WO2024059965A1 - 基于双通道信息互补融合堆叠自编码器的产品质量预测方法 - Google Patents

基于双通道信息互补融合堆叠自编码器的产品质量预测方法 Download PDF

Info

Publication number
WO2024059965A1
WO2024059965A1 PCT/CN2022/119575 CN2022119575W WO2024059965A1 WO 2024059965 A1 WO2024059965 A1 WO 2024059965A1 CN 2022119575 W CN2022119575 W CN 2022119575W WO 2024059965 A1 WO2024059965 A1 WO 2024059965A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
fusion
channel
dual
product quality
Prior art date
Application number
PCT/CN2022/119575
Other languages
English (en)
French (fr)
Inventor
张新民
朱泓宇
何柏村
宋执环
Original Assignee
浙江大学
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 浙江大学 filed Critical 浙江大学
Priority to PCT/CN2022/119575 priority Critical patent/WO2024059965A1/zh
Priority to US18/354,595 priority patent/US20240094716A1/en
Publication of WO2024059965A1 publication Critical patent/WO2024059965A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/4183—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by data acquisition, e.g. workpiece identification
    • 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
    • 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
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32181—Monitor production, assembly apparatus with multiple sensors
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32193—Ann, neural base quality management
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/32—Operator till task planning
    • G05B2219/32194—Quality prediction

