WO2021238136A1 - 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统 - Google Patents

一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统 Download PDF

Info

Publication number
WO2021238136A1
WO2021238136A1 PCT/CN2020/133724 CN2020133724W WO2021238136A1 WO 2021238136 A1 WO2021238136 A1 WO 2021238136A1 CN 2020133724 W CN2020133724 W CN 2020133724W WO 2021238136 A1 WO2021238136 A1 WO 2021238136A1
Authority
WO
WIPO (PCT)
Prior art keywords
hmm
model
heavy vehicle
heavy
vehicle
Prior art date
Application number
PCT/CN2020/133724
Other languages
English (en)
French (fr)
Inventor
朱天军
尹晓轩
蔡超明
梁建国
李伟豪
Original Assignee
肇庆学院
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 肇庆学院 filed Critical 肇庆学院
Priority to JP2020571534A priority Critical patent/JP7162919B2/ja
Priority to PCT/CN2020/133724 priority patent/WO2021238136A1/zh
Publication of WO2021238136A1 publication Critical patent/WO2021238136A1/zh

Links

Images

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
    • B60W30/02—Control of vehicle driving stability
    • B60W30/04—Control of vehicle driving stability related to roll-over prevention

Definitions

  • the invention relates to the technical field of vehicle state prediction, in particular to a heavy vehicle rollover warning method and system based on an HMM-RF hybrid model.
  • Heavy-duty vehicles have the characteristics of high center of gravity, large weight and volume, and too narrow wheelbase relative to the height of the center of mass. Therefore, their rollover stability limit is low and rollover accidents are extremely prone to occur.
  • UMTRI University of Michigan Transportation Research Center
  • UMTRI University of Michigan Transportation Research Center
  • the number of people who die from heavy vehicle rollover accidents each year has increased from 5,314 in 2016 to 5,537 in 2019.
  • a single car rollover accident may cause great losses, including casualties, property losses, and damage to public facilities such as roads and bridges, causing serious environmental pollution and more serious indirect consequences. . Therefore, the rollover of heavy vehicles has become an important issue affecting transportation safety.
  • the present invention aims to provide a heavy-duty vehicle based on the HMM-RF (HMM: Hidden Markov Model, Hidden Markov; RF: Random Forest) hybrid model. Vehicle rollover warning method and system.
  • HMM-RF Hidden Markov Model, Hidden Markov; RF: Random Forest
  • a heavy vehicle rollover warning method based on the HMM-RF hybrid model including:
  • Step 1 Collect heavy vehicle status data in real time
  • Step 2 According to the collected heavy vehicle state data, predict the state of the heavy vehicle based on the HMM-RF hybrid model, and obtain the prediction result of the heavy vehicle state;
  • Step 3 Determine whether there is a rollover risk based on the prediction result of the heavy vehicle status, and if so, send out the corresponding warning message.
  • the heavy vehicle status data includes the vehicle speed, lateral acceleration, and steering wheel angle signals of the heavy vehicle.
  • step 1 also includes:
  • the method also includes:
  • the HMM-RF hybrid model contains 6 HMM models and 1 RF model, of which 6 HMM models ⁇ HMM1, HMM2, HMM3, HMM4, HMM5, HMM6 ⁇ correspond to 6 driving conditions of heavy vehicles, of which driving conditions Including straight-line driving, normal left-turn driving, normal right-turn driving, left-turn rollover dangerous driving, right-turn rollover dangerous driving and alternate driving;
  • the multi-dimensional probability output of each HMM model is used as the input vector of the RF model, and the result of the RF model output is the prediction result of the heavy vehicle state;
  • the training process of the above-mentioned HMM-RF hybrid model includes:
  • the multi-dimensional probability output of the corresponding type of HMM model is used as the input vector of the random forest RF model for model training, and the final HMM-RF hybrid model is established;
  • step 3 also includes:
  • the heavy vehicle driving state at the future time is predicted, and the heavy vehicle driving state at the future time is judged whether there is a rollover risk, and if so, the corresponding warning message is issued.
  • step 3 also includes:
  • a heavy-duty vehicle rollover warning system based on the HMM-RF hybrid model, which includes: a data acquisition module, a processing module, and a warning module; among them,
  • the data acquisition module is used to collect the status data of heavy vehicles in real time
  • the processing module is used to predict the status of the heavy vehicle based on the HMM-RF hybrid model based on the collected heavy vehicle status data, and obtain the prediction result of the heavy vehicle status; judge whether there is a rollover risk according to the prediction result of the heavy vehicle status, and if so, control the warning module to issue a corresponding Warning information.
  • the heavy vehicle rollover warning method proposed in the present invention uses the HMM_RF hybrid model with rigorous mathematical structure and reliable calculation performance. It is not necessary to accurately know the vehicle load, center of gravity and other parameter information, that is, it can easily, accurately and quickly predict the side of the heavy vehicle. It is dangerous, and the model has good robustness and real-time performance.
  • the traditional vehicle early warning system is a static early warning system, which cannot dynamically and accurately predict the danger of an approaching rollover.
  • the dynamic early warning algorithm based on the HMM-RF hybrid model proposed by the present invention uses the current vehicle driving state and vehicle state conversion probability calculation to dynamically and accurately predict the occurrence of heavy vehicle rollover danger in a period of time in the future.
  • Figure 1 is a schematic diagram of a heavy vehicle rollover warning method according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of an estimation algorithm of a heavy-duty measurement rollover state estimation model according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of a construction process of an HMM-RF model according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of the algorithm framework of a hierarchical HMM-RF state prediction hybrid model according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of the structure of the driving state probability switching layer shown in FIG. 4;
