
其中，P fc_max为燃料电池最大输出功率；P b_min和P b_max为动力电池最大充电和放电功率；优化目标为找到整个驾驶循环内的最优控制变量P fc_k使得能量消耗量J最小：
Figure PCTCN2020079244-appb-000013
将状态变量SOC在SOC min和SOC max范围内按一定步长划分为N个节点，每个节点都保存了到达该节点的最优轨迹。第k步节点i的计算过程为：首先找出第k-1步在约束条件下能够转移到节点i的全部节点，计算发生这些状态转移的累积能耗f k，使之最小的状态转移即为第k步经过节点i的最优策略。递推到循环终点，找出f k最小的节点，通过节点保存的轨迹信息即可找出最优SOC参考轨迹。
将当前工况片段特征参数X i作为增量学习模型的输入，将下一工况片段的最优SOC参考轨迹Y i+1作为增量学习模型的输出。为减小输入数据间数量级差别较大引起的网络预测误 差，对输入数据作归一化处理，归一化数据的取值范围为[0,1]。采用离差标准化方法进行归一化，转换式为：
Figure PCTCN2020079244-appb-000014
其中，max为样本数据的最大值，min为样本数据的最小值，X为原始训练数据(包括大量的当前工况片段特征参数X i)，X *为归一化数据。
运用SVM模型对下一工况片段最优SOC参考轨迹进行预测之后，还需对预测的结果按式(3)进行反归一化处理，使预测得到的数据符合实际范围和意义。据此，下一工况片段最优SOC参考轨迹的表达式为：
Figure PCTCN2020079244-appb-000015
式中，
Figure PCTCN2020079244-appb-000016
是从输入空间到高位特征空间的非线性映射；权重W i和偏差b由下式得到：
Figure PCTCN2020079244-appb-000017
式中W＝{W 1,W 2,,W i,…W N}，
Figure PCTCN2020079244-appb-000018
为正则化部分；第二项中
Figure PCTCN2020079244-appb-000019
是经验风险，由下式给出的不敏感损失函数L ε来度量，ε为回归允许最大误差；C为用来平衡两者的权重参数，称为正则化参数。
Figure PCTCN2020079244-appb-000020
为了得到W i和b，通过RBF核函数K(X i,X j)将式(6)进行转化为：
Figure PCTCN2020079244-appb-000021
式中α、α *是对偶参数。
于是，回归函数变成了下面的精确形式：
Figure PCTCN2020079244-appb-000022
得到增量学校模型的预测输出之后，将预测值进行反归一化，还原为下一工况片段最优SOC参考轨迹的预测值：
Y *＝min+f(X)(max-min)      (25)
结合附图6，本发明一种燃料电池公交车实时全局优化智能控制方法，具体包括以下步骤：
步骤一：速度传感器14、加速度传感器15预先采集燃料电池公交车1的工况信息发送给整车控制器VCU 2，整车控制器VCU 2发送给速度信息接收模块20，速度信息接收模块20输送给动态规划模块21，得到下一工况最优SOC参考轨迹。特征参数计算模块22接收速度信息接收模块20和动态规划模块21发送的工况信息及下一工况最优SOC参考轨迹，并分别计算其特征参数。
步骤二：利用增量学习模型8对得到的工况信息和最优SOC参考轨迹进行训练，进而得到最优SOC预测模型参数，下载至最优SOC预测模型模块9。
步骤三：将当前工况片段特征参数输入到最优SOC预测模型模块9，进而输出下一工况片段最优SOC参考轨迹预测值。
步骤四：将下一工况片段最优SOC参考轨迹预测值输入到MPC预测控制模块10中，输出相应的转速、转矩，进而得到功率P参考值，功率P参考值包括最大值P max和最小值P min。
步骤五：电池管理系统BMS 19监测电池组实时SOC值，并与步骤三输出的下一工况片段TS i+1最优SOC参考轨迹预测值进行比较。
步骤六：驱动电机实时功率计算模块16由驱动电机18的转速、转矩，进而测得实时功率P，并与步骤四得到的驱动电机功率参考值进行比较。