Definitions

  • the invention belongs to the field of industrial process control, and in particular relates to a product quality prediction method based on dual-channel information complementary fusion stacked autoencoders.
  • the data describes the real situation of each production stage of manufacturing, provides valuable data resources for understanding, analyzing and optimizing the manufacturing process, and is the source of intelligence for realizing intelligent manufacturing. Therefore, how to rationally use the data information accumulated in the manufacturing process to establish an intelligent analysis model suitable for industrial processes to better serve the intelligent decision-making and quality control of the manufacturing process has become a hot issue in the industry.
  • models suitable for industrial process product quality prediction can be mainly divided into mechanism-based modeling, expert experience-based modeling and data-driven modeling. Since industrial process systems are often complex black box states, in recent years, data Driving models have become the main research direction, and they also have better prediction effects.
  • the data-driven quality prediction method uses intelligent analysis technologies such as machine learning and deep learning to deeply mine, model and analyze industrial data to provide users and industry with real-time and accurate product quality prediction results.
  • the data-driven model can solve the problem of unclear mechanism of the object under study.
  • Data-driven models can be built using only one algorithm or integrated using multiple algorithms.
  • the algorithm used in the data-driven model can flexibly adjust the internal hyperparameters according to actual problems or use optimization algorithms to improve it.
  • Data-driven models are usually established using historical data, aiming to describe the internal operating rules and related pattern differences of the research object, and can be combined with parameter input from online data to achieve online output of the model.
  • the purpose of the present invention is to provide a product quality prediction method based on dual-channel information complementary fusion stacked autoencoders in view of the shortcomings of the existing technology.
  • the method includes the following steps:
  • a product quality prediction method based on dual-channel information complementary fusion stacked autoencoders includes the following steps:
  • Step 1 Collect sensor data and product quality data of industrial processes to obtain a training data set Among them, x represents the input sample, y represents the sample label, and M represents the number of labeled samples;
  • Step 2 Construct a dual-channel information complementary fusion stacked autoencoder model for product quality prediction, and use the training set to train the dual-channel information complementary fusion stacked autoencoder model;
  • the dual-channel information complementary fusion stacked autoencoder model includes a stacked autoencoder, a top-down information transfer and fusion channel, a bottom-up information transfer and fusion channel, and a gating structure;
  • the top-down information transmission and fusion channel includes multiple information complementary fusion modules; the information complementary fusion module extracts the hidden layer information of the stacked autoencoder, performs information fusion, and performs regression on the fused information to obtain the corresponding The output value of the information complementary fusion module; at the same time, the fusion information is transferred from the upper layer to the lower layer;
  • the bottom-up information transmission and fusion channel includes multiple information complementary fusion modules; the information complementary fusion module extracts the hidden layer information of the stacked autoencoder, performs information fusion, and performs regression on the fused information to obtain Corresponds to the output value of the information complementary fusion module; at the same time, the fusion information is transferred from the lower layer to the upper layer;
  • the gate control structure is used to collect fusion information and output values of multiple information complementary fusion modules in two information transmission and fusion channels, then use the fusion information to calculate the weight of the output value, and finally use the weight and output value to perform weighted fusion. Calculate the overall prediction result of the model;
  • Step 3 Collect industrial field work data and input the trained dual-channel information complementary fusion stacked autoencoder model to output the corresponding product quality prediction results.
  • the stacked autoencoder structure includes a multi-layer encoder and a multi-layer decoder, and the output of each layer of the stacked autoencoder is a reconstruction of the input;
  • W (e) , W (d) are the weight coefficient matrices of the encoder and decoder, b (e) , b (d) are the deviation coefficient matrices of the encoder and decoder;
  • the output of the upper layer of encoder is used as the input of the next layer of encoder, and its calculation formula is as follows:
  • L represents the number of hidden layers, and is the weight coefficient matrix and bias coefficient matrix of the corresponding hidden layer.
  • the information complementary fusion module includes a dimension reduction fusion channel, an ascending dimension fusion channel and a regressor, high-dimensional information passes through the dimensionality reduction fusion channel, and low-dimensional information passes through the ascending dimension fusion channel;
  • f DR is the function of dimensionality reduction fusion channel
  • h i high-dimensional information
  • h j is low-dimensional information
  • W i DR are the weight coefficient matrix and the deviation coefficient matrix respectively;
  • fDA is the dimension-upgrading fusion channel function, is the information after h j is upgraded to a new dimension, and They are weight coefficient matrix and bias coefficient matrix respectively; The information after fusion;
  • the fused information is passed through the regressor to obtain the corresponding module output value.
  • Its calculation formula is as follows:
  • f R is the regressor function, W i R , and are the weight coefficient matrix and bias coefficient matrix of the regressor function respectively.
  • f ICFB represents the functional expression of the information complementary fusion module
  • g k is the gating function
  • W g,k and b g,k are the weight coefficient matrix and deviation coefficient matrix output by the corresponding module
  • the gated structure uses weights and output values to perform weighted fusion calculations to obtain the overall prediction result of the model, and the calculation expression is as follows:
  • the stochastic gradient descent algorithm is used to train the dual-channel information complementary fusion stacked autoencoder.
  • the model training loss function is the mean square error function (MSE), and the expression is as follows:
  • xi represents the i-th data
  • the present invention proposes a product quality prediction method based on dual-channel information complementary fusion and stacked autoencoders.
  • This method stacks autoencoders.
  • a top-down information transfer and fusion channel, a bottom-up information transfer and fusion channel, and a gate control structure are designed externally.
  • Both information transfer and fusion channels include multiple information complementary fusion modules, which can stack auto-encoding
  • the hidden layer information inside the device is extracted and fused and transferred in two directions to enhance the utilization of effective information and suppress noise; furthermore, the output value and output information of the information complementary fusion module are calculated and weighted through the gating structure Fusion improves the adaptive ability and prediction accuracy of the model.
  • Figure 1 is a schematic diagram of an autoencoder
  • Figure 2 is a schematic diagram of the stacked autoencoder
  • Figure 3 is a schematic diagram of the information complementary fusion module
  • Figure 4 is a schematic diagram of a dual-channel information complementary fusion stacked autoencoder
  • Figure 5 is a schematic structural diagram of the gate control module
  • FIG6 is a graph showing the product quality prediction results of different models.
  • the present invention proposes a product quality prediction method based on dual-channel information complementary fusion and stacked autoencoders.
  • This method first adopts stacking The design of the autoencoder receives and trains the input data information, and obtains the information of the hidden layer of the stacked autoencoder. After that, outside the stacked autoencoder, a top-down information transfer and fusion channel, a bottom-up information transfer and fusion channel, and a gating structure were designed.
  • This structural design can transfer the hidden layer information inside the stacked autoencoder. Extract and perform fusion passes in both directions.
  • module output value and output information are calculated and weighted fused through the gate control structure to obtain the final product quality prediction result.