  • Fig. 6 is a flow chart of a method for early warning of a heavy vehicle rollover according to an embodiment of the present invention.
  • Figure 1 shows a heavy vehicle rollover warning method based on the HMM-RF hybrid model, including:
  • Step 1 Collect heavy vehicle status data in real time.
  • the heavy vehicle state data includes the vehicle speed, lateral acceleration, and steering wheel angle signals of the heavy vehicle.
  • the above-mentioned heavy vehicle status data after real-time collection of the above-mentioned heavy vehicle status data, it also includes filtering processing of the acquired heavy vehicle status data, including signal conditioning and analog-to-digital conversion processing.
  • Step 2 According to the collected heavy vehicle state data, predict the heavy vehicle state based on the HMM-RF hybrid model, and obtain the heavy vehicle state prediction result.
  • the HMM-RF hybrid model contains 6 HMM models and 1 RF model, of which 6 HMM models ⁇ HMM1, HMM2, HMM3, HMM4, HMM5, HMM6 ⁇ correspond to 6 driving conditions of heavy vehicles (heavy vehicle status) ,
  • the driving conditions include straight driving, normal left-turn driving, normal right-turn driving, left-turn rollover dangerous driving, right-turn rollover dangerous driving and alternate driving;
  • the multi-dimensional probability output of the HMM model is used as the input vector of the RF model, and the output result of the RF model is the prediction result of the heavy vehicle state.
  • the training process of the above-mentioned HMM-RF hybrid model includes:
  • the multi-dimensional probability output of the corresponding type of HMM model is used as the input vector of the random forest RF model for model training, and the final HMM-RF hybrid model is established;
  • the forward comparison algorithm determines the HMM-RF model that can best describe the current vehicle state.
  • the corresponding driving conditions are the prediction results of the current heavy vehicle state.
  • the invention uses the built dynamic model to focus on the dynamic stability of vehicle motion and the factors affecting rollover stability.
  • certain components of the vehicle in real vehicles, certain components of the vehicle (suspension and tires) often show strong nonlinear characteristics, and have a great impact on the dynamic characteristics of the handling and stability of heavy vehicles.
  • the RF random forest algorithm and HMM hidden Markov pattern recognition technology are used to accurately establish the heavy vehicle rollover warning model.
  • the estimation algorithm of the heavy-duty measurement rollover state estimation model is shown in Figure 2.
  • the algorithm contains six HMM models corresponding to the driving conditions of heavy vehicles. Each HMM model in turn corresponds to heavy vehicles driving in a straight line, normal left-turn driving, normal right-turn driving, left-turning dangerous driving, and right-turning dangerous driving. There are six working conditions of driving and transitional driving. Among them, the alternate driving condition is the alternate operating condition between the vehicle traveling in a straight line, turning left and turning right.
  • the input of the algorithm is a real-time observation sequence, and the HMM model that can best describe the current vehicle state is determined by the Forward Algorithm forward comparison algorithm, that is, the current vehicle state is estimated.
  • the process of establishing the HMM-RF model is as follows: (1) Using the sliding window method to divide the entire time series into time series segments. In view of the correspondence between the time series period and each state, according to the boundary value of each observation value sequence corresponding to each state, the membership function is applied to fuzz each observation value sequence, and then each observation value sequence is comprehensively considered to discretize the time series period; (2) After repeated iterations, until the HMM model converges or reaches the maximum number of iterations, at the same time, according to the set maximum order of the autoregressive model, use the AR autoregressive model to calculate the predicted observable time series of the HMM model; (3) Apply Viterbi decoding The algorithm makes the trend prediction of the vehicle state; (4) The vehicle state corresponding to the predicted output of the HMM model of each order is compared with the actual vehicle measurement state, and the optimal autoregressive model order is finally determined.
  • the establishment process of the N-order AR-HMM rollover estimation model is realized.
  • the multi-dimensional probability output of the original calculation of the maximum likelihood value is used as the input vector of the random forest RF model for model training.
  • the estimation accuracy of the model is verified by online real vehicle test data under different vehicle speeds, lateral accelerations and various steering inputs, which lays the foundation for the heavy-duty vehicle dynamic prediction algorithm.
  • Step 3 Determine whether there is a rollover risk based on the prediction result of the heavy vehicle status, and if so, send out the corresponding warning message.
  • step 3 also includes:
  • the heavy vehicle driving state at the future time is predicted, and the heavy vehicle driving state at the future time is judged whether there is a rollover risk, and if so, the corresponding warning message is issued.
  • step 3 also includes:
  • this application also proposes a heavy vehicle state prediction algorithm based on the HMM-RF hybrid model.
  • the algorithm framework includes the highest layer, a hidden layer, and a state observation layer;
  • the state observation layer includes sensor data collection such as lateral motion state (lateral acceleration), speed level (vehicle speed), steering motion state (steering wheel angle), etc.; the data is classified into observation sequences in vector form as a hidden layer The input observation sequence of the HMM model;
  • Each x node in the hidden layer is defined as an HMM model, which corresponds to a HMM model of a heavy vehicle state in the HMM-RF hybrid model.
  • Each HMM model inputs the output multi-dimensional probability to the highest layer, and the highest layer corresponds to the RF model
  • the RF model obtains the likelihood value of each driving state according to the multi-dimensional probability of each HMM model, and selects the state with the largest likelihood value as the predicted current vehicle driving state, and judges whether there is a heavy vehicle based on the predicted current vehicle driving state. Turn over the risk.