步骤七：燃料电池控制单元FCU 11根据功率跟随式能量控制策略，综合对SOC值与功率进行判断，具体为：
首先判断电池组实时SOC值是否大于下一工况片段最优SOC参考轨迹预测值的最大值SOC max，若SOC＞SOC max，则判断实时功率P是否小于功率参考值P min，若P＜P min，则燃料电池控制单元FCU 11关闭不再充电；如果P≥P min，则继续判断实时功率是否大于功率参考值P max，若P＞P max，则燃料电池控制单元FCU 11开始工作，当P min≤P≤P max，则燃料电池控制单元FCU 11保持状态不变。当实时SOC＜SOC min或者SOC min≤SOC≤SOC max且P＞P max，则燃料电池控制单元FCU 11使燃料电池电堆13开启并放电，否则保持状态不变。
步骤八：整车控制器VCU 2判断燃料电池公交车1是否到达终点完成一次行程，若行程未结束，则返回步骤三循环进行；若结束，则云端分析工作站7通过增量学习模型8对新增训练数据完成增量学习并将更新的预测模型参数重新下载至整车控制器VCU 2中的最优SOC预测模型模块9中进行最优SOC预测模型更新。
下面结合附图7来具体描述本发明的工作流程：燃料电池公交车1在行驶过程中，整车控制器VCU 2对工况信息进行采样并按照时间划分片段，在每段工况片段的起始时间点开始 进行控制；将当前工况片段特征参数输入最优SOC预测模型模块9中，得到下一工况片段最优SOC参考轨迹预测值作为输入传送给MPC预测控制模块10，MPC预测控制模块10得到驱动电机功率参考值，电池管理系统BMS模块19实时对燃料电池车电池组SOC值进行监测，驱动电机实时功率计算模块16通过电机控制器MCU 17对车辆行驶时的电机转速转矩进行监测进而计算得到驱动电机实时功率；将测得的实时电池组SOC值、驱动电机实时功率与最优SOC预测模型模块9、MPC预测控制模块10预测输出的下一片段最优SOC参考轨迹预测值、驱动电机功率参考值共同输入燃料电池控制单元FCU 11，根据功率跟随式能量控制策略判断燃料电池的工作状态，进而完成下一工况片段的控制。在行驶过程中燃料电池公交车1不断通过车辆行驶通信系统上传各片段的工况信息，每完成一次行程之后，云端分析工作站7通过实时上传的工况信息进行增量学习训练。当原行驶路线由于施工改造造成道路工况前后发生变化，上传的工况信息也随之发生变化。在行程结束，云端分析工作站7利用增量学习模型8对新增数据进行增量学习训练，进而对模型参数进行更新，把训练后的模型参数下载至燃料电池整车控制器VCU 2。当车辆在相同突发工况下再次运行时，即可适应新环境的变化。
上述列出的一系列的详细说明仅仅是针对本发明的可行性实施方式的具体说明，它们并非用以限制本发明的保护范围，凡未脱离本发明技艺精神所作的等效实施方式或变更均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种燃料电池公交车实时全局优化智能控制系统，其特征在于：包括车辆行驶通信单元、车辆行驶信息预测分析单元、燃料电池整车控制单元以及车辆行驶信息采集单元，所述燃料电池整车控制单元通过车辆行驶通信单元与车辆行驶信息预测分析单元信号连接，所述燃料电池整车控制单元还与车辆行驶信息采集单元信号连接；所述车辆行驶信息预测分析单元获取预测模型参数，下载至燃料电池整车控制单元进行最优SOC预测模型更新；所述燃料电池整车控制单元控制燃料电池的工作状态。
  2. 根据权利要求1所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述燃料电池整车控制单元控制燃料电池的工作状态是根据驱动电机(18)的功率参考值和实时功率、车辆下一工况片段最优SOC参考轨迹预测值和电池组实时SOC值进行控制的。
  3. 根据权利要求2所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述电池组实时SOC是由电池管理系统BMS(19)实时检测。