  • the design of the model enables the model to extract and utilize information more effectively, enhance the utilization of effective information, suppress noise, and improve the effect of product quality prediction.
  • Step 1 Collect sensor data and product quality data of industrial processes to obtain a training data set Among them, x represents the input sample, y represents the sample label, and M represents the number of labeled samples;
  • Step 2 Construct a dual-channel information complementary fusion stacked autoencoder model for product quality prediction, and use the industrial process data sample set to train the dual-channel information complementary fusion stacked autoencoder model;
  • J AE ( ⁇ ) is the mean square error formula used in the training process of a single-layer autoencoder. During training, the J AE ( ⁇ ) result is required to be as close to 0 as possible.
  • the output of the upper layer encoder can be used as the output of the next layer encoder, as shown in Figure 2, and its calculation formula is as follows:
  • L represents the number of hidden layers, and is the weight coefficient matrix and bias coefficient matrix of the corresponding hidden layer.
  • the information complementary fusion module includes a dimensionality reduction fusion channel, a dimensionality enhancement fusion channel and a regressor. As shown in Figure 3, the information complementary fusion module extracts and collects two different information, where h i is high-dimensional information and h j is low-dimensional information.
  • High-dimensional information is reduced to a dimension equivalent to the low-dimensional information through the dimensionality reduction fusion channel, and the low-dimensional information is upgraded to the dimension of the high-dimensional information through the dimension-raising fusion channel, and the unity of dimensions is achieved through the two channels, and then Fusion of information of the same dimension is performed to obtain the fused information.
  • the fused information will include both high-dimensional information and low-dimensional information;
  • f DR is the function of dimensionality reduction fusion channel, is the information after dimensionality reduction of h i , is the fused information;
  • W i DR and are respectively the weight coefficient matrix and the bias coefficient matrix in the dimensionality reduction fusion channel formula;
  • f DA is the function of the ascending dimension fusion channel, is the information after h j is dimensioned, For the information after fusion, and are respectively the weight coefficient matrix and the bias coefficient matrix in the ascending dimension fusion channel formula;
  • the fused information is passed through the regressor to obtain the corresponding module output value.
  • Its calculation formula is as follows:
  • f R is the regressor function, W i R , and are the weight coefficient matrix and bias coefficient matrix in the regressor function respectively.
  • the present invention completes the information extraction and fusion of the entire stack auto-encoding by designing top-down information transfer and fusion channels and bottom-up information transfer and fusion channels. and directional transmission. The specific structures of these two parts of the channel are shown in Figure 4.
  • the expressions of the bottom-up information transmission and the fusion channel for information transmission are:
  • f ICFB represents the functional expression of the information complementary fusion module.
  • the information of the bottom two adjacent hidden layers h 1 and h 2 is first extracted by the bottom layer information complementary fusion module.
  • the fused information is obtained through fusion
  • the lowest layer information fusion module passes the fusion information to the second information complementary fusion module from bottom to top.
  • the second information complementary fusion module extracts the third layer hidden layer information h 3 and puts it into After information fusion with h 3 , it is passed to the third information complementary fusion module from bottom to top.
  • the hidden layer information in the stacked autoencoder completes the process of being extracted, fused and transmitted from the lower layer to the upper layer;
  • f ICFB represents the functional expression of the information complementary fusion module.
  • the uppermost information complementary fusion module first extracts the values of the two adjacent hidden layers h L and h L-1 in the top layer. Information, after fusion, the fused information is obtained Then the top-level information fusion module passes the fusion information to the second information complementary fusion module from top to bottom. At the same time, the second information complementary fusion module extracts the L-2 hidden layer information h L-2 , and After information fusion with h L-2 , it is passed to the third information complementary fusion module from top to bottom.
  • the hidden layer information in the stacked autoencoder is extracted, fused and transmitted from the upper layer to the lower layer. the process of;
  • g k is the gating function
  • W g,k and b g,k are the weight coefficient matrix and deviation coefficient matrix output by the corresponding module
  • the overall prediction structure includes two parts. One part is the output value of each information complementary fusion module in the top-down information fusion and transmission channel. and the weighted fusion of its calculated weight, and the other part is the weighted fusion of the output value of each information complementary fusion module in the bottom-up information fusion and transmission channel and its calculated weight.
  • the gate control structure uses the weight sum The output values are weighted and fused to calculate the overall prediction result of the model.
  • the calculation expression is as follows:
  • xi represents the i-th data
  • Step three Collect industrial field work data and input it into the dual-channel information complementary fusion stacked autoencoder model for product quality prediction, and output the corresponding product quality prediction results.
  • the effectiveness of the method of the present invention is verified below with a specific industrial process example.
  • sensor data from blast furnace ironmaking in the actual production process were collected and data collected.
  • the data set has a total of 110 process variables.
  • the quality variable to be predicted is silicon content. Silicon content is one of the important hot metal quality indicators in the blast furnace ironmaking process and can well reflect the operating status and product quality of the blast furnace ironmaking industrial system. .
  • the collected data includes a total of 20,000 pieces of data, of which 12,000 pieces of data are used as the training set, 4,000 pieces of data are used as the verification set, and 4,000 pieces of data are used as the test set.
  • the training data set is used for model training
  • the validation data set is used for model parameter selection
  • the test data set is used for model testing.
  • One of the evaluation criteria is the root mean square error, which is defined as follows:
  • N represents the total data volume of the test set, yi and are the true values and predicted values of the output quality variables respectively.
  • Another evaluation criterion is the coefficient of determination, referred to as R 2 , which is defined as follows:
  • Table 1 The test results of the comparative experiment are shown in Table 1, where SAE represents the stacked autoencoder model, GSAE represents the gated multi-expert hybrid stacked autoencoder model, and DC-ICF-SAE represents the dual-channel information complementary fusion stacked autoencoder of the present invention. device model.
  • SAE represents the stacked autoencoder model
  • GSAE represents the gated multi-expert hybrid stacked autoencoder model
  • DC-ICF-SAE represents the dual-channel information complementary fusion stacked autoencoder of the present invention. device model.
  • Table 1 As can be seen from Table 1, compared with the SAE and GSAE models, the DC-ICF-SAE model proposed by the present invention obtained the smallest RMSE and the largest R 2 , indicating that the method of the present invention has better prediction performance.
  • Figure 6 shows the product quality prediction result curves of SAE, GSAE and DC-ICF-SAE models.
  • the predicted values output by the SAE and GSAE models are consistent with the actual measured values, and have better prediction effects and smaller prediction deviations in the parts where the quality variable values change rapidly.
  • Experimental results show that the product quality prediction method proposed by the present invention based on dual-channel information complementary fusion stacked autoencoders can better process data information, extract more effective information, and have better prediction effects on product quality.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Manufacturing & Machinery (AREA)
  • Quality & Reliability (AREA)
  • Automation & Control Theory (AREA)
  • Theoretical Computer Science (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，该方法首先采用了堆叠自编码器的设计，对输入的数据信息进行接收和训练，得到了堆叠自编码器隐藏层的信息。之后，在堆叠自编码器外部，设计了信息互补融合模块与底层到顶层与顶层到底层双通道信息融合层的结构，这种结构设计能够将堆叠自编码器内部隐藏层信息提取并且进行在两个方向上的融合传递。再者，通过门控结构将模块输出值和输出信息进行计算并进行加权融合，得到最终的产品质量预测结果。该方法在对数据信息的处理上，提取出了更有效的信息，减少了噪声，提高了对信息的利用效率，有着更好的对产品质量的预测能力。

