WO2021109644A1 - Hybrid vehicle working condition prediction method based on meta-learning - Google Patents

Hybrid vehicle working condition prediction method based on meta-learning Download PDF

Info

Publication number
WO2021109644A1
WO2021109644A1 PCT/CN2020/112611 CN2020112611W WO2021109644A1 WO 2021109644 A1 WO2021109644 A1 WO 2021109644A1 CN 2020112611 W CN2020112611 W CN 2020112611W WO 2021109644 A1 WO2021109644 A1 WO 2021109644A1
Authority
WO
WIPO (PCT)
Prior art keywords
vehicle speed
data
training
model
prediction
Prior art date
Application number
PCT/CN2020/112611
Other languages
French (fr)
Chinese (zh)
Inventor
何洪文
曹剑飞
Original Assignee
北京理工大学
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 北京理工大学 filed Critical 北京理工大学
Publication of WO2021109644A1 publication Critical patent/WO2021109644A1/en

Links

Images

Classifications

    • 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
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • 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
    • Y02T10/80—Technologies aiming to reduce greenhouse gasses emissions common to all road transportation technologies
    • Y02T10/84—Data processing systems or methods, management, administration

Definitions

  • the present invention relates to the technical field of vehicle working condition prediction, in particular to a meta-learning-based working condition prediction technology suitable for hybrid vehicles.
  • model predictive energy management strategy is generally considered to be a hybrid energy management strategy that can simultaneously take into account online application and optimizing control effects, and has been extensively studied.
  • model predictive control an accurate predictive model is of great significance for solving the optimal control sequence.
  • driving behavior itself has great uncertainty and randomness, there are still some hidden rules in the driving process. These rules are reflected in the probability of different future speed sequences corresponding to similar historical speed sequences. . Therefore, if the above-mentioned laws can be effectively used, a new approach to the technical problem of improving the fuel economy of hybrid electric vehicles can be provided.
  • the present invention provides a method for predicting working conditions of hybrid electric vehicles based on meta-learning, which specifically includes the following steps:
  • Step 1 Establish a vehicle speed initial prediction model based on deep neural network, using historical vehicle speed sequence and predicted future vehicle speed sequence as the input and output of the model, and the number of sequence elements is the same as the number of neurons in the input layer and output layer. Number correspondence
  • Step 2 Using the measured operating condition data collected under different traffic conditions, perform offline pre-training on the initial vehicle speed prediction model for repeated updating of model parameters to obtain a stable vehicle speed prediction base model after pre-training;
  • Step 3 Perform online fine-tuning training on the vehicle speed prediction base model using specific cyclic operating conditions or actual measured operating conditions data, so that the base model meets the accuracy verification requirements, so as to obtain a fine-tuned and trained vehicle speed prediction fine-tuning model;
  • Step 4 Based on the vehicle speed prediction fine-tuning model that meets the accuracy verification requirements, use real-time collected vehicle driving data to predict the future vehicle speed.
  • step one specifically includes:
  • the initial vehicle speed prediction model can be expressed by the following formula:
  • F(*) represents the mapping relationship function from historical vehicle speed sequence to future vehicle speed sequence
  • V(t ⁇ *) represents vehicle speed information per second
  • ⁇ H and ⁇ P represent the number of neurons in the input layer and output layer of the neural network ;
  • the accuracy of vehicle speed prediction is defined by the following formula:
  • V t+i and Respectively represent the predicted vehicle speed and the reference vehicle speed at time t, and i represents the serial number of the vehicle speed sequence.
  • the second step specifically includes:
  • Each set of data in each training data pool has the following format:
  • the input data sequence is:
  • n H and n P The dimensions of input data and label data are: n H and n P ;
  • L task is the overall loss function
  • ⁇ is a reference coefficient to determine whether the data in the query set is stable
  • the parameters of the initial prediction model of vehicle speed are updated by the following formula.
  • ⁇ meta represents the learning rate of the meta- learning process, which is a small positive number
  • step three specifically includes:
  • n the number of query sets
  • ⁇ Loss represents the allowable loss error.
  • an initial vehicle speed prediction model can be established first, and subsequent offline pre-training and online fine-tuning training steps can be executed in sequence, and finally used for actual measurement. It is also possible to establish an initial prediction model of the same driving during pre-training and fine-tuning training, and to integrate the results of pre-training and fine-tuning in the actual measurement link to adjust the final future vehicle speed prediction process.
  • the stable prediction model parameters are obtained through pre-training, and the prediction accuracy can meet the predetermined requirements through fine-tuning training. Both of these steps include but are not limited to the detailed steps mentioned above. For those skilled in the art based on this The specific multiple implementation manners that can be conceived of the inventive concept of the invention also fall within the protection scope of the independent claims of the present invention.
  • the meta-learning-based hybrid vehicle operating condition prediction method combines a multi-task training method on the basis of a deep neural network.
  • the model training process is divided into two parts: offline execution pre-training and online execution Fine-tuning training.
  • Pre-training performs parallel training for multiple working conditions to obtain a base model with better generalization performance.
  • Fine-tuning training is based on the basic model for specific working conditions, with low time cost and can be applied to the online correction of the model.
  • an online application framework of a vehicle speed prediction model consisting of offline training, online training and real-time prediction is further given, which can be applied to the task of predicting working conditions under actual traffic conditions.
  • Figure 1 is the deep neural network architecture on which the present invention is based
  • Fig. 2 is a schematic diagram of the basic flow of pre-training in the method of the present invention.
  • Figure 3 is a schematic diagram of the basic flow of fine-tuning training in the method of the present invention.
  • Fig. 4 is an online application framework corresponding to an example of the present invention.
  • the method for predicting working conditions of hybrid electric vehicles based on meta-learning specifically includes the following steps:
  • Step 1 Establish a vehicle speed initial prediction model based on deep neural network, using historical vehicle speed sequence and predicted future vehicle speed sequence as the input and output of the model, and the number of sequence elements is the same as the number of neurons in the input layer and output layer. Number correspondence; the deep neural network architecture is shown in Figure 1.
  • Step 2 Using the measured operating condition data collected under different traffic conditions, perform offline pre-training on the initial vehicle speed prediction model for repeated updating of model parameters to obtain a stable vehicle speed prediction base model after pre-training;
  • Step 3 Perform online fine-tuning training on the vehicle speed prediction base model using specific cyclic operating conditions or actual measured operating conditions data, so that the base model meets the accuracy verification requirements, so as to obtain a fine-tuned and trained vehicle speed prediction fine-tuning model;
  • Step 4 Based on the vehicle speed prediction fine-tuning model that meets the accuracy verification requirements, use real-time collected vehicle driving data to predict the future vehicle speed.
  • the step one specifically includes:
  • the initial vehicle speed prediction model can be expressed by the following formula:
  • F(*) represents the mapping relationship function from historical vehicle speed sequence to future vehicle speed sequence
  • V(t ⁇ *) represents vehicle speed information per second
  • ⁇ H and ⁇ P represent the number of neurons in the input layer and output layer of the neural network ;
  • the accuracy of vehicle speed prediction is defined by the following formula:
  • V t+i and Respectively represent the predicted vehicle speed and the reference vehicle speed at time t, and i represents the serial number of the vehicle speed sequence.
  • the second step specifically includes:
  • Each set of data in each training data pool has the following format:
  • the input data sequence is:
  • n H and n P The dimensions of input data and label data are: n H and n P ;
  • L task is the overall loss function
  • ⁇ is a reference coefficient to determine whether the data in the query set is stable
  • the parameters of the initial prediction model of vehicle speed are updated by the following formula.
  • ⁇ meta represents the learning rate of the meta- learning process, which is a small positive number
  • the step three specifically includes:
  • n the number of query sets
  • ⁇ Loss represents the allowable loss error.
  • Fig. 4 shows a preferred embodiment of the present invention.
  • an initial vehicle speed prediction model can be established first, and subsequent offline pre-training and online fine-tuning training steps can be executed in sequence, and finally used for actual measurement. It is also possible to establish an initial prediction model of the same driving during pre-training and fine-tuning training, and to integrate the results of pre-training and fine-tuning in the actual measurement link to adjust the final future vehicle speed prediction process.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Business, Economics & Management (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Marketing (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Traffic Control Systems (AREA)
  • Feedback Control In General (AREA)
  • Electric Propulsion And Braking For Vehicles (AREA)