  4. 根据权利要求2所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述下一工况片段最优SOC参考轨迹预测值是由燃料电池整车控制单元中的最优SOC预测模型模块(9)接收当前工况片段特征参数，由Y *＝min+f(X)(max-min)得到，其中f(X)是回归函数，max为样本数据的最大值，min为样本数据的最小值，Y *为下一工况片段最优SOC参考轨迹预测值。
  5. 根据权利要求4所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述驱动电机(18)的功率参考值是燃料电池整车控制单元中的MPC预测控制模块(10)根据下一工况片段最优SOC参考轨迹预测值计算得到。
  6. 根据权利要求2所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述驱动电机(18)的实时功率是由驱动电机实时功率计算模块(16)计算得到。
  7. 根据权利要求1所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述预测模型参数是由特征参数计算模块(22)计算出的特征参数传送至增量学习模型(8)，经过训练后得到；所述特征参数计算模块(22)接收速度信息接收模块(20)和动态规划模块(21)发送的车辆行驶工况信息和最优SOC参考轨迹。
  8. 根据权利要求7所述的燃料电池公交车实时全局优化智能控制系统，其特征在于：所述增量学习模型(8)对特征参数进行训练，得到回归模型SVM1和支持向量集SV1，对新增工况信息进行整合，并与支持向量集SV1合成新样本库数据，继续训练得到全新的回归模型SVM2和支持向量集SV2作为最终模型。
  9. 一种燃料电池公交车实时全局优化智能控制方法，其特征在于：包括以下步骤：
    步骤一，利用增量学习模型(8)对得到的工况信息和最优SOC参考轨迹进行训练，得到最优SOC预测模型参数，下载至最优SOC预测模型模块(9)进行最优SOC预测模型更新；
    步骤二，将当前工况片段特征参数输入到最优SOC预测模型模块(9)，输出下一工况片段最优SOC参考轨迹预测值；
    步骤三，将下一工况片段最优SOC参考轨迹预测值输入到MPC预测控制模块(10)中，得到驱动电机功率参考值；
    步骤四，燃料电池控制单元FCU(11)对SOC值与功率判断，控制燃料电池的工作状态；
    步骤五，整车控制器VCU(2)判断公交车是否完成一次行程，若行程未结束，则返回步骤二循环进行；若结束，则云端分析工作站(7)将更新的预测模型参数重新下载至整车控制器VCU(2)中进行最优SOC预测模型更新。
  10. 根据权利要求9所述的燃料电池公交车实时全局优化智能控制方法，其特征在于：所述步骤四具体为：判断电池组实时SOC值是否大于下一工况片段最优SOC参考轨迹预测值的最大值SOC max，若SOC＞SOC max，则判断实时功率P是否小于功率参考值P min，若P＜P min，则燃料电池控制单元FCU(11)关闭不再充电；如果P≥P min，则继续判断实时功率P是否大于功率参考值P max，若P＞P max，则燃料电池控制单元FCU(11)开始工作，当P min≤P≤P max，则燃料电池控制单元FCU(11)保持状态不变；当实时SOC＜SOC min或者SOC min≤SOC≤SOC max且P＞P max，则燃料电池控制单元FCU(11)使燃料电池电堆(13)开启并放电，否则保持状态不变。
PCT/CN2020/079244 2019-10-15 2020-03-13 一种燃料电池公交车实时全局优化智能控制系统及方法 WO2021073036A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CH01054/20A CH717533B1 (de) 2019-10-15 2020-03-13 Intelligentes Steuerungssystem und -Verfahren mit einer globalen Echtzeit-Optimierung für einen Brennstoffzellenbus.