Description

基于双通道信息互补融合堆叠自编码器的产品质量预测方法 技术领域
本发明属于工业过程控制领域，特别涉及一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法。
背景技术
随着现代工业技术的不断发展进步，如何保证工业生产过程中的产品质量成为了工业过程控制领域重点关注和需要解决的关键问题之一。在工业生产过程中，对产品质量变量的实时测量和准确预报对从业人员非常重要，能够通过质量变量的变化对生产过程状态进行实时的监控，对过程控制策略进行及时地调整，同时也能够达到降低生产成本、提高生产效率、优化产品质量等。随着现代测量手段的不断发展和进步，可以从过程中测量得到更多的数据，工业生产过程积累了大量的数据。数据描述了制造各生产阶段的真实情况，为读懂、分析和优化制造过程提供了宝贵的数据资源，是实现智能制造的智能来源。因此，如何合理地利用制造过程积累的数据信息，建立适用于工业过程的智能分析模型，以更好地为制造过程的智能决策与质量控制服务，是工业界较为关注的热点问题。现阶段适用于工业过程产品质量预测的模型主要可以分为基于机理的建模，基于专家经验的建模以及基于数据驱动的建模，由于工业过程系统往往是复杂的黑箱状态，因此近年来数据驱动模型成为主要的研究方向，同时也有着更好的预测效果。数据驱动的质量预测方法利用机器学习、深度学习等智能分析技术，对工业数据深入挖掘、建模和分析，为用户和工业提供实时准确的产品质量预测结果。数据驱动模型作为一种黑箱模型可以解决所研究对象机理模糊不清的问题。数据驱动模型可以仅采用一种算法建立，也可以采用多种算法集成建立。同时，数据驱动模型采用的算法可以根据实际问题灵活调整内部超参数或利用优化算法进行改良。数据驱动模型通常利用历史数据建立，旨在描述研究对象的内部运行规律和相关模式区别，并可结合在线数据的参数输入实现模型的在线输出。但是，现有的一些数据驱动的建模方法，在对工业过程数据信息的处理和利用上并不完善，有效信息利用率较低。因此，亟需提供一种有更好的对数据信息利用率的产品质量预测方法，使得模型能够更有效地提取有效信息，抑制噪声，提高对产品质量预测的精度和效率。
发明内容
本发明的目的在于针对现有技术的不足，提供一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，该方法包括如下步骤：
一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，该方法包括以下步骤：
步骤一：收集工业过程的传感器数据与产品质量数据，得到训练数据集
Figure PCTCN2022119575-appb-000001
其中，x代表输入样本，y代表样本标签，M表示有标签样本个数；
步骤二：构建用于产品质量预测的双通道信息互补融合堆叠自编码器模型，并利用所述训练集对双通道信息互补融合堆叠自编码器模型进行训练；
所述双通道信息互补融合堆叠自编码器模型包括堆叠自编码器、自上而下信息传递和融合通道、自下而上信息传递和融合通道以及一个门控结构；
所述自上而下信息传递和融合通道包括多个信息互补融合模块；所述信息互补融合模块提取所述堆叠自编码器的隐藏层信息，进行信息融合，并对融合信息进行回归，得到对应信息互补融合模块的输出值；同时将融合信息从上层向下层传递；
所述自下而上的信息传递和融合通道包括多个信息互补融合模块；所述信息互补融合模块提取所述堆叠自编码器的隐藏层信息，进行信息融合，并对融合信息进行回归，得到对应信息互补融合模块的输出值；同时将融合信息从下层向上层传递；
所述门控结构用于收集两个信息传递和融合通道中多个信息互补融合模块的融合信息和输出值，然后利用融合信息计算所述输出值的权重，最后利用权重和输出值进行加权融合计算得到模型的总体预测结果；
步骤三：采集工业现场工作数据并输入训练后的双通道信息互补融合堆叠自编码器模型，输出对应的产品质量预测结果。
进一步地，所述步骤二中，所述堆叠自编码器结构包含多层编码器和多层解码器，堆叠自编码器每层的输出是对输入的重构；
对单层的自编码器，计算公式如下：
Figure PCTCN2022119575-appb-000002
Figure PCTCN2022119575-appb-000003
其中，
Figure PCTCN2022119575-appb-000004
为编码器函数，
Figure PCTCN2022119575-appb-000005
为解码器函数，x代表输入的数据集，h代表编码器的输出同时为解码器的输入，
Figure PCTCN2022119575-appb-000006
为解码器的输出；W (e),W (d)为编码器和解码器的权重系数矩阵，b (e),b (d)为编码器和解码器的偏差系数矩阵；
对于堆叠的自编码器来说，上一层编码器的输出作为下一层编码器的输入，其计算公式如下：
Figure PCTCN2022119575-appb-000007
其中，L代表隐藏层的层数，
Figure PCTCN2022119575-appb-000008
和
Figure PCTCN2022119575-appb-000009
为对应隐藏层的权重系数矩阵和偏差系数矩阵。
进一步地，所述信息互补融合模块包括降维融合通道、升维融合通道和回归器，高维信息通过降维融合通道，低维信息通过升维融合通道；
其中，降维融合的计算公式为：
Figure PCTCN2022119575-appb-000010
Figure PCTCN2022119575-appb-000011
其中，f DR为降维融合通道的函数，h i为高维信息，h j为低维信息，
Figure PCTCN2022119575-appb-000012
为h i降维之后的信息，
Figure PCTCN2022119575-appb-000013
为融合之后的信息；W i DR与
Figure PCTCN2022119575-appb-000014
分别为权重系数矩阵与偏差系数矩阵；
升维融合的计算公式为：
Figure PCTCN2022119575-appb-000015
Figure PCTCN2022119575-appb-000016
其中，f DA为升维融合通道函数，
Figure PCTCN2022119575-appb-000017
为h j升维之后的信息，
Figure PCTCN2022119575-appb-000018
与
Figure PCTCN2022119575-appb-000019
分别为权重系数矩阵与偏差系数矩阵；
Figure PCTCN2022119575-appb-000020
为融合之后的信息；
融合之后的信息经过回归器得到对应的模块输出值
Figure PCTCN2022119575-appb-000021
和
Figure PCTCN2022119575-appb-000022
其计算公式如下所示：
Figure PCTCN2022119575-appb-000023
Figure PCTCN2022119575-appb-000024
其中，f R为回归器函数，W i R，
Figure PCTCN2022119575-appb-000025
和
Figure PCTCN2022119575-appb-000026