Abstract

A hybrid vehicle working condition prediction method based on meta-learning. A model training process is divided into two parts: pre-training executed offline and fine-tuning training executed online; the pre-training relates to implementing parallel training for various working conditions to obtain a base model having good generalization performance; the fine-tuning training combines multi-task training on the basis of a deep neural network, and relates to training for a specific working condition on the basis of the base model. The time cost is low, and the method can be applied to a model online correction link. In addition, on the basis of said process, further provided is a vehicle speed prediction model online application framework composed of three parts, i.e., offline training, online training, and real-time prediction, which can be applied to a working condition prediction task under actual traffic conditions.

Description

一种基于元学习的混合动力车辆工况预测方法A Method for Predicting Working Conditions of Hybrid Electric Vehicles Based on Meta-learning 技术领域Technical field
本发明涉及车辆工况预测技术领域，尤其涉及一种适用于混合动力车辆的基于元学习实现的工况预测技术。The present invention relates to the technical field of vehicle working condition prediction, in particular to a meta-learning-based working condition prediction technology suitable for hybrid vehicles.
背景技术Background technique
混合动力车辆相较于传统燃油车，其燃油经济性的提升常需要通过整车能量管理策略来实现。目前，模型预测能量管理策略被普遍认为是一种可以同时兼顾在线应用及优化控制效果的混合动力能量管理策略，得到了广泛研究。在模型预测控制框架下，一个精准的预测模型对于求解最优控制序列具有重要意义。虽然驾驶行为本身具有很大的不确定性和随机性，但在驾驶过程中仍存在一些隐藏规则，这些规律反映在由相似的历史速度序列所对应的不同的未来速度序列在概率上也是相似的。因此，如果能够有效地利用上述规律，则对于提升混合动力电动汽车的燃油经济性能这一技术问题，可提供新型的实现途径。Compared with traditional fuel vehicles, hybrid vehicles often need to improve their fuel economy through vehicle energy management strategies. At present, the model predictive energy management strategy is generally considered to be a hybrid energy management strategy that can simultaneously take into account online application and optimizing control effects, and has been extensively studied. Under the framework of model predictive control, an accurate predictive model is of great significance for solving the optimal control sequence. Although the driving behavior itself has great uncertainty and randomness, there are still some hidden rules in the driving process. These rules are reflected in the probability of different future speed sequences corresponding to similar historical speed sequences. . Therefore, if the above-mentioned laws can be effectively used, a new approach to the technical problem of improving the fuel economy of hybrid electric vehicles can be provided.
发明内容Summary of the invention
针对现有技术中存在的技术问题，本发明提供了一种基于元学习的混合动力车辆工况预测方法，具体包括以下步骤：In view of the technical problems existing in the prior art, the present invention provides a method for predicting working conditions of hybrid electric vehicles based on meta-learning, which specifically includes the following steps:
步骤一、建立基于深度神经网络的车速初始预测模型，分别以历史车速序列与将预测的未来车速序作为所述模型的输入与输出，序列元素个数分别与输入层和输出层的神经元个数对应；Step 1. Establish a vehicle speed initial prediction model based on deep neural network, using historical vehicle speed sequence and predicted future vehicle speed sequence as the input and output of the model, and the number of sequence elements is the same as the number of neurons in the input layer and output layer. Number correspondence
步骤二、利用不同交通条件下采集的实测工况数据，对所述车速初始预测模型进行离线预训练，用于对模型参数进行重复更新，以得到经预训练后稳定的车速预测基模型；Step 2: Using the measured operating condition data collected under different traffic conditions, perform offline pre-training on the initial vehicle speed prediction model for repeated updating of model parameters to obtain a stable vehicle speed prediction base model after pre-training;
步骤三、利用特定的循环工况或实测工况数据对所述车速预测基模型进行在线微调训练，使所述基模型满足精度校验要求，以得到经微调训练的车速预测微调模型；Step 3: Perform online fine-tuning training on the vehicle speed prediction base model using specific cyclic operating conditions or actual measured operating conditions data, so that the base model meets the accuracy verification requirements, so as to obtain a fine-tuned and trained vehicle speed prediction fine-tuning model;
步骤四、基于满足精度校验要求的车速预测微调模型，利用实时采集的车辆行驶数据对未来车速进行预测。Step 4: Based on the vehicle speed prediction fine-tuning model that meets the accuracy verification requirements, use real-time collected vehicle driving data to predict the future vehicle speed.
进一步地，所述步骤一具体包括：Further, the step one specifically includes:
将历史车速序列作为输入参数，将预测的未来车速序列作为输出参数，车速初始预测模型可以由以下公式表示：Taking the historical vehicle speed sequence as the input parameter and the predicted future vehicle speed sequence as the output parameter, the initial vehicle speed prediction model can be expressed by the following formula:
[V t+1,V t+2,...,V t+ΔP]＝F(V t-ΔH,...V t-1,V t) [V t+1 ,V t+2 ,...,V t+ΔP ]=F(V t-ΔH ,...V t-1 ,V t )
其中，F(*)表示由历史车速序列到未来车速序列的映射关系函数；V(t±*)表示每一秒的车速信息；ΔH和ΔP表示神经网络输入层和输出层的神经元个数；Among them, F(*) represents the mapping relationship function from historical vehicle speed sequence to future vehicle speed sequence; V(t±*) represents vehicle speed information per second; ΔH and ΔP represent the number of neurons in the input layer and output layer of the neural network ；
车速预测的精度由以下公式定义：The accuracy of vehicle speed prediction is defined by the following formula:
Figure PCTCN2020112611-appb-000001
Figure PCTCN2020112611-appb-000001
其中，V t+i和
Figure PCTCN2020112611-appb-000002
分别表示时间t时刻的预测车速和参考车速，i表示车速序列的序号。预测误差值越小，则表示预测精度越高。 Where V t+i and
Figure PCTCN2020112611-appb-000002
Respectively represent the predicted vehicle speed and the reference vehicle speed at time t, and i represents the serial number of the vehicle speed sequence. The smaller the prediction error value, the higher the prediction accuracy.