  • the highest level further calculates the probability of switching the vehicle driving state in the future based on the current vehicle state prediction result and the vehicle driving state at the previous moment, so as to accurately predict the vehicle driving state in the future.
  • the structure of the highest level predicting the driving state of the vehicle at the future time adopts the commonly used grid (Trellis) structure in the Viterbi algorithm.
  • the specific driving state probability switching layer structure is shown in Figure 5.
  • the heavy vehicle rollover state prediction algorithm based on the HMM-RF hybrid model uses a hierarchical structure to accurately predict the vehicle driving state in the future based on the current state of the vehicle estimated by the hidden layer and the driving state probability switching layer. So as to realize the dynamic prediction function of the heavy vehicle rollover hazard.
  • this application proposes an online warning method for heavy vehicle rollover as shown in FIG. 6.
  • the preprocessed (filtered) signals of vehicle speed, steering wheel angle and lateral acceleration signals collected in real time according to vehicle sensor signals are passed through the heavy vehicle rollover state estimation and prediction module, and then online according to the trained HMM-RF hybrid model and Viterbi decoding algorithm Predict the state corresponding to the observation sequence, if it is a rollover state, trigger the heavy vehicle early warning device.
  • a rollover early warning threshold value of X seconds (corresponding to a larger order of the autoregressive model, that is, a multi-order autoregressive model). If you are predicting the future X Within seconds, if the driving state of the heavy vehicle is always in a normal state, the calculation will be stopped for a period of time (for example, X seconds), and the warning will continue after this time has passed.
  • X is 2 seconds.
  • the heavy-duty vehicle rollover warning algorithm based on the HMM-RF hybrid model is not affected by changes in vehicle parameters (weight, center of mass position, and moment of inertia). According to the on-board sensor signal and the trained HMM-RF hybrid model, the probability is used The statistical method accurately and dynamically calculates the vehicle rollover status. At the same time, the early warning algorithm has good real-time performance, robustness and strong anti-interference. It is very suitable for heavy vehicle rollover warning applications and helps to improve the safety of heavy vehicle driving.
  • the present invention also proposes a heavy vehicle rollover warning system based on the HMM-RF hybrid model, which includes: a data acquisition module, a processing module, and a warning module; among them,
  • the data acquisition module is used to collect the status data of heavy vehicles in real time
  • the processing module is used to predict the status of the heavy vehicle based on the HMM-RF hybrid model based on the collected heavy vehicle status data, and obtain the prediction result of the heavy vehicle status; judge whether there is a rollover risk according to the prediction result of the heavy vehicle status, and if so, control the warning module to issue a corresponding Warning information.
  • the data collection module includes lateral acceleration sensors, steering wheel sensors, and vehicle speed sensors installed on heavy vehicles.
  • the data collection module and the processing module implement data interaction through a signal transmission module, and the vehicle's driving status is collected in real time through sensors.
  • the processing module is mainly used to process the received heavy vehicle status data. According to the real-time vehicle status, run the rollover warning method based on the HMM-RF hybrid model, predict the upcoming rollover hazard in the current state, and pass the warning The module (warning light and buzzer) warns the driver to correct improper operation in time, so as to avoid the occurrence of vehicle rollover accidents.
  • the processing module is also used to construct a HMM-RF hybrid model.
  • the heavy vehicle rollover warning system proposed above can also be used to implement the steps in the heavy vehicle rollover warning method proposed above and the specific implementations corresponding to each step. This application will not be repeated here.
  • each functional unit/module in each embodiment of the present invention may be integrated into one processing unit/module, or each unit/module may exist alone physically, or it may be two or more units/modules.
  • the module is integrated in a unit/module.
  • the above-mentioned integrated unit/module can be implemented in the form of hardware or software functional unit/module.
  • the embodiments described herein can be implemented by hardware, software, firmware, middleware, code, or any appropriate combination thereof.
  • the processor can be implemented in one or more of the following units: application specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field Programmable gate array (FPGA), processor, controller, microcontroller, microprocessor, other electronic unit designed to implement the functions described herein, or a combination thereof.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • DSPD digital signal processing device
  • PLD programmable logic device
  • FPGA field Programmable gate array
  • processor controller, microcontroller, microprocessor, other electronic unit designed to implement the functions described herein, or a combination thereof.
  • part or all of the procedures of the embodiments can be completed by computer programs instructing relevant hardware.
  • the above-mentioned program may be stored in a computer-readable medium or transmitted as one or more instructions or codes on the computer-readable medium.
  • the computer-readable medium includes a computer storage medium and a communication medium, where the communication medium includes any medium that facilitates the transfer of a computer program from one place to another.
  • the storage medium may be any available medium that can be accessed by a computer.
  • the computer-readable medium may include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or can be used to carry or store desired program codes in the form of instructions or data structures And any other medium that can be accessed by the computer.