分别为回归器函数的权重系数矩阵与偏差系数矩阵。
进一步地，所述自下而上信息传递和融合通道进行信息的传递的表达式为：
Figure PCTCN2022119575-appb-000027
Figure PCTCN2022119575-appb-000028
其中，f ICFB代表了信息互补融合模块的函数表达；
所述自上而下信息传递和融合通道进行信息的传递的表达式为：
Figure PCTCN2022119575-appb-000029
Figure PCTCN2022119575-appb-000030
进一步地，所述门控结构利用融合信息计算所述输出值的权重的计算表达如下所示：
Figure PCTCN2022119575-appb-000031
其中，g k为门控函数，W g,k和b g,k为对应模块输出的权重系数矩阵与偏差系数矩阵；
所述门控结构利用权重和输出值进行加权融合计算得到模型的总体预测结果的计算表达式如下：
Figure PCTCN2022119575-appb-000032
进一步地，所述步骤二中，采用随机梯度下降算法对所述双通道信息互补融合堆叠自编码器进行训练，其模型训练损失函数为均方误差函数(MSE)，表达式如下所示：
Figure PCTCN2022119575-appb-000033
其中，x i代表第i个数据，
Figure PCTCN2022119575-appb-000034
代表堆叠自编码器对数据的重构。
本发明的有益效果如下：
本发明在针对现有产品质量预测模型在对工业过程数据信息的利用率较低的问题，提出了一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，该方法堆叠自编码器外部设计了自上而下信息传递和融合通道、自下而上信息传递和融合通道以及一个门控结构，两个信息传递和融合通道中均包括多个信息互补融合模块，能够将堆叠自编码器内部隐藏层信息提取并且进行在两个方向上的融合传递，加强对有效信息的利用率，抑制噪声；再者，通过门控结构将信息互补融合模块输出值和输出信息进行计算并进行加权融合，提高了模型的自适应能力和预测精度。
附图说明
图1为自编码器示意图；
图2为堆叠自编码器示意图；
图3为信息互补融合模块示意图；
图4为双通道信息互补融合堆叠自编码器示意图；
图5为门控模块结构示意图；
图6为不同模型的产品质量预测结果曲线图。
具体实施方式
下面根据附图和优选实施例详细描述本发明，本发明的目的和效果将变得更加明白，应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
本发明在针对现有产品质量预测模型在对工业过程数据信息的利用率较低的问题，提出了一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，该方法首先采用了堆叠自编码器的设计，对输入的数据信息进行接收和训练，得到了堆叠自编码器隐藏层的信息。之后，在堆叠自编码器外部，设计了自上而下信息传递和融合通道、自下而上信息传递和融合通道以及一个门控结构，这种结构设计能够将堆叠自编码器内部隐藏层信息提取并且进行在两个方向上的融合传递。再者，通过门控结构将模块输出值和输出信息进行计算并进行加权融合，得到最终的产品质量预测结果。模型的设计使得模型能够更有效地进行信息提取和利用，加强对有效信息的利用率，抑制噪声，提高了产品质量预测的效果。
本发明的方法具体步骤如下：
步骤一：收集工业过程的传感器数据与产品质量数据，得到训练数据集
Figure PCTCN2022119575-appb-000035
其中，x代表输入样本，y代表样本标签，M表示有标签样本个数；
步骤二：构建用于产品质量预测的双通道信息互补融合堆叠自编码器模型，并利用工业过程数据样本集对双通道信息互补融合堆叠自编码器模型进行训练；
(2.1)构建堆叠自编码器模型，包含多层编码器和解码器，模型输出是对输入的重构，对单层的自编码器，如图1所示，计算公式如下：
Figure PCTCN2022119575-appb-000036
Figure PCTCN2022119575-appb-000037
Figure PCTCN2022119575-appb-000038
其中，
Figure PCTCN2022119575-appb-000039
为编码器函数，
Figure PCTCN2022119575-appb-000040
为解码器函数，x代表输入的数据集，h代表了编码器的输出同时为解码器的输入，
Figure PCTCN2022119575-appb-000041
为解码器的输出。W (e),W (d)为编码器和解码器的权重系数矩阵，b (e),b (d)为编码器和解码器的偏差系数矩阵。J AE(θ)为单层自编码器训练过程中使用的均方误差公式，训练时要求J AE(θ)结果趋近于0越好。对于堆叠的自编码器来说，上一层编码器的输出可以作为下一层编码器的输出，如图2所示，其计算公式如下：
Figure PCTCN2022119575-appb-000042
其中，L代表隐藏层的层数，
Figure PCTCN2022119575-appb-000043
和
Figure PCTCN2022119575-appb-000044
为 对应隐藏层的权重系数矩阵和偏差系数矩阵。
(2.2)建立信息互补融合模块进行对堆叠自编码器中隐藏层信息的提取和融合，并进行融合信息的进一步传递。所述信息互补融合模块包括降维融合通道、升维融合通道和回归器，如图3所示，信息互补融合模块提取收集两个不同的信息，其中h i为高维信息，h j为低维信息，高维信息通过降维融合通道降维至等同于低维信息的维度，低维信息通过升维融合通道升维至高维信息的维度，通过两个通道来达到维度上的统一，再进行相同维度信息的融合来得到融合之后的信息，融合后的信息会同时包括高维信息和低维信息中两者的信息；
其中，降维融合通道的计算公式为：
Figure PCTCN2022119575-appb-000045
Figure PCTCN2022119575-appb-000046
其中，f DR为降维融合通道的函数，
Figure PCTCN2022119575-appb-000047
为h i降维之后的信息，
Figure PCTCN2022119575-appb-000048
为融合之后的信息；W i DR与
Figure PCTCN2022119575-appb-000049
分别为降维融合通道公式中的权重系数矩阵与偏差系数矩阵；
此外，升维融合通道的计算公式为：
Figure PCTCN2022119575-appb-000050
Figure PCTCN2022119575-appb-000051
其中，f DA为升维融合通道的函数，
Figure PCTCN2022119575-appb-000052
为h j升维之后的信息，
Figure PCTCN2022119575-appb-000053
为融合之后的信息，
Figure PCTCN2022119575-appb-000054
与
Figure PCTCN2022119575-appb-000055
分别为升维融合通道公式中的权重系数矩阵与偏差系数矩阵；
融合之后的信息经过回归器得到对应的模块输出值
Figure PCTCN2022119575-appb-000056