进一步地，所述步骤二具体包括：Further, the second step specifically includes:
2.1、初始化模型网络参数为：θ k，k＝1,2,3,4...； 2.1. The network parameters of the initialization model are: θ k , k = 1, 2, 3, 4...;
2.2、对于参与训练的每一组工况，都定义一个与之对应的数据池，并将其命名为：2.2. For each group of working conditions participating in training, define a corresponding data pool and name it:
Pool i,i＝1,2,3,...,n Pool i ,i=1,2,3,...,n
每个训练数据池中的每一组数据都具有如下的格式：Each set of data in each training data pool has the following format:
Figure PCTCN2020112611-appb-000003
Figure PCTCN2020112611-appb-000003
其中，输入数据序列为：Among them, the input data sequence is:
Figure PCTCN2020112611-appb-000004
Figure PCTCN2020112611-appb-000004
标签数据序列为：The label data sequence is:
Figure PCTCN2020112611-appb-000005
Figure PCTCN2020112611-appb-000005
输入数据和标签数据的维度依次为：n H和n P； The dimensions of input data and label data are: n H and n P ;
2.3、对于每一组训练数据集，均执行一次如下的流程：2.3. For each training data set, perform the following process once:
2.3.1、通过随机采样，从训练数据池中选择一个批次的数据，然后将数据划分为两组：一组被命名为支持集(support set)，另一组被命名为查询集(query set)；2.3.1. Through random sampling, select a batch of data from the training data pool, and then divide the data into two groups: one group is named as the support set, and the other group is named as the query set. set);
2.3.2、通过支持集的数据来更新车速初始预测模型的原始参数，将网络参数θ k更新为
Figure PCTCN2020112611-appb-000006
(采用梯度下降法来更新，更新次数可以是1次，也可以是多次，但是更新次数这个值是固定的)。 2.3.2. Update the original parameters of the initial vehicle speed prediction model through the data of the support set, and update the network parameters θ k to
Figure PCTCN2020112611-appb-000006
(Using the gradient descent method to update, the number of updates can be 1 time or multiple times, but the value of the number of updates is fixed).
2.3.3、基于更新后的参数，通过初始预测模型和由查询集提供的训练数据来计算对应的损失函数L i。 2.3.3. Based on the updated parameters, calculate the corresponding loss function L i through the initial prediction model and the training data provided by the query set.
2.4、按照如下公式计算整体损失函数：2.4. Calculate the overall loss function according to the following formula:
Figure PCTCN2020112611-appb-000007
Figure PCTCN2020112611-appb-000007
其中，L task为整体损失函数，ε是一个参考系数，用以确定查询集的数据是否稳定； Among them, L task is the overall loss function, and ε is a reference coefficient to determine whether the data in the query set is stable;
然后计算整体损失函数L task关于模型参数θ的导数，并将其记录为
Figure PCTCN2020112611-appb-000008
Then calculate the derivative of the overall loss function L task with respect to the model parameter θ, and record it as
Figure PCTCN2020112611-appb-000008
通过如下公式更新车速初始预测模型参数。The parameters of the initial prediction model of vehicle speed are updated by the following formula.
Figure PCTCN2020112611-appb-000009
Figure PCTCN2020112611-appb-000009
其中，η meta表示元学习过程的学习率，是一个很小的正数； Among them, η meta represents the learning rate of the meta- learning process, which is a small positive number;
2.5、令θ k＝θ k+1并重复步骤2.2-2.4，直至训练得到一个稳定的车速预测基模型。 2.5. Let θ k = θ k+1 and repeat steps 2.2-2.4 until a stable base model for vehicle speed prediction is obtained through training.
进一步地，所述步骤三具体包括：Further, the step three specifically includes:
利用所述车速预测基模型参数，并依次执行：Use the vehicle speed to predict the base model parameters, and execute them in sequence:
3.1、利用特定的循环工况或实测工况对应的行驶数据定义训练数据池，其格式与预训练时的训练池数据格式相同；3.1. Use the driving data corresponding to specific cycle conditions or measured conditions to define the training data pool, the format of which is the same as the training pool data format during pre-training;
3.2、通过随机采样，从训练数据池中抽取一定批次的数据，然后将数据划分为若干组(超过2组)；将其中一组数据被定义为支持集(support set)，另外的其他所有组数据均被定义为查询集(query set)；3.2. Through random sampling, extract a certain batch of data from the training data pool, and then divide the data into several groups (more than 2 groups); define one group of data as a support set, and all others Group data are all defined as query set;
3.3、通过支持集的数据，执行一次梯度下降，将模型参数从θ k更新至θ k+1； 3.3. Perform a gradient descent based on the data of the support set, and update the model parameters from θ k to θ k+1 ;
3.4、基于更新后的参数，通过基模型来计算每一组查询集中数据对应的损失函数值，并通过如下公式计算平均预测误差：3.4. Based on the updated parameters, calculate the loss function value corresponding to each set of query data through the base model, and calculate the average prediction error through the following formula:
Figure PCTCN2020112611-appb-000010
Figure PCTCN2020112611-appb-000010
其中，n表示查询集的个数；Among them, n represents the number of query sets;
3.5、令θ k＝θ k+1，并重复执行步骤3.2-3.4直到如下公式被满足： 3.5. Let θ k = θ k+1 , and repeat steps 3.2-3.4 until the following formula is satisfied:
Figure PCTCN2020112611-appb-000011
Figure PCTCN2020112611-appb-000011
其中，ΔLoss表示允许的损失误差。Among them, ΔLoss represents the allowable loss error.
上述方法在执行过程中，可首先建立车速预测初始模型并依次执行后续的离线预训练与在线微调训练步骤，最后用于实测。也可在预训练与微调训练时分别建立相同行驶的初始预测模型，并在实测环节中融合预训练及微调的结果对最终的未来车速预测过程进行调整。During the execution of the above method, an initial vehicle speed prediction model can be established first, and subsequent offline pre-training and online fine-tuning training steps can be executed in sequence, and finally used for actual measurement. It is also possible to establish an initial prediction model of the same driving during pre-training and fine-tuning training, and to integrate the results of pre-training and fine-tuning in the actual measurement link to adjust the final future vehicle speed prediction process.