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910977702.8 2019-10-15
CN201910977702.8A CN110852482B (zh) 2019-10-15 2019-10-15 一种燃料电池公交车实时全局优化智能控制系统及方法

Publications (1)

Publication Number Publication Date
WO2021073036A1 true WO2021073036A1 (zh) 2021-04-22

Family

ID=69596613

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/079244 WO2021073036A1 (zh) 2019-10-15 2020-03-13 一种燃料电池公交车实时全局优化智能控制系统及方法

Country Status (3)

Country Link
CN (1) CN110852482B (zh)
CH (1) CH717533B1 (zh)
WO (1) WO2021073036A1 (zh)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113486698A (zh) * 2021-04-30 2021-10-08 华中科技大学 一种氢燃料电池工作的识别预测方法、存储介质及系统
CN113569947A (zh) * 2021-07-27 2021-10-29 合肥阳光智维科技有限公司 电弧检测方法和系统
CN113733936A (zh) * 2021-08-18 2021-12-03 中车唐山机车车辆有限公司 一种混合驱动有轨电车的功率控制方法、装置及存储介质
CN113904354A (zh) * 2021-09-24 2022-01-07 国网安徽省电力有限公司经济技术研究院 氢燃料电池热电联供单元的控制方法、系统及设备
CN114179641A (zh) * 2021-12-29 2022-03-15 上海重塑能源科技有限公司 一种用于电动叉车的燃料电池复合电源系统
CN114228696A (zh) * 2021-12-10 2022-03-25 燕山大学 一种考虑制冷系统耗能的冷链混动车实时优化控制方法
CN114291067A (zh) * 2021-12-29 2022-04-08 山东大学 基于预测的混合动力汽车凸优化能量控制方法及系统
CN114500504A (zh) * 2022-01-19 2022-05-13 博雷顿科技有限公司 电机效率恶化的智能预警方法、计算机存储介质及车辆
CN114821854A (zh) * 2022-03-28 2022-07-29 中汽研汽车检验中心（天津）有限公司 一种工况切换对车辆油耗影响评估方法
CN114843561A (zh) * 2022-05-13 2022-08-02 中国第一汽车股份有限公司 燃料电池的控制方法及装置
CN115000471A (zh) * 2022-05-18 2022-09-02 西安交通大学 一种基于机器学习的燃料电池催化层预测-分析-优化方法
CN115587654A (zh) * 2022-10-17 2023-01-10 国网河北省电力有限公司邯郸供电分公司 作业车放电优化方法及系统、电子设备及可读存储介质
CN115632145A (zh) * 2022-10-26 2023-01-20 上海汉翱新能源科技有限公司 一种氢燃料电池供氢压力优化方法及系统
CN115649015A (zh) * 2022-10-19 2023-01-31 苏州市华昌能源科技有限公司 一种基于时段的车载燃料电池能量管理方法
CN115906622A (zh) * 2022-11-08 2023-04-04 杭州润氢科技有限公司 一种基于模型强化学习的燃料电池电动汽车能量控制策略
CN116522498A (zh) * 2023-04-28 2023-08-01 重庆大学 增程式电动车能耗和排放协同优化方法和增程式电动车控制方法
CN116702978A (zh) * 2023-06-07 2023-09-05 西安理工大学 考虑突发事件特征的电动汽车充电负荷预测方法及装置

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110852482B (zh) * 2019-10-15 2020-12-18 江苏大学 一种燃料电池公交车实时全局优化智能控制系统及方法
CN113428049B (zh) * 2021-08-26 2021-11-09 北京理工大学 一种考虑电池老化抑制的燃料电池混动汽车能量管理方法
CN113525179B (zh) * 2021-08-27 2024-01-12 潍柴动力股份有限公司 一种燃料电池的动态响应控制方法
CN114103971B (zh) * 2021-11-23 2023-10-10 北京理工大学 一种燃料电池汽车节能驾驶优化方法及装置
CN114714987B (zh) * 2022-03-08 2024-06-25 佛山仙湖实验室 一种燃料电池混合动力汽车的能量管理优化方法及系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107818377A (zh) * 2016-09-12 2018-03-20 法乐第(北京)网络科技有限公司 基于云计算平台的车辆全局优化控制方法、系统、车辆和云计算平台
CN108177648A (zh) * 2018-01-02 2018-06-19 北京理工大学 一种基于智能预测的插电式混合动力车辆的能量管理方法
CN108427985A (zh) * 2018-01-02 2018-08-21 北京理工大学 一种基于深度强化学习的插电式混合动力车辆能量管理方法
WO2018158965A1 (en) * 2017-03-01 2018-09-07 Nec Corporation Apparatus and method for prediction of alarm, irregularity and undesired mode
CN109927709A (zh) * 2017-12-15 2019-06-25 郑州宇通客车股份有限公司 一种车辆行驶路线工况确定方法、能量管理方法及系统
CN110309978A (zh) * 2019-07-09 2019-10-08 南京邮电大学 基于二次动态调整的神经网络光伏功率预测模型及方法
CN110852482A (zh) * 2019-10-15 2020-02-28 江苏大学 一种燃料电池公交车实时全局优化智能控制系统及方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107425534B (zh) * 2017-08-25 2020-07-31 电子科技大学 一种基于优化蓄电池充放电策略的微电网调度方法
CN108197726B (zh) * 2017-12-11 2021-11-09 浙江科技学院 一种基于改进进化算法的家庭能量数据优化方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107818377A (zh) * 2016-09-12 2018-03-20 法乐第(北京)网络科技有限公司 基于云计算平台的车辆全局优化控制方法、系统、车辆和云计算平台
WO2018158965A1 (en) * 2017-03-01 2018-09-07 Nec Corporation Apparatus and method for prediction of alarm, irregularity and undesired mode
CN109927709A (zh) * 2017-12-15 2019-06-25 郑州宇通客车股份有限公司 一种车辆行驶路线工况确定方法、能量管理方法及系统
CN108177648A (zh) * 2018-01-02 2018-06-19 北京理工大学 一种基于智能预测的插电式混合动力车辆的能量管理方法
CN108427985A (zh) * 2018-01-02 2018-08-21 北京理工大学 一种基于深度强化学习的插电式混合动力车辆能量管理方法
CN110309978A (zh) * 2019-07-09 2019-10-08 南京邮电大学 基于二次动态调整的神经网络光伏功率预测模型及方法
CN110852482A (zh) * 2019-10-15 2020-02-28 江苏大学 一种燃料电池公交车实时全局优化智能控制系统及方法