和
Figure PCTCN2022119575-appb-000057
其计算公式如下所示：
Figure PCTCN2022119575-appb-000058
Figure PCTCN2022119575-appb-000059
其中，f R为回归器函数，W i R，
Figure PCTCN2022119575-appb-000060
和
Figure PCTCN2022119575-appb-000061
分别为回归器函数中的权重系数矩阵与偏差系数矩阵。
(2.3)对于信息互补融合模块提取融合得到的融合信息，本发明通过设计自上而下信息传递和融合通道以及自下而上信息传递和融合通道来完成对整个堆叠自编码的信息提取，融合和方向性上的传递，这两部分通道的具体结构如图4所示，所述自下而上信息传递和融合通道进行信息的传递的表达式为：
Figure PCTCN2022119575-appb-000062
Figure PCTCN2022119575-appb-000063
其中，f ICFB代表了信息互补融合模块的函数表达，自下而上的信息融合和传递通道中，首先由最下层信息互补融合模块提取最下层两个相邻隐藏层h 1和h 2的信息，经过融合得到融合信息
Figure PCTCN2022119575-appb-000064
之后最下层信息融合模块将融合信息传递给从下往上的第二信息互补融合模块中，同时第二信息互补融合模块提取第三层隐藏层信息h 3，将
Figure PCTCN2022119575-appb-000065
和h 3进行信息融合后传递给从下往上的第三信息互补融合模块，以此类推，堆叠自编码器中的隐藏层信息从下层开始至上层完成了被提取，融合和传递的过程；
所述自上而下信息传递和融合通道进行信息的传递的表达式为：
Figure PCTCN2022119575-appb-000066
Figure PCTCN2022119575-appb-000067
其中，f ICFB代表了信息互补融合模块的函数表达，自上而下信息融合和传递通道中，首先由最上层信息互补融合模块提取最上层两个相邻隐藏层h L和h L-1的信息，经过融合得到融合信息
Figure PCTCN2022119575-appb-000068
之后最上层的信息融合模块将融合信息传递给从上往下第二信息互补融合模块中，同时第二信息互补融合模块提取第L-2层隐藏层信息h L-2，将
Figure PCTCN2022119575-appb-000069
和h L-2进行信息融合后，传递给从上往下的第三信息互补融合模块，以此类推，堆叠自编码器中的隐藏层信息从上层开始至下层完成了被提取，融合和传递的过程；
(2.4)门控结构利用融合信息计算所述输出值的权重的计算表达如下所示：
Figure PCTCN2022119575-appb-000070
其中，g k为门控函数，W g,k和b g,k为对应模块输出的权重系数矩阵与偏差系数矩阵；
门控结构的具体结构如附图5所示，在计算模型总体预测结果时，总的预测结构包括两部分，一部分为自上而下信息融合和传递通道中每一个信息互补融合模块的输出值和其计算得到的权重的加权融合，另一部分为自下而上信息融合和传递通道中每一个信息互补融合模块的输出值和其计算得到的权重的加权融合，所述门控结构利用权重和输出值进行加权融合计算得到模型的总体预测结果的计算表达式如下：
Figure PCTCN2022119575-appb-000071
其中，
Figure PCTCN2022119575-appb-000072
为自下而上信息融合和传递通道中每一个信息互补融合模块的输出值和其计算得到的权重的加权融合；
Figure PCTCN2022119575-appb-000073
为自上而下信息融合和传递通道中每一个信息互补融合模块的输出值和其计算得到的权重的加权融合。
(2.5)采用步骤一构建的工业过程样本数据集，采用随机梯度下降算法对所述双通道门控多专家混合堆叠自编码器模型进行训练，其模型训练损失函数为均方误差函数(MSE)，表达式如下所示：
Figure PCTCN2022119575-appb-000074
其中，x i代表第i个数据，
Figure PCTCN2022119575-appb-000075
代表堆叠自编码器对数据的重构。
步骤三：采集工业现场工作数据并输入所述用于产品质量预测的双通道信息互补融合堆叠自编码器模型，输出对应的产品质量预测结果。
下面以一个具体工业过程实例验证本发明的方法的有效性。在实验中采集了实际生产过程中高炉炼铁中的传感器数据并进行了数据收集。数据集一共有110个过程变量，待预测的质量变量为硅含量，硅含量是高炉炼铁过程中重要的铁水质量指标之一，能够很好地反映高炉炼铁工业系统的运行状态和产品质量。
采集到的数据一共包括20000条数据，其中12000条数据作为训练集，4000条数据作为验证集，4000条数据作为测试集。训练数据集用于模型训练，验证数据集用于模型参数选择，测试数据集用于模型测试。为了评价所设计产品质量预测模型的预测性能，实验过程中使用的评价标准主要有两个，其中一个评价标准为均方根误差，其定义如下：
Figure PCTCN2022119575-appb-000076
其中，N代表测试集的总数据量，y i与
Figure PCTCN2022119575-appb-000077
分别为输出的质量变量的真实值和预测值。另一个评价标准为决定系数，简称为R 2，其定义如下：
Figure PCTCN2022119575-appb-000078
其中，
Figure PCTCN2022119575-appb-000079
代表了质量变量的平均值，对于上述的评价标准，RMSE取值越小和R 2取值越大，表示模型的预测性能越高。
对比实验的测试结果如表1所示，其中SAE代表堆叠自编码器模型，GSAE代表门控多专家混合堆叠自编码器模型，DC-ICF-SAE代表本发明的双通道信息互补融合堆叠自编码器模型。从表1可以看出，与SAE和GSAE模型相比，本发明所提出的DC-ICF-SAE模型获得了最小的RMSE和最大的R 2，表明本发明方法具有更好的预测性能。
表1
模型 RMSE R 2
SAE 0.0401 0.9041
GSAE 0.0389 0.9185
DC-ICF-SAE 0.0286 0.9504
此外，图6给出了SAE、GSAE和DC-ICF-SAE模型的产品质量预测结果曲线图。从图6可以看出，SAE和GSAE模型输出的预测值与实际测量值之间存在比较明显的偏差，并且在质量变量数值变化较快的地方预测效果有较为明显的下降。相比之下，本发明提出的DC-ICF-SAE模型输出的预测值与实际测量值的吻合度较高，并且在质量变量数值变化较快的部分有更好的预测效果，预测偏差更小。由实验结果说明，本发明提出的基于双通道信息互补融合堆叠自编码器的产品质量预测方法更好地对数据信息进行了处理，提取了更有效的信息，对产品质量的预测效果更优。
本领域普通技术人员可以理解，以上所述仅为发明的优选实例而已，并不用于限制发明，尽管参照前述实例对发明进行了详细的说明，对于本领域的技术人员来说，其依然可以对前述各实例记载的技术方案进行修改，或者对其中部分技术特征进行等同替换。凡在发明的精神和原则之内，所做的修改、等同替换等均应包含在发明的保护范围之内。