Landscapes

  • Engineering & Computer Science (AREA)
  • Automation & Control Theory (AREA)
  • Transportation (AREA)
  • Mechanical Engineering (AREA)
  • Traffic Control Systems (AREA)
  • Electric Propulsion And Braking For Vehicles (AREA)
  • Emergency Alarm Devices (AREA)
  • Control Of Driving Devices And Active Controlling Of Vehicle (AREA)

Abstract

本发明提供一种基于HMM-RF混合模型的重型车辆侧翻预警方法和系统。所述一种基于HMM-RF混合模型的重型车辆侧翻预警方法包括：实时采集重型车辆状态数据；根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；根据重型车辆状态预测结果判断是否存在侧翻风险，若是则发出相应的预警信息。本发明能够方便、精确、迅捷的预测重型车辆侧翻危险，有助于提高重型车辆驾驶的安全性。

Description

一种基于HMM-RF混合模型的重型车辆侧翻预警方法和系统 技术领域
本发明涉及车辆状态预测技术领域，特别是一种基于HMM-RF混合模型的重型车辆侧翻预警方法和系统。
背景技术
随着国民经济连续多年的高速发展，尤其是国家对基础设施建设投入的逐年加大，使得各种重型车辆为载体的道路运输业呈现爆发式发展。重型车辆由于具有运输效率高、运输成本低的特点，逐渐成为道路交通运输的宠儿。据预测，到2020年我国重型车辆产销量将达到200万辆，销售额将达2300亿元，将成为世界上重型载车辆市场成长最快的国家，重型车辆的生产与开发成为国内车辆生产厂家竞争的焦点。汽车产业核心技术的进步和人民生活水平的进一步提高，使重型车辆用户对车辆的性能水平要求越来越高，如何保证高速行驶重型车辆行驶安全性和操纵稳定性，已成为重型车辆产品研发机构所关注的重要课题。
重型车辆具有重心位置高、重量和体积大、轮距相对于质心高度过窄等特点，因此其侧翻稳定极限低，极易发生侧翻事故。据美国密歇根大学交通研究中心(UMTRI)统计,2016年至2019年,美国每年各类重型车辆的侧翻事故平均有6900起。同时每年死于重型车辆侧翻事故的人数，也从2016年的5314人增加到2019年的5537人。另外，一个单独的汽车侧翻事故可能会导致极大的损失，其中除包括人员伤亡，财产损失，还可能对道路桥梁等公共设施的破坏，造成环境的严重污染，产生更为严重的间接后果。因此，重型车辆侧翻已经成为影响交通运输安全的重要问题。
随着我国进入WTO，国外各大汽车公司竞相将众多高技术含量的重型车辆产品投入中国市场，重型车辆市场已经从以价格为主导逐步转变为以性能品质取胜。目前尽管我国已成为重型车辆生产大国，也掌握了一定的自主开发技术。然而，高档次产品和关键技术仍然依赖国外技术，特别是在重型车辆安全性能方面，由于缺乏主动安全装置，交通事故频发，带来巨大的经济损失和人员伤亡，2019年，我国重型车辆事故造成32104人死亡，是欧美发达国家的五倍以上。为了尽快加强我国重型车辆主动安全产品的自主开发能力，防止汽车产业核心技术空心 化势在必行。虽然目前也有一些针对车辆的预警控制方法研究，但是目前的研究重点通常在车辆超重和超速上的预警，而针对重型车辆侧翻的预测预警研究依然存在空白。
发明内容
针对上述提出目前重型车辆侧翻预警技术存在欠缺的问题，本发明旨在提供一种基于HMM-RF(HMM：Hidden Markov Model，隐马尔科夫；RF：Random Forest，随机森林)混合模型的重型车辆侧翻预警方法和系统。
本发明的目的采用以下技术方案来实现：
第一方面，提出一种基于HMM-RF混合模型的重型车辆侧翻预警方法，包括：
步骤1：实时采集重型车辆状态数据；
步骤2：根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；
步骤3：根据重型车辆状态预测结果判断是否存在侧翻风险，若是则发出相应的预警信息。
进一步，重型车辆状态数据包括重型车辆的车速、侧向加速度和方向盘转角信号。
进一步，步骤1还包括：
对获取的重型车辆状态数据进行滤波处理，包括信号调理和模数转换处理。
进一步，该方法还包括：
构建HMM-RF混合模型，包括：
其中HMM-RF混合模型包含6个HMM模型和1个RF模型，其中6个HMM模型{HMM1,HMM2,HMM3,HMM4,HMM5,HMM6}分别对应重型车辆的6个行驶工况，其中行驶工况包括直线行驶、正常左转弯行驶、正常右转弯行驶、左转弯侧翻危险行驶、右转弯侧翻危险行驶和交替行驶；
车辆状态数据分别输入到6个HMM模型后，将各HMM模型的多维概率输出作为RF模型的输入矢量，RF模型输出的结果即为重型车辆状态预测结果；
其中上述HMM-RF混合模型的训练过程包括：
1)采用滑动窗方法将训练数据从时间序分割成时间段序列；采用隶属函数把训练数据中各个观测值序列模糊化，然后综合考虑各个观测值序列，使时间序列段离散化；
2)将训练数据分别输入到各个HMM模型中对HMM模型进行迭代训练，直到HMM模型收敛或者达到最大迭代次数，根据设置的自回归模型的最大阶数，利用AR自回归模型计算HMM模型的预测可观测时间序列；
3)应用Viterbi解码算法做出车辆状态的趋势预测；
4)把对应于各阶HMM模型预测输出的车辆状态和实际车辆测量状态进行比较，最终确定最优的自回归模型阶数N，建立N阶HMM侧翻估计模型；
5)经过训练得到的各类型的HMM模型后，将训练数据经对应类型的HMM模型的多维概率输出作为随机森林RF模型的输入矢量进行模型训练，建立最终的HMM-RF混合模型；
进一步，步骤3还包括：
根据获取的重型车辆状态预测结果和前一时刻重型车辆状态预测未来时刻的重型车辆行驶状态，根据未来时刻的重型车辆行驶状态判断是否存在侧翻风险，若是则发出相应的预警信息。