上述方法在执行过程中，通过预训练得到稳定的预测模型参数，以及通过微调训练使预测精度达到预定要求，均可采用包括但不限于上述所提到的详细步骤，对于本领域技术人员基于本发明的发明构思所能够想到的具体多种实现方式，同样落入本发明独立权利要求的保护范围。During the execution of the above method, the stable prediction model parameters are obtained through pre-training, and the prediction accuracy can meet the predetermined requirements through fine-tuning training. Both of these steps include but are not limited to the detailed steps mentioned above. For those skilled in the art based on this The specific multiple implementation manners that can be conceived of the inventive concept of the invention also fall within the protection scope of the independent claims of the present invention.
本发明所提供的基于元学习的混合动力车辆工况预测方法，在深度神经网络的基础上结合了多任务训练的方式，模型训练过程被划分为两个部分：离线执行的预训练与在线执行的微调训练。预训练针对多种工况进行并行训练，以获得泛化性能较好的基模型。微调训练在基模型基础上针对特定工况进行训练，时间成本低，可以应用于模型的在线修正环节。此外，基于上述流程，还进一步给出了一种由离线训练、在线训练和实时预测三部分组成的车速预测模型在线应用框架，可应用于实际交通条件下的工况预测任务。The meta-learning-based hybrid vehicle operating condition prediction method provided by the present invention combines a multi-task training method on the basis of a deep neural network. The model training process is divided into two parts: offline execution pre-training and online execution Fine-tuning training. Pre-training performs parallel training for multiple working conditions to obtain a base model with better generalization performance. Fine-tuning training is based on the basic model for specific working conditions, with low time cost and can be applied to the online correction of the model. In addition, based on the above process, an online application framework of a vehicle speed prediction model consisting of offline training, online training and real-time prediction is further given, which can be applied to the task of predicting working conditions under actual traffic conditions.
附图说明Description of the drawings
图1是本发明所基于的深度神经网络构架；Figure 1 is the deep neural network architecture on which the present invention is based;
图2是本发明的方法中的预训练基本流程示意图；Fig. 2 is a schematic diagram of the basic flow of pre-training in the method of the present invention;
图3是本发明的方法中的微调训练基本流程示意图；Figure 3 is a schematic diagram of the basic flow of fine-tuning training in the method of the present invention;
图4是本发明的一实例所对应的在线应用框架。Fig. 4 is an online application framework corresponding to an example of the present invention.
具体实施方式Detailed ways
下面将结合附图对本发明的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.
本发明所提供的一种基于元学习的混合动力车辆工况预测方法，具体包括以下步骤：The method for predicting working conditions of hybrid electric vehicles based on meta-learning provided by the present invention specifically includes the following steps:
步骤一、建立基于深度神经网络的车速初始预测模型，分别以历史车速序列与将预测的未来车速序作为所述模型的输入与输出，序列元素个数分别与输入层和输出层的神经元个数对应；所述深度神经网络架构如图1所示。Step 1. Establish a vehicle speed initial prediction model based on deep neural network, using historical vehicle speed sequence and predicted future vehicle speed sequence as the input and output of the model, and the number of sequence elements is the same as the number of neurons in the input layer and output layer. Number correspondence; the deep neural network architecture is shown in Figure 1.
步骤二、利用不同交通条件下采集的实测工况数据，对所述车速初始预测模型进行离线预训练，用于对模型参数进行重复更新，以得到经预训练后稳定的车速预测基模型；Step 2: Using the measured operating condition data collected under different traffic conditions, perform offline pre-training on the initial vehicle speed prediction model for repeated updating of model parameters to obtain a stable vehicle speed prediction base model after pre-training;
步骤三、利用特定的循环工况或实测工况数据对所述车速预测基模型进行在线微调训练，使所述基模型满足精度校验要求，以得到经微调训练的车速预测微调模型；Step 3: Perform online fine-tuning training on the vehicle speed prediction base model using specific cyclic operating conditions or actual measured operating conditions data, so that the base model meets the accuracy verification requirements, so as to obtain a fine-tuned and trained vehicle speed prediction fine-tuning model;
步骤四、基于满足精度校验要求的车速预测微调模型，利用实时采集的车辆行驶数据对未来车速进行预测。Step 4: Based on the vehicle speed prediction fine-tuning model that meets the accuracy verification requirements, use real-time collected vehicle driving data to predict the future vehicle speed.
在本发明的一优选实施例中，所述步骤一具体包括：In a preferred embodiment of the present invention, the step one specifically includes:
将历史车速序列作为输入参数，将预测的未来车速序列作为输出参数，车速初始预测模型可以由以下公式表示：Taking the historical vehicle speed sequence as the input parameter and the predicted future vehicle speed sequence as the output parameter, the initial vehicle speed prediction model can be expressed by the following formula:
[V t+1,V t+2,...,V t+ΔP]＝F(V t-ΔH,...V t-1,V t) [V t+1 ,V t+2 ,...,V t+ΔP ]=F(V t-ΔH ,...V t-1 ,V t )
其中，F(*)表示由历史车速序列到未来车速序列的映射关系函数；V(t±*)表示每一秒的车速信息；ΔH和ΔP表示神经网络输入层和输出层的神经元个数；Among them, F(*) represents the mapping relationship function from historical vehicle speed sequence to future vehicle speed sequence; V(t±*) represents vehicle speed information per second; ΔH and ΔP represent the number of neurons in the input layer and output layer of the neural network ；
车速预测的精度由以下公式定义：The accuracy of vehicle speed prediction is defined by the following formula:
Figure PCTCN2020112611-appb-000012
Figure PCTCN2020112611-appb-000012
其中，V t+i和
Figure PCTCN2020112611-appb-000013
分别表示时间t时刻的预测车速和参考车速，i表示车速序列的序号。预测误差值越小，则表示预测精度越高。 Where V t+i and
Figure PCTCN2020112611-appb-000013
Respectively represent the predicted vehicle speed and the reference vehicle speed at time t, and i represents the serial number of the vehicle speed sequence. The smaller the prediction error value, the higher the prediction accuracy.
在本发明的一优选实施例中，如图2所示，所述步骤二具体包括：In a preferred embodiment of the present invention, as shown in FIG. 2, the second step specifically includes:
2.1、初始化模型网络参数为：θ k，k＝1,2,3,4...； 2.1. The network parameters of the initialization model are: θ k , k = 1, 2, 3, 4...;