Claims (6)

  1. 一种基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，该方法包括以下步骤：
    步骤一：收集工业过程的传感器数据与产品质量数据，得到训练数据集
    Figure PCTCN2022119575-appb-100001
    其中，x代表输入样本，y代表样本标签，M表示有标签样本个数；
    步骤二：构建用于产品质量预测的双通道信息互补融合堆叠自编码器模型，并利用所述训练集对双通道信息互补融合堆叠自编码器模型进行训练；
    所述双通道信息互补融合堆叠自编码器模型包括堆叠自编码器、自上而下信息传递和融合通道、自下而上信息传递和融合通道以及一个门控结构；
    所述自上而下信息传递和融合通道包括多个信息互补融合模块；所述信息互补融合模块提取所述堆叠自编码器的隐藏层信息，进行信息融合，并对融合信息进行回归，得到对应信息互补融合模块的输出值；同时将融合信息从上层向下层传递；
    所述自下而上的信息传递和融合通道包括多个信息互补融合模块；所述信息互补融合模块提取所述堆叠自编码器的隐藏层信息，进行信息融合，并对融合信息进行回归，得到对应信息互补融合模块的输出值；同时将融合信息从下层向上层传递；
    所述门控结构用于收集两个信息传递和融合通道中多个信息互补融合模块的融合信息和输出值，然后利用融合信息计算所述输出值的权重，最后利用权重和输出值进行加权融合计算得到模型的总体预测结果；
    步骤三：采集工业现场工作数据并输入训练后的双通道信息互补融合堆叠自编码器模型，输出对应的产品质量预测结果。
  2. 根据权利要求1所述的基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，所述步骤二中，所述堆叠自编码器结构包含多层编码器和多层解码器，堆叠自编码器每层的输出是对输入的重构；
    对单层的自编码器，计算公式如下：
    Figure PCTCN2022119575-appb-100002
    Figure PCTCN2022119575-appb-100003
    其中，
    Figure PCTCN2022119575-appb-100004
    为编码器函数，
    Figure PCTCN2022119575-appb-100005
    为解码器函数，x代表输入的数据集，h代表编码器的输出同时为解码器的输入，
    Figure PCTCN2022119575-appb-100006
    为解码器的输出；W (e),W (d)为编码器和解码器的权重系数矩阵， b (e),b (d)为编码器和解码器的偏差系数矩阵；
    对于堆叠的自编码器来说，上一层编码器的输出作为下一层编码器的输入，其计算公式如下：
    Figure PCTCN2022119575-appb-100007
    其中，L代表隐藏层的层数，
    Figure PCTCN2022119575-appb-100008
    和
    Figure PCTCN2022119575-appb-100009
    为对应隐藏层的权重系数矩阵和偏差系数矩阵。
  3. 根据权利要求2所述的基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，所述信息互补融合模块包括降维融合通道、升维融合通道和回归器，高维信息通过降维融合通道，低维信息通过升维融合通道；
    其中，降维融合的计算公式为：
    Figure PCTCN2022119575-appb-100010
    Figure PCTCN2022119575-appb-100011
    其中，f DR为降维融合通道的函数，h i为高维信息，h j为低维信息，
    Figure PCTCN2022119575-appb-100012
    为h i降维之后的信息，
    Figure PCTCN2022119575-appb-100013
    为融合之后的信息；W i DR与
    Figure PCTCN2022119575-appb-100014
    分别为权重系数矩阵与偏差系数矩阵；
    升维融合的计算公式为：
    Figure PCTCN2022119575-appb-100015
    Figure PCTCN2022119575-appb-100016
    其中，f DA为升维融合通道函数，
    Figure PCTCN2022119575-appb-100017
    为h j升维之后的信息，W j DA与
    Figure PCTCN2022119575-appb-100018
    分别为权重系数矩阵与偏差系数矩阵；
    Figure PCTCN2022119575-appb-100019
    为融合之后的信息；
    融合之后的信息经过回归器得到对应的模块输出值
    Figure PCTCN2022119575-appb-100020
    和
    Figure PCTCN2022119575-appb-100021
    其计算公式如下所示：
    Figure PCTCN2022119575-appb-100022
    Figure PCTCN2022119575-appb-100023
    其中，f R为回归器函数，W i R，
    Figure PCTCN2022119575-appb-100024
    和
    Figure PCTCN2022119575-appb-100025
    分别为回归器函数的权重系数矩阵与偏差系数矩阵。
  4. 根据权利要求3所述的基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，
    所述自下而上信息传递和融合通道进行信息的传递的表达式为：
    Figure PCTCN2022119575-appb-100026
    Figure PCTCN2022119575-appb-100027
    其中，f ICFB代表了信息互补融合模块的函数表达；
    所述自上而下信息传递和融合通道进行信息的传递的表达式为：
    Figure PCTCN2022119575-appb-100028
    Figure PCTCN2022119575-appb-100029
  5. 根据权利要求4所述的基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，
    所述门控结构利用融合信息计算所述输出值的权重的计算表达如下所示：
    Figure PCTCN2022119575-appb-100030
    其中，g k为门控函数，W g,k和b g,k为对应模块输出的权重系数矩阵与偏差系数矩阵；
    所述门控结构利用权重和输出值进行加权融合计算得到模型的总体预测结果的计算表达式如下：
    Figure PCTCN2022119575-appb-100031
  6. 根据权利要求5所述的基于双通道信息互补融合堆叠自编码器的产品质量预测方法，其特征在于，所述步骤二中，采用随机梯度下降算法对所述双通道信息互补融合堆叠自编码器进行训练，其模型训练损失函数为均方误差函数，表达式如下所示：
    Figure PCTCN2022119575-appb-100032
    其中，x i代表第i个数据，
    Figure PCTCN2022119575-appb-100033
    代表堆叠自编码器对数据的重构。
PCT/CN2022/119575 2022-09-19 2022-09-19 基于双通道信息互补融合堆叠自编码器的产品质量预测方法 WO2024059965A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
PCT/CN2022/119575 WO2024059965A1 (zh) 2022-09-19 2022-09-19 基于双通道信息互补融合堆叠自编码器的产品质量预测方法
US18/354,595 US20240094716A1 (en) 2022-09-19 2023-07-18 Product quality prediction method based on dual-channel information complementary fusion stacked auto-encoder