进一步，步骤3还包括：
当预测到未来时刻的重型车辆行驶状态不存在侧翻风险时，则停止对采集的重型车辆状态数据进行重型车辆状态预测处理，并到达该未来时刻时重新开始对采集的重型车辆状态数据进行重型车辆状态预测处理。
第二方面，提出一种基于HMM-RF混合模型的重型车辆侧翻预警系统，包括：数据采集模块、处理模块和警示模块；其中，
数据采集模块用于实时采集重型车辆状态数据；
处理模块用于根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；根据重型车辆状态预测结果判断是否存在侧翻风险，若是则控制警示模块发出相应的预警信息。
本发明的有益效果为：
1)本发明提出的重型车辆侧翻预警方法，使用具有严谨数学结构和可靠计 算性能的HMM_RF混合模型，不必精确获知车辆载荷、重心等参数信息，即可以方便、精确、迅捷的预测重型车辆侧翻危险，同时该模型具有良好的鲁棒性和实时性。
2)针对传统车辆预警系统属于静态预警系统，不能动态精确预测迫近的侧翻危险程度。本发明提出的基于HMM-RF混合模型的动态预警算法，该算法利用当前时刻车辆行驶状态和车辆状态转化概率计算可以实时动态精确预测未来一段时间内重型车辆侧翻危险的发生。
附图说明
利用附图对本发明作进一步说明，但附图中的实施例不构成对本发明的任何限制，对于本领域的普通技术人员，在不付出创造性劳动的前提下，还可以根据以下附图获得其它的附图。
图1为本发明一种实施例所示的重型车辆侧翻预警方法示意图；
图2为本发明一种实施例所示的重型测量侧翻状态估计模型的估计算法示意图；
图3为本发明一种实施例所示的HMM-RF模型搭建流程示意图；
图4为本发明一种实施例所示的分层HMM-RF状态预测混合模型的算法框架示意图；
图5为图4所示的行驶状态概率切换层结构示意图；
图6为本发明一种实施例所示的重型车辆侧翻预警方法流程图。
具体实施方式
结合以下应用场景对本发明作进一步描述。
参见图1，其示出一种基于HMM-RF混合模型的重型车辆侧翻预警方法，包括：
步骤1：实时采集重型车辆状态数据。
优选的，重型车辆状态数据包括重型车辆的车速、侧向加速度和方向盘转角信号。
通过传感器实时采集重型车辆的车速、侧向加速度和方向盘转角信号。
其中，在实时采集上述重型车辆状态数据后，还包括对获取的重型车辆状态数据进行滤波处理，包括信号调理和模数转换处理。
步骤2：根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果。
其中HMM-RF混合模型包含6个HMM模型和1个RF模型，其中6个HMM模型{HMM1,HMM2,HMM3,HMM4,HMM5,HMM6}分别对应重型车辆的6个行驶工况(重型车辆状态)，其中行驶工况包括直线行驶、正常左转弯行驶、正常右转弯行驶、左转弯侧翻危险行驶、右转弯侧翻危险行驶和交替行驶；
车辆状态数据分别输入到6个HMM模型后，将HMM模型的多维概率输出作为RF模型的输入矢量，RF模型输出的结果即为重型车辆状态预测结果。
其中上述HMM-RF混合模型的训练过程包括：
1)采用滑动窗方法将训练数据从时间序分割成时间段序列；采用隶属函数把训练数据中各个观测值序列模糊化，然后综合考虑各个观测值序列，使时间序列段离散化；
2)将训练数据分别输入到各个HMM模型中对HMM模型进行迭代训练，直到HMM模型收敛或者达到最大迭代次数，根据设置的自回归模型的最大阶数，利用AR自回归模型计算HMM模型的预测可观测时间序列；
3)应用Viterbi解码算法做出车辆状态的趋势预测；
4)把对应于各阶HMM模型预测输出的车辆状态和实际车辆测量状态进行比较，最终确定最优的自回归模型阶数N，建立N阶HMM侧翻估计模型；
5)经过训练得到的各类型的HMM模型后，将训练数据经对应类型的HMM模型的多维概率输出作为随机森林RF模型的输入矢量进行模型训练，建立最终的HMM-RF混合模型；
将重型车辆状态数据的实时观测序列输入到HMM-RF混合模型，经过前向对比算法判定能最好描述当前的车辆状态对应的HMM模型，其对应的行驶工况即为出当前的重型车辆侧翻状态。
分别输入到6个HMM模型中，经过前向对比算法判定能最好描述当前的车辆状态的HMM-RF模型，其对应的行驶工况即为出当前的重型车辆状态预测结 果。
其中，针对重型车辆的载荷、质心位置、转动惯量等变化范围大等特点，发明利用所建动力学模型重点研究车辆运动动力学稳定性，以及侧翻稳定性影响因素。在实车上，车辆某些组成部分(悬架和轮胎)往往显示很强的非线性特性，而且对重型车辆操纵稳定性的动态特性影响很大。为了使建立的HMM-RF混合模型能够精确的描述车辆稳态和动态过程，选用RF随机森林算法和HMM隐Markov模式识别技术来实现精确建立重型车辆侧翻预警模型的工作。其中，重型测量侧翻状态估计模型的估计算法如图2所示。
该算法中包含与重型车辆行驶工况对应的六个HMM模型，每个HMM模型依次对应重型车辆直线行驶、正常左转弯行驶、正常右转弯行驶、左转弯侧翻危险行驶、右转弯侧翻危险行驶和过渡行驶六种工况。其中交替行驶工况为车辆直线行驶、左转弯、右转弯之间的交替工况。算法输入为实时的观测序列，经过Forward Algorithm前向对比算法判定能最好描述当前的车辆状态的HMM模型，即估计出当前的车辆状态。
其中HMM-RF混合模型搭建流程如图3所示：
HMM-RF模型建立过程如下：(1)采用滑动窗方法将整个时间序列划分成时间序列段。鉴于时间序列段和各个状态的对应关系，根据各个状态对应的各个观察值序列的分界值，应用隶属函数把各个观测值序列模糊化，然后综合考虑各个观察值序列，使时间序列段离散化；(2)经过反复迭代，直至HMM模型收敛或达到最大迭代次数，同时根据设置的自回归模型的最大阶数，利用AR自回归模型计算HMM模型的预测可观测时间序列；(3)应用Viterbi解码算法做出车辆状态的趋势预测；(4)把对应于各阶HMM模型预测输出的车辆状态和实际车辆测量状态进行比较，最终确定最优的自回归模型阶数。这样就实现了N阶AR-HMM侧翻估计模型建立过程。(5)经过训练得到的对应类型的HMM模型后，将原本计算最大似然值的多维概率输出作为随机森林RF模型的输入矢量进行模型训练。建立HMM-RF混合模型后，通过不同车速、侧向加速度和各种转向输入下的在线实车试验数据来验证该模型的估计精度，为重型车辆动态预测算法打下基础。