2.2、对于参与训练的每一组工况，都定义一个与之对应的数据池，并将其命名为：2.2. For each group of working conditions participating in training, define a corresponding data pool and name it:
Pool i,i＝1,2,3,...,n Pool i ,i=1,2,3,...,n
每个训练数据池中的每一组数据都具有如下的格式：Each set of data in each training data pool has the following format:
Figure PCTCN2020112611-appb-000014
Figure PCTCN2020112611-appb-000014
其中，输入数据序列为：Among them, the input data sequence is:
Figure PCTCN2020112611-appb-000015
Figure PCTCN2020112611-appb-000015
标签数据序列为：The label data sequence is:
Figure PCTCN2020112611-appb-000016
Figure PCTCN2020112611-appb-000016
输入数据和标签数据的维度依次为：n H和n P； The dimensions of input data and label data are: n H and n P ;
2.3、对于每一组训练数据集，均执行一次如下的流程：2.3. For each training data set, perform the following process once:
2.3.1、通过随机采样，从训练数据池中选择一个批次的数据，然后将数据划分为两组：一组被命名为支持集(support set)，另一组被命名为查询集(query set)；2.3.1. Through random sampling, select a batch of data from the training data pool, and then divide the data into two groups: one group is named as the support set, and the other group is named as the query set. set);
2.3.2、通过支持集的数据来更新车速初始预测模型的原始参数，将网络参数θ k更新为
Figure PCTCN2020112611-appb-000017
(采用梯度下降法来更新，更新次数可以是1次，也可以是多次，但是更新次数这个值是固定的)。 2.3.2. Update the original parameters of the initial vehicle speed prediction model through the data of the support set, and update the network parameters θ k to
Figure PCTCN2020112611-appb-000017
(Using the gradient descent method to update, the number of updates can be 1 time or multiple times, but the value of the number of updates is fixed).
2.3.3、基于更新后的参数，通过初始预测模型和由查询集提供的训练数据来计算对应的损失函数L i。 2.3.3. Based on the updated parameters, calculate the corresponding loss function L i through the initial prediction model and the training data provided by the query set.
2.4、按照如下公式计算整体损失函数：2.4. Calculate the overall loss function according to the following formula:
Figure PCTCN2020112611-appb-000018
Figure PCTCN2020112611-appb-000018
其中，L task为整体损失函数，ε是一个参考系数，用以确定查询集的数据是否稳定； Among them, L task is the overall loss function, and ε is a reference coefficient to determine whether the data in the query set is stable;
然后计算整体损失函数L task关于模型参数θ的导数，并将其记录为
Figure PCTCN2020112611-appb-000019
Then calculate the derivative of the overall loss function L task with respect to the model parameter θ, and record it as
Figure PCTCN2020112611-appb-000019
通过如下公式更新车速初始预测模型参数。The parameters of the initial prediction model of vehicle speed are updated by the following formula.
Figure PCTCN2020112611-appb-000020
Figure PCTCN2020112611-appb-000020
其中，η meta表示元学习过程的学习率，是一个很小的正数； Among them, η meta represents the learning rate of the meta- learning process, which is a small positive number;
2.5、令θ k＝θ k+1并重复步骤2.2-2.4，直至训练得到一个稳定的车速预测基模型。 2.5. Let θ k = θ k+1 and repeat steps 2.2-2.4 until a stable base model for vehicle speed prediction is obtained through training.
在本发明的一优选实施例中，如图3所示，所述步骤三具体包括：In a preferred embodiment of the present invention, as shown in FIG. 3, the step three specifically includes:
利用所述车速预测基模型参数，并依次执行：Use the vehicle speed to predict the base model parameters, and execute them in sequence:
3.1、利用特定的循环工况或实测工况对应的行驶数据定义训练数据池，其格式与预训练时的训练池数据格式相同；3.1. Use the driving data corresponding to specific cycle conditions or measured conditions to define the training data pool, the format of which is the same as the training pool data format during pre-training;
3.2、通过随机采样，从训练数据池中抽取一定批次的数据，然后将数据划分为若干组(超过2组)；将其中一组数据被定义为支持集(support set)，另外的其他所有组数据均被定义为查询集(query set)；3.2. Through random sampling, extract a certain batch of data from the training data pool, and then divide the data into several groups (more than 2 groups); define one group of data as a support set, and all others Group data are all defined as query set;
3.3、通过支持集的数据，执行一次梯度下降，将模型参数从θ k更新至θ k+1； 3.3. Perform a gradient descent based on the data of the support set, and update the model parameters from θ k to θ k+1 ;
3.4、基于更新后的参数，通过基模型来计算每一组查询集中数据对应的损失函数值，并通过如下公式计算平均预测误差：3.4. Based on the updated parameters, calculate the loss function value corresponding to each set of query data through the base model, and calculate the average prediction error through the following formula:
Figure PCTCN2020112611-appb-000021
Figure PCTCN2020112611-appb-000021
其中，n表示查询集的个数；Among them, n represents the number of query sets;
3.5、令θ k＝θ k+1，并重复执行步骤3.2-3.4直到如下公式被满足： 3.5. Let θ k = θ k+1 , and repeat steps 3.2-3.4 until the following formula is satisfied:
Figure PCTCN2020112611-appb-000022
Figure PCTCN2020112611-appb-000022
其中，ΔLoss表示允许的损失误差。Among them, ΔLoss represents the allowable loss error.
图4示出了本发明的一优选实施方式，方法在执行过程中，可首先建立车速预测初始模型并依次执行后续的离线预训练与在线微调训练步骤，最后用于实测。也可在预训练与微调训练时分别建立相同行驶的初始预测模型，并在实测环节中融合预训练及微调的结果对最终的未来车速预测过程进行调整。Fig. 4 shows a preferred embodiment of the present invention. During the execution of the method, an initial vehicle speed prediction model can be established first, and subsequent offline pre-training and online fine-tuning training steps can be executed in sequence, and finally used for actual measurement. It is also possible to establish an initial prediction model of the same driving during pre-training and fine-tuning training, and to integrate the results of pre-training and fine-tuning in the actual measurement link to adjust the final future vehicle speed prediction process.
尽管已经示出和描述了本发明的实施例，对于本领域的普通技术人员而言，可以理解在不脱离本发明的原理和精神的情况下可以对这些实施例进行多种变化、修改、替换和变型，本发明的范围由所附权利要求及其等同物限定。Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, and substitutions can be made to these embodiments without departing from the principle and spirit of the present invention. And variations, the scope of the present invention is defined by the appended claims and their equivalents.