Cited By (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113486698B (zh) * 2021-04-30 2023-09-26 华中科技大学 一种氢燃料电池工作的识别预测方法、存储介质及系统
CN113486698A (zh) * 2021-04-30 2021-10-08 华中科技大学 一种氢燃料电池工作的识别预测方法、存储介质及系统
CN113569947A (zh) * 2021-07-27 2021-10-29 合肥阳光智维科技有限公司 电弧检测方法和系统
CN113733936A (zh) * 2021-08-18 2021-12-03 中车唐山机车车辆有限公司 一种混合驱动有轨电车的功率控制方法、装置及存储介质
CN113904354A (zh) * 2021-09-24 2022-01-07 国网安徽省电力有限公司经济技术研究院 氢燃料电池热电联供单元的控制方法、系统及设备
CN114228696B (zh) * 2021-12-10 2023-08-15 燕山大学 一种考虑制冷系统耗能的冷链混动车实时优化控制方法
CN114228696A (zh) * 2021-12-10 2022-03-25 燕山大学 一种考虑制冷系统耗能的冷链混动车实时优化控制方法
CN114291067A (zh) * 2021-12-29 2022-04-08 山东大学 基于预测的混合动力汽车凸优化能量控制方法及系统
CN114291067B (zh) * 2021-12-29 2024-04-02 山东大学 基于预测的混合动力汽车凸优化能量控制方法及系统
CN114179641B (zh) * 2021-12-29 2023-10-20 上海重塑能源科技有限公司 一种用于电动叉车的燃料电池复合电源系统
CN114179641A (zh) * 2021-12-29 2022-03-15 上海重塑能源科技有限公司 一种用于电动叉车的燃料电池复合电源系统
CN114500504A (zh) * 2022-01-19 2022-05-13 博雷顿科技有限公司 电机效率恶化的智能预警方法、计算机存储介质及车辆
CN114821854B (zh) * 2022-03-28 2023-06-09 中汽研汽车检验中心（天津）有限公司 一种工况切换对车辆油耗影响评估方法
CN114821854A (zh) * 2022-03-28 2022-07-29 中汽研汽车检验中心（天津）有限公司 一种工况切换对车辆油耗影响评估方法
CN114843561A (zh) * 2022-05-13 2022-08-02 中国第一汽车股份有限公司 燃料电池的控制方法及装置
CN115000471A (zh) * 2022-05-18 2022-09-02 西安交通大学 一种基于机器学习的燃料电池催化层预测-分析-优化方法
CN115587654A (zh) * 2022-10-17 2023-01-10 国网河北省电力有限公司邯郸供电分公司 作业车放电优化方法及系统、电子设备及可读存储介质
CN115587654B (zh) * 2022-10-17 2024-03-01 国网河北省电力有限公司邯郸供电分公司 作业车放电优化方法及系统、电子设备及可读存储介质
CN115649015B (zh) * 2022-10-19 2023-12-15 苏州市华昌能源科技有限公司 一种基于时段的车载燃料电池能量管理方法
CN115649015A (zh) * 2022-10-19 2023-01-31 苏州市华昌能源科技有限公司 一种基于时段的车载燃料电池能量管理方法
CN115632145A (zh) * 2022-10-26 2023-01-20 上海汉翱新能源科技有限公司 一种氢燃料电池供氢压力优化方法及系统
CN115632145B (zh) * 2022-10-26 2023-05-16 上海汉翱新能源科技有限公司 一种氢燃料电池供氢压力优化方法及系统
CN115906622A (zh) * 2022-11-08 2023-04-04 杭州润氢科技有限公司 一种基于模型强化学习的燃料电池电动汽车能量控制策略
CN116522498A (zh) * 2023-04-28 2023-08-01 重庆大学 增程式电动车能耗和排放协同优化方法和增程式电动车控制方法
CN116522498B (zh) * 2023-04-28 2024-02-02 重庆大学 增程式电动车能耗和排放协同优化方法和增程式电动车控制方法
CN116702978A (zh) * 2023-06-07 2023-09-05 西安理工大学 考虑突发事件特征的电动汽车充电负荷预测方法及装置
CN116702978B (zh) * 2023-06-07 2024-02-13 西安理工大学 考虑突发事件特征的电动汽车充电负荷预测方法及装置