步骤3：根据重型车辆状态预测结果判断是否存在侧翻风险，若是则发出相应的预警信息。
优选的，步骤3还包括：
根据获取的重型车辆状态预测结果和前一时刻重型车辆状态预测未来时刻的重型车辆行驶状态，根据未来时刻的重型车辆行驶状态判断是否存在侧翻风险，若是则发出相应的预警信息。
优选的，步骤3还包括：
当预测到未来时刻的重型车辆行驶状态不存在侧翻风险时，则停止对采集的重型车辆状态数据进行分析，并到达该未来时刻时重新开始对采集的重型车辆状态数据进行分析。
基于本申请提出的HMM-RF混合模型，本申请还提出一种基于HMM-RF混合模型的重型车辆状态预测算法，该算法框架包含最高层、隐含层和状态观测层；
其中状态观测层包括侧向运动状态(侧向加速度)、速度等级(车速)、转向运动状态(方向盘转角)等传感器数据采集；并将数据进行分类处理为向量形式的观察序列，作为隐含层的HMM模型的输入观测序列；
其中隐含层中每个x节点定义为一个HMM模型，与HMM-RF混合模型中的一个重型车辆状态的HMM模型对应，各HMM模型将输出的多维概率输入到最高层，最高层对应RF模型，RF模型根据各HMM模型的多维概率获取每个行驶状态的似然值，并选取似然值最大的状态作为预测的当前车辆行驶状态，并根据预测的当前车辆行驶状态判断重型车辆是否存在侧翻风险。
同时最高层还进一步根据当前的车辆状态预测结果和前一时刻车辆行驶状态计算未来时刻车辆行驶状态切换的概率，从而精确预测未来时刻车辆行驶状态。
其中最高层预测未来时刻车辆行驶状态的结构采用Viterbi算法中常用的网格(Trellis)结构体形式，具体行驶状态概率切换层结构如图5所示.
通过上述分析可知：基于HMM-RF混合模型的重型车辆侧翻状态预测算法利用分层结构，依据隐含层估计出来的当前时刻车辆状态和行驶状态概率切换层精确预测出未来时刻车辆行驶状态，从而实现重型车辆侧翻危险动态预测功能。
基于上述方法，本申请提出一种重型车辆侧翻在线预警方法如图6所示。
根据车辆传感器信号实时采集的车速、方向盘转角和侧向加速度信号预处理后(滤波)信号，经过重型车辆侧翻状态估计和预测模块，然后根据训练好的HMM-RF混合模型以及Viterbi解码算法在线预测观测序列所对应的状态，假如为侧翻状态，触发重型车辆预警装置。
为了减小模型计算量，保证预警的实时性，需要设定一个侧翻预警门槛值为X秒(对应自回归模型的阶数较大，即多阶自回归模型)，如果在预测未来的X秒内，重型车辆的行驶状态一直是正常态，则停止计算一段时间(例如X秒)，超过该时间后继续预警。
一种场景中，X为2秒。
由上可知，基于HMM-RF混合模型的重型车辆侧翻预警算法不受车辆参数(重量、质心位置和转动惯量)变化的影响，根据车载传感器信号及训练好的HMM-RF混合模型，利用概率统计方法精确动态计算得到车辆侧翻状态，同时该预警算法的实时性较好、鲁棒性及抗干扰性强，十分适合重型车辆侧翻预警应用，有助于提高重型车辆驾驶的安全性。
基于上述提出的重型车辆侧翻预警方法，本发明还提出一种基于HMM-RF混合模型的重型车辆侧翻预警系统，包括：数据采集模块、处理模块和警示模块；其中，
数据采集模块用于实时采集重型车辆状态数据；
处理模块用于根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；根据重型车辆状态预测结果判断是否存在侧翻风险，若是则控制警示模块发出相应的预警信息。
其中，数据采集模块包括设置在重型车辆上的侧向加速度传感器、方向盘传感器和车速传感器，其中数据采集模块与处理模块之间通过信号传输模块实现数据交互，通过传感器实时采集车辆的行驶状态。
处理模块主要用于对接收到的重型车辆状态数据进行处理，根据实时的车辆状态，运行基于HMM-RF混合模型的侧翻预警方法，在当前状态下预测即将发生的侧翻危险，并通过警示模块(警示灯和蜂鸣器)警示驾驶员及时修正不当操 作，从而避免车辆侧翻事故发生。
优选的，处理模块还用于构建HMM-RF混合模型。
其中，上述提出的重型车辆侧翻预警系统还能够用于实现上述提出的重型车辆侧翻预警方法中各步骤以及与各步骤相应的具体实施方式。本申请在此不再重复叙述。
需要说明的是，在本发明各个实施例中的各功能单元/模块可以集成在一个处理单元/模块中，也可以是各个单元/模块单独物理存在，也可以是两个或两个以上单元/模块集成在一个单元/模块中。上述集成的单元/模块既可以采用硬件的形式实现，也可以采用软件功能单元/模块的形式实现。
通过以上的实施方式的描述，所属领域的技术人员可以清楚地了解应当理解，可以以硬件、软件、固件、中间件、代码或其任何恰当组合来实现这里描述的实施例。对于硬件实现，处理器可以在一个或多个下列单元中实现：专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、处理器、控制器、微控制器、微处理器、设计用于实现这里所描述功能的其他电子单元或其组合。对于软件实现，实施例的部分或全部流程可以通过计算机程序来指令相关的硬件来完成。实现时，可以将上述程序存储在计算机可读介质中或作为计算机可读介质上的一个或多个指令或代码进行传输。计算机可读介质包括计算机存储介质和通信介质，其中通信介质包括便于从一个地方向另一个地方传送计算机程序的任何介质。存储介质可以是计算机能够存取的任何可用介质。计算机可读介质可以包括但不限于RAM、ROM、EEPROM、CD-ROM或其他光盘存储、磁盘存储介质或者其他磁存储设备、或者能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质。
最后应当说明的是，以上实施例仅用以说明本发明的技术方案，而非对本发明保护范围的限制，尽管参照较佳实施例对本发明作了详细地说明，本领域的普通技术人员应当分析，可以对本发明的技术方案进行修改或者等同替换，而不脱离本发明技术方案的实质和范围。