Claims (4)

  1. 一种基于元学习的混合动力车辆工况预测方法，其特征在于：具体包括以下步骤：A method for predicting working conditions of hybrid electric vehicles based on meta-learning is characterized in that it specifically includes the following steps:
    步骤一、建立基于深度神经网络的车速初始预测模型，分别以历史车速序列与将预测的未来车速序作为所述模型的输入与输出，序列元素个数分别与输入层和输出层的神经元个数对应；Step 1. Establish a vehicle speed initial prediction model based on deep neural network, using historical vehicle speed sequence and predicted future vehicle speed sequence as the input and output of the model, and the number of sequence elements is the same as the number of neurons in the input layer and output layer. Number correspondence
    步骤二、利用不同交通条件下采集的实测工况数据，对所述车速初始预测模型进行离线预训练，用于对模型参数进行重复更新，以得到经预训练后稳定的车速预测基模型；Step 2: Using the measured operating condition data collected under different traffic conditions, perform offline pre-training on the initial vehicle speed prediction model for repeated updating of model parameters to obtain a stable vehicle speed prediction base model after pre-training;
    步骤三、利用特定的循环工况或实测工况数据对所述车速预测基模型进行在线微调训练，使所述基模型满足精度校验要求，以得到经微调训练的车速预测微调模型；Step 3: Use specific cycle operating conditions or measured operating conditions data to perform online fine-tuning training on the vehicle speed prediction base model, so that the base model meets the accuracy verification requirements, so as to obtain a fine-tuned and trained vehicle speed prediction fine-tuning model;
    步骤四、基于满足精度校验要求的车速预测微调模型，利用实时采集的车辆行驶数据对未来车速进行预测。Step 4: Based on the vehicle speed prediction fine-tuning model that meets the accuracy verification requirements, use real-time collected vehicle driving data to predict the future vehicle speed.
  2. 如权利要求1所述的方法，其特征在于：所述步骤一具体包括：The method according to claim 1, wherein said step one specifically comprises:
    将历史车速序列作为输入参数，将预测的未来车速序列作为输出参数，车速初始预测模型由以下公式表示：Taking the historical vehicle speed sequence as the input parameter and the predicted future vehicle speed sequence as the output parameter, the initial vehicle speed prediction model is expressed by the following formula:
    [V t+1,V t+2,.. .,V t+ΔP]＝F(V t-ΔH,.. .V t-1,V t) [V t + 1, V t + 2, ..., V t + ΔP] = F (V t-ΔH, ... V t-1, V t)
    其中，F(*)表示由历史车速序列到未来车速序列的映射关系函数；V(t±*)表示每一秒的车速信息；ΔH和ΔP表示神经网络输入层和输出层的神经元个数；Among them, F(*) represents the mapping relationship function from historical vehicle speed sequence to future vehicle speed sequence; V(t±*) represents vehicle speed information per second; ΔH and ΔP represent the number of neurons in the input layer and output layer of the neural network ；
    车速预测的精度由以下公式定义：The accuracy of vehicle speed prediction is defined by the following formula:
    Figure PCTCN2020112611-appb-100001
    Figure PCTCN2020112611-appb-100001
    其中，V t+i和
    Figure PCTCN2020112611-appb-100002
    分别表示时间t时刻的预测车速和参考车速，i表示车速序列的序号。     Where V t+i and
    Figure PCTCN2020112611-appb-100002
    Respectively represent the predicted vehicle speed and the reference vehicle speed at time t, and i represents the serial number of the vehicle speed sequence.
  3. 如权利要求2所述的方法，其特征在于：所述步骤二具体包括：The method according to claim 2, wherein the step two specifically includes:
    2.1、初始化模型网络参数为：θ k，k＝1,2,3,4...； 2.1. The network parameters of the initialization model are: θ k , k = 1, 2, 3, 4...;
    2.2、对于参与训练的每一组工况，都定义一个与之对应的数据池，并将其定义为：2.2. For each group of working conditions participating in training, define a corresponding data pool and define it as:
    Pool i,i＝1,2,3,...,n Pool i ,i=1,2,3,...,n
    每个训练数据池中的每一组数据都具有如下的格式：Each set of data in each training data pool has the following format:
    Figure PCTCN2020112611-appb-100003
    Figure PCTCN2020112611-appb-100003
    其中，输入数据序列为：Among them, the input data sequence is:
    Figure PCTCN2020112611-appb-100004
    Figure PCTCN2020112611-appb-100004
    标签数据序列为：The label data sequence is:
    Figure PCTCN2020112611-appb-100005
    Figure PCTCN2020112611-appb-100005
    输入数据和标签数据的维度依次为：n H和n P； The dimensions of input data and label data are: n H and n P ;