Also Published As

Publication number Publication date
CH717533B1 (de) 2022-01-14
CN110852482B (zh) 2020-12-18
CN110852482A (zh) 2020-02-28

Similar Documents

Publication Publication Date Title
WO2021073036A1 (zh) 一种燃料电池公交车实时全局优化智能控制系统及方法
Xiong et al. Towards a smarter hybrid energy storage system based on battery and ultracapacitor-A critical review on topology and energy management
Li et al. Adaptive energy management strategy for fuel cell/battery hybrid vehicles using Pontryagin's Minimal Principle
Yuan et al. Intelligent energy management strategy based on hierarchical approximate global optimization for plug-in fuel cell hybrid electric vehicles
CN102729987B (zh) 一种混合动力公交车能量管理方法
Lin et al. Online recursive power management strategy based on the reinforcement learning algorithm with cosine similarity and a forgetting factor
Wu et al. Hierarchical predictive control for electric vehicles with hybrid energy storage system under vehicle-following scenarios
Fu et al. Energy management strategy for fuel cell/battery/ultracapacitor hybrid electric vehicles using deep reinforcement learning with action trimming
Liu et al. Energy management for hybrid electric vehicles based on imitation reinforcement learning
Lin et al. Real-time optimization strategy by using sequence quadratic programming with multivariate nonlinear regression for a fuel cell electric vehicle
CN105882648A (zh) 一种基于模糊逻辑算法的混合动力系统能量管理方法
CN113085860B (zh) 一种跟车环境下的燃料电池混合动力汽车的能量管理方法
Li et al. Energy management strategy for a series hybrid electric vehicle using improved deep Q-network learning algorithm with prioritized replay
Lin et al. Intelligent energy management strategy based on an improved reinforcement learning algorithm with exploration factor for a plug-in PHEV
Wang et al. Real-time energy management strategy for a plug-in hybrid electric bus considering the battery degradation
Hu et al. Energy management optimization method of plug-in hybrid-electric bus based on incremental learning
CN115588242B (zh) 一种基于物联网的储能电池测试系统及方法
Shi et al. Cloud-based dynamic programming for an electric city bus energy management considering real-time passenger load prediction
Zhou et al. Genetic algorithm-based parameter optimization of energy management strategy and its analysis for fuel cell hybrid electric vehicles
WO2022028257A1 (zh) 新能源车辆的能耗回收比预测方法、节能控制方法和系统
CN112837532B (zh) 一种新能源公交协同调度与节能驾驶系统及其控制方法
CN115571113A (zh) 一种基于速度规划的混合动力车辆能量优化控制方法及系统
Ren et al. Battery longevity-conscious energy management predictive control strategy optimized by using deep reinforcement learning algorithm for a fuel cell hybrid electric vehicle
Romaus et al. Self-optimization energy management considering stochastic influences for a hybrid energy storage of an electric road vehicle
US11247571B2 (en) Intelligent energy management system for a vehicle and corresponding method

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20877446

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20877446

Country of ref document: EP

Kind code of ref document: A1