Claims (7)

  1. 一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，包括：
    步骤1：实时采集重型车辆状态数据；
    步骤2：根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；
    步骤3：根据重型车辆状态预测结果判断是否存在侧翻风险，若是则发出相应的预警信息。
  2. 根据权利要求1所述的一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，所述重型车辆状态数据包括重型车辆的车速、侧向加速度和方向盘转角信号。
  3. 根据权利要求2所述的一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，所述步骤1还包括：
    对获取的所述重型车辆状态数据进行滤波处理，包括信号调理和模数转换处理。
  4. 根据权利要求1所述的一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，还包括：
    构建所述HMM-RF混合模型，
    HMM-RF混合模型包含6个HMM模型和1个RF模型，其中6个HMM模型分别对应所述重型车辆的6个行驶工况，其中行驶工况包括直线行驶、正常左转弯行驶、正常右转弯行驶、左转弯侧翻危险行驶、右转弯侧翻危险行驶和交替行驶；
    所述车辆状态数据分别输入到6个HMM模型后，将各HMM模型的多维概率输出作为RF模型的输入矢量，RF模型输出的结果即为所述重型车辆状态预测结果；
    其中所述HMM-RF混合模型的训练过程包括：
    1)采用滑动窗方法将训练数据从时间序分割成时间段序列；采用隶属函数把训练数据中各个观测值序列模糊化，然后综合考虑各个观测值序列，使时间序列段离散化；
    2)将训练数据分别输入到各个HMM模型中对HMM模型进行迭代训练， 直到HMM模型收敛或者达到最大迭代次数，根据设置的自回归模型的最大阶数，利用AR自回归模型计算HMM模型的预测可观测时间序列；
    3)应用Viterbi解码算法做出车辆状态的趋势预测；
    4)把对应于各阶HMM模型预测输出的车辆状态和实际车辆测量状态进行比较，最终确定最优的自回归模型阶数N，建立N阶HMM侧翻估计模型；
    5)经过训练得到的各类型的HMM模型后，将训练数据经对应类型的HMM模型的多维概率输出作为随机森林RF模型的输入矢量进行模型训练，建立最终的HMM-RF混合模型。
  5. 根据权利要求4所述的一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，所述步骤3还包括：
    根据获取的所述重型车辆状态预测结果和前一时刻重型车辆状态，预测未来时刻的重型车辆行驶状态，根据未来时刻的重型车辆行驶状态判断是否存在侧翻风险，若是则发出相应的预警信息。
  6. 根据权利要求5所述的一种基于HMM-RF混合模型的重型车辆侧翻预警方法，其特征在于，所述步骤3还包括：
    当预测到所述未来时刻的重型车辆行驶状态不存在侧翻风险时，则停止对采集的所述重型车辆状态数据进行重型车辆状态预测处理，并到达该未来时刻时重新开始对采集的重型车辆状态数据进行重型车辆状态预测处理。
  7. 一种基于HMM-RF混合模型的重型车辆侧翻预警系统，其特征在于，包括：数据采集模块、处理模块和警示模块；其中，
    数据采集模块用于实时采集重型车辆状态数据；
    处理模块用于根据采集的重型车辆状态数据，基于HMM-RF混合模型预测重型车辆状态，获取重型车辆状态预测结果；根据重型车辆状态预测结果判断是否存在侧翻风险，若是则控制警示模块发出相应的预警信息。
PCT/CN2020/133724 2020-12-04 2020-12-04 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统 WO2021238136A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2020571534A JP7162919B2 (ja) 2020-12-04 2020-12-04 Ｈｍｍ－ｒｆハイブリッドモデルに基づく大型車両横転警告方法
PCT/CN2020/133724 WO2021238136A1 (zh) 2020-12-04 2020-12-04 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/133724 WO2021238136A1 (zh) 2020-12-04 2020-12-04 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统

Publications (1)

Publication Number Publication Date
WO2021238136A1 true WO2021238136A1 (zh) 2021-12-02

Family

ID=78745552

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/133724 WO2021238136A1 (zh) 2020-12-04 2020-12-04 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统

Country Status (2)