    2.3、对于每一组训练数据集，均执行一次如下的流程：2.3. For each training data set, perform the following process once:
    2.3.1、通过随机采样，从训练数据池中选择一个批次的数据，然后将数据划分为两组：一组定义为支持集，另一组定义为查询集；2.3.1. Through random sampling, select a batch of data from the training data pool, and then divide the data into two groups: one group is defined as the support set, and the other group is defined as the query set;
    2.3.2、通过支持集的数据来更新车速初始预测模型的原始参数，将网络参数θ k更新为
    Figure PCTCN2020112611-appb-100006
    2.3.2. Update the original parameters of the initial vehicle speed prediction model through the data of the support set, and update the network parameters θ k to
    Figure PCTCN2020112611-appb-100006
    2.3.3、基于更新后的参数，通过初始预测模型和由查询集提供的训练数据来计算对应的损失函数L i； 2.3.3. Based on the updated parameters, calculate the corresponding loss function L i through the initial prediction model and the training data provided by the query set;
    2.4、按照如下公式计算整体损失函数：2.4. Calculate the overall loss function according to the following formula:
    Figure PCTCN2020112611-appb-100007
    Figure PCTCN2020112611-appb-100007
    其中，L task为整体损失函数，ε是一个参考系数，用以确定查询集的数据是否稳定； Among them, L task is the overall loss function, and ε is a reference coefficient to determine whether the data in the query set is stable;
    然后计算整体损失函数L task关于网络参数θ的导数，并将其记录为
    Figure PCTCN2020112611-appb-100008
    Then calculate the derivative of the overall loss function L task with respect to the network parameter θ, and record it as
    Figure PCTCN2020112611-appb-100008
    通过如下公式更新车速初始预测模型参数：Update the parameters of the initial prediction model of vehicle speed through the following formula:
    Figure PCTCN2020112611-appb-100009
    Figure PCTCN2020112611-appb-100009
    其中，η meta表示元学习过程的学习率，是一个很小的正数； Among them, η meta represents the learning rate of the meta- learning process, which is a small positive number;
    2.5、令θ k＝θ k+1并重复步骤2.2-2.4，直至训练得到一个稳定的车速预测基模型。 2.5. Let θ k = θ k+1 and repeat steps 2.2-2.4 until a stable base model for vehicle speed prediction is obtained through training.
  4. 如权利要求2所述的方法，其特征在于：所述步骤三具体包括：The method according to claim 2, wherein the step three specifically includes:
    利用所述车速预测基模型参数，并依次执行：Use the vehicle speed to predict the base model parameters, and execute them in sequence:
    3.1、利用特定的循环工况或实测工况对应的行驶数据定义训练数据池，其格式与预训练时的训练池数据格式相同；3.1. Use the driving data corresponding to specific cycle conditions or measured conditions to define the training data pool, the format of which is the same as the training pool data format during pre-training;
    3.2、通过随机采样，从训练数据池中抽取一定批次的数据，然后将数据划分为若干组；将其中一组数据定义为支持集，另外的其他所有组数据均定义为查询集；3.2. Through random sampling, extract a certain batch of data from the training data pool, and then divide the data into several groups; define one group of data as a support set, and all other groups of data as a query set;
    3.3、通过支持集的数据，执行一次梯度下降，将模型网络参数从θ k更新至θ k+1； 3.3. Perform a gradient descent through the data of the support set, and update the model network parameters from θ k to θ k+1 ;
    3.4、基于更新后的参数，通过基模型来计算每一组查询集中数据对应的损失函数值，并通过如下公式计算平均预测误差：3.4. Based on the updated parameters, calculate the loss function value corresponding to each set of query data through the base model, and calculate the average prediction error through the following formula:
    Figure PCTCN2020112611-appb-100010
    Figure PCTCN2020112611-appb-100010
    其中，L i为损失函数，n表示查询集的个数； Among them, Li is the loss function, and n is the number of query sets;
    3.5、令θ k＝θ k+1，并重复执行步骤3.2-3.4直到如下公式被满足： 3.5. Let θ k = θ k+1 , and repeat steps 3.2-3.4 until the following formula is satisfied:
    Figure PCTCN2020112611-appb-100011
    Figure PCTCN2020112611-appb-100011
    其中，ΔLoss表示允许的损失误差。Among them, ΔLoss represents the allowable loss error.
PCT/CN2020/112611 2019-12-06 2020-08-31 Hybrid vehicle working condition prediction method based on meta-learning WO2021109644A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911239316.5 2019-12-06
CN201911239316.5A CN111047085B (en) 2019-12-06 2019-12-06 Hybrid vehicle working condition prediction method based on meta-learning