Country Link
JP (1) JP7162919B2 (zh)
WO (1) WO2021238136A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114379540A (zh) * 2022-02-21 2022-04-22 东南大学 考虑前方障碍物影响的大型营运车辆防侧翻驾驶决策方法
CN115320307A (zh) * 2022-10-17 2022-11-11 浙江孔辉汽车科技有限公司 一种车辆的减振方法、减振装置、电子设备及存储介质
CN115938128A (zh) * 2023-03-15 2023-04-07 天津所托瑞安汽车科技有限公司 交通事故预测方法、装置、终端及存储介质
CN117541066A (zh) * 2024-01-09 2024-02-09 巨野金牛车业有限公司 一种自卸半挂车侧翻评价指标及评价方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004262316A (ja) * 2003-02-28 2004-09-24 Calsonic Kansei Corp 車両横転検出装置及びこれを用いた車両乗員保護システム
KR20080086649A (ko) * 2007-03-23 2008-09-26 주식회사 만도 차량 안정성 제어 시스템에서의 차량 전복 상태 감지방법
CN104401323A (zh) * 2014-11-04 2015-03-11 河北工程大学 一种重型车辆侧翻预警方法及装置
CN104951764A (zh) * 2015-06-17 2015-09-30 浙江工业大学 基于二次谱聚类和hmm-rf混合模型的高速车辆行为识别方法
CN109664890A (zh) * 2018-12-24 2019-04-23 河北工程大学 基于svm模型的汽车侧翻预警算法
CN111094099A (zh) * 2017-09-28 2020-05-01 大陆-特韦斯贸易合伙股份公司及两合公司 用于求得车辆的重心的位置的方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH04191181A (ja) * 1990-11-26 1992-07-09 Hino Motors Ltd 車両用横転警報装置
JP2009113717A (ja) 2007-11-08 2009-05-28 Fuji Heavy Ind Ltd 状態推定システム

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004262316A (ja) * 2003-02-28 2004-09-24 Calsonic Kansei Corp 車両横転検出装置及びこれを用いた車両乗員保護システム
KR20080086649A (ko) * 2007-03-23 2008-09-26 주식회사 만도 차량 안정성 제어 시스템에서의 차량 전복 상태 감지방법
CN104401323A (zh) * 2014-11-04 2015-03-11 河北工程大学 一种重型车辆侧翻预警方法及装置
CN104951764A (zh) * 2015-06-17 2015-09-30 浙江工业大学 基于二次谱聚类和hmm-rf混合模型的高速车辆行为识别方法
CN111094099A (zh) * 2017-09-28 2020-05-01 大陆-特韦斯贸易合伙股份公司及两合公司 用于求得车辆的重心的位置的方法
CN109664890A (zh) * 2018-12-24 2019-04-23 河北工程大学 基于svm模型的汽车侧翻预警算法

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114379540A (zh) * 2022-02-21 2022-04-22 东南大学 考虑前方障碍物影响的大型营运车辆防侧翻驾驶决策方法
CN114379540B (zh) * 2022-02-21 2024-04-30 东南大学 考虑前方障碍物影响的大型营运车辆防侧翻驾驶决策方法
CN115320307A (zh) * 2022-10-17 2022-11-11 浙江孔辉汽车科技有限公司 一种车辆的减振方法、减振装置、电子设备及存储介质
CN115320307B (zh) * 2022-10-17 2023-01-20 浙江孔辉汽车科技有限公司 一种车辆的减振方法、减振装置、电子设备及存储介质
CN115938128A (zh) * 2023-03-15 2023-04-07 天津所托瑞安汽车科技有限公司 交通事故预测方法、装置、终端及存储介质
CN115938128B (zh) * 2023-03-15 2023-10-03 天津所托瑞安汽车科技有限公司 交通事故预测方法、装置、终端及存储介质
CN117541066A (zh) * 2024-01-09 2024-02-09 巨野金牛车业有限公司 一种自卸半挂车侧翻评价指标及评价方法
CN117541066B (zh) * 2024-01-09 2024-03-29 巨野金牛车业有限公司 一种自卸半挂车侧翻评价指标及评价方法

Also Published As

Publication number Publication date
JP2022535629A (ja) 2022-08-10
JP7162919B2 (ja) 2022-10-31

Similar Documents

Publication Publication Date Title
WO2021238136A1 (zh) 一种基于hmm-rf混合模型的重型车辆侧翻预警方法和系统
US20190155291A1 (en) Methods and systems for automated driving system simulation, validation, and implementation
CN103578227B (zh) 基于gps定位信息的疲劳驾驶检测方法
CN102167041B (zh) 一种基于加速度传感器的车辆行驶状态的判定方法
CN105599773B (zh) 一种基于车辆运动姿态的驾驶员状态提示装置及其方法
CN109878509B (zh) 基于模糊逻辑的整体式罐车多源信息融合侧翻预警方法
CN107479061A (zh) 一种基于图像识别与雷达测距的安全预警方法及系统
CN106875510A (zh) 一种车辆侧翻预警方法和系统
CN114157513B (zh) 基于改进卷积神经网络的车联网入侵检测方法及设备
CN110660194A (zh) 一种驾驶监控预警方法及系统
CN111813083B (zh) 场景任务复杂度量化评估方法
CN112373483B (zh) 一种基于前向神经网络的车辆速度和转向预测方法
CN101327795A (zh) 一种车辆行驶智能预警系统
CN106960189A (zh) 一种基于隐含马尔可夫模型的驾驶意图判定方法
CN112560782A (zh) 一种基于随机森林算法的车辆换道行为识别方法
CN112356849A (zh) 一种危险驾驶的预警方法及装置
CN104200704A (zh) 车辆避撞预警方法及设备
CN110021192A (zh) 车道偏离告警方法、智能告警装置以及车辆
CN104111095B (zh) 基于汽车行驶状态的汽车油耗分析方法及系统
CN113222211B (zh) 一种多区域柴油车污染物排放因子预测方法及系统
CN103273914A (zh) 一种基于动态优化的汽车自动制动装置
Singh et al. Application of machine learning & deep learning techniques in the context of use cases relevant for the tire industry
CN105329239A (zh) 攻击性驾驶行为的检测和处理
CN113240901B (zh) 一种智能车换道行为风险等级确定方法及装置
CN113945224A (zh) 智能驾驶adas测试场景自动化生成方法及系统

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2020571534

Country of ref document: JP

Kind code of ref document: A

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

Ref document number: 20937514

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: 20937514

Country of ref document: EP

Kind code of ref document: A1