Publications (1)

Publication Number Publication Date
WO2021109644A1 true WO2021109644A1 (en) 2021-06-10

Family

ID=70235038

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/112611 WO2021109644A1 (en) 2019-12-06 2020-08-31 Hybrid vehicle working condition prediction method based on meta-learning

Country Status (2)

Country Link
CN (1) CN111047085B (en)
WO (1) WO2021109644A1 (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113223705A (en) * 2021-05-22 2021-08-06 杭州医康慧联科技股份有限公司 Offline prediction method suitable for privacy computing platform
CN117474928A (en) * 2023-12-28 2024-01-30 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117744504A (en) * 2024-02-20 2024-03-22 成都理工大学 Flood discharge atomization rain intensity analysis model building method and device

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111047085B (en) * 2019-12-06 2022-09-06 北京理工大学 Hybrid vehicle working condition prediction method based on meta-learning
CN112182962A (en) * 2020-09-24 2021-01-05 青岛大学 Hybrid electric vehicle running speed prediction method
CN112180726A (en) * 2020-09-29 2021-01-05 北京航空航天大学 Spacecraft relative motion trajectory planning method based on meta-learning
CN112269930B (en) * 2020-10-26 2023-10-24 北京百度网讯科技有限公司 Regional heat prediction model and regional heat prediction method and device
CN112529104B (en) * 2020-12-23 2024-06-18 东软睿驰汽车技术（沈阳）有限公司 Vehicle fault prediction model generation method, fault prediction method and device
CN115204417B (en) * 2022-09-13 2022-12-27 鱼快创领智能科技(南京)有限公司 Vehicle weight prediction method and system based on ensemble learning and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107284452A (en) * 2017-07-18 2017-10-24 吉林大学 Merge the following operating mode forecasting system of hybrid vehicle of intelligent communication information
CN107463992A (en) * 2017-08-01 2017-12-12 北京理工大学 The operating mode Forecasting Methodology that a kind of hybrid car travel operating mode is trained based on fragment waveform
CN108346293A (en) * 2018-04-18 2018-07-31 湖南大学 A kind of arithmetic for real-time traffic flow Forecasting Approach for Short-term
CN111047085A (en) * 2019-12-06 2020-04-21 北京理工大学 Hybrid vehicle working condition prediction method based on meta-learning

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2019137269A (en) * 2018-02-13 2019-08-22 株式会社豊田中央研究所 Driving force predictor and control device
CN109159785B (en) * 2018-07-19 2020-05-01 重庆科技学院 Automobile driving condition prediction method based on Markov chain and neural network
CN109829577B (en) * 2019-01-17 2021-10-01 北京交通大学 Rail train running state prediction method based on deep neural network structure model

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107284452A (en) * 2017-07-18 2017-10-24 吉林大学 Merge the following operating mode forecasting system of hybrid vehicle of intelligent communication information
CN107463992A (en) * 2017-08-01 2017-12-12 北京理工大学 The operating mode Forecasting Methodology that a kind of hybrid car travel operating mode is trained based on fragment waveform
CN108346293A (en) * 2018-04-18 2018-07-31 湖南大学 A kind of arithmetic for real-time traffic flow Forecasting Approach for Short-term
CN111047085A (en) * 2019-12-06 2020-04-21 北京理工大学 Hybrid vehicle working condition prediction method based on meta-learning

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113223705A (en) * 2021-05-22 2021-08-06 杭州医康慧联科技股份有限公司 Offline prediction method suitable for privacy computing platform
CN117474928A (en) * 2023-12-28 2024-01-30 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117474928B (en) * 2023-12-28 2024-03-19 东北大学 Ceramic package substrate surface defect detection method based on meta-learning model
CN117744504A (en) * 2024-02-20 2024-03-22 成都理工大学 Flood discharge atomization rain intensity analysis model building method and device

Also Published As

Publication number Publication date
CN111047085B (en) 2022-09-06
CN111047085A (en) 2020-04-21

Similar Documents

Publication Publication Date Title
WO2021109644A1 (en) Hybrid vehicle working condition prediction method based on meta-learning
CN110341690B (en) PHEV energy management method based on deterministic strategy gradient learning
CN104834215B (en) A kind of BP neural network pid control algorithm of mutation particle swarm optimization
CN108520155B (en) Vehicle behavior simulation method based on neural network
CN113255206B (en) Hydrologic prediction model parameter calibration method based on deep reinforcement learning
CN110059867B (en) Wind speed prediction method combining SWLSTM and GPR
CN106021990A (en) Method for achieving classification and self-recognition of biological genes by means of specific characters
CN104539601B (en) Dynamic network attack process analysis method for reliability and system
CN102034133A (en) Quantum neural network-based comprehensive evaluation method for multi-factor system
CN113156900B (en) Machining deformation control method based on meta reinforcement learning
CN116992779B (en) Simulation method and system of photovoltaic energy storage system based on digital twin model
CN110276483A (en) Sugaring raw material prediction technique neural network based
CN115600669A (en) High-efficiency deep pulse neural network learning method based on local classifier
CN104616072A (en) Method for improving concentration of glutamic acid fermented product based on interval optimization
CN110880044A (en) Markov chain-based load prediction method
CN116819372A (en) Method for calculating health state of energy storage battery by neural network
CN114384931B (en) Multi-target optimal control method and equipment for unmanned aerial vehicle based on strategy gradient
CN116106751A (en) Lithium ion battery state of charge estimation method based on Informar
CN109146145A (en) BP neural network building energy consumption prediction technique based on MapReduce
CN115062528A (en) Prediction method for industrial process time sequence data
CN114943277A (en) Method for identifying operation condition and predicting operation performance of comprehensive energy system
CN109543263B (en) Method for establishing integrated atmospheric distillation process proxy model
CN113788007A (en) Layered real-time energy management method and system
Sun et al. Application of neural network model combining information entropy and ant colony clustering theory for short-term load forecasting
CN111369046A (en) Wind-solar complementary power prediction method based on grey neural network

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: 20896478

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: 20896478

Country of ref document: EP

Kind code of ref document: A1