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2022/119575 WO2024059965A1 (zh) 2022-09-19 2022-09-19 基于双通道信息互补融合堆叠自编码器的产品质量预测方法

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US18/354,595 Continuation US20240094716A1 (en) 2022-09-19 2023-07-18 Product quality prediction method based on dual-channel information complementary fusion stacked auto-encoder

Publications (1)

Publication Number Publication Date
WO2024059965A1 true WO2024059965A1 (zh) 2024-03-28

Family

ID=90244737

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/119575 WO2024059965A1 (zh) 2022-09-19 2022-09-19 基于双通道信息互补融合堆叠自编码器的产品质量预测方法

Country Status (2)

Country Link
US (1) US20240094716A1 (zh)
WO (1) WO2024059965A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2020102208A4 (en) * 2020-09-10 2020-10-22 C, Vijayalakshmi Mrs An intelligent embedded tool recognition system for cnc machines using deep learning algorithms
CN112257341A (zh) * 2020-10-20 2021-01-22 浙江大学 一种基于异源数据差补融合的定制产品性能预测方法
CN114004346A (zh) * 2021-08-17 2022-02-01 中国地质大学（武汉） 基于门控堆叠同构自编码器的软测量建模方法及存储介质
US20220113695A1 (en) * 2020-02-21 2022-04-14 Shandong University Rolling early warning method and system of dynamic security risk situation for large scale hybrid ac/dc grids

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220113695A1 (en) * 2020-02-21 2022-04-14 Shandong University Rolling early warning method and system of dynamic security risk situation for large scale hybrid ac/dc grids
AU2020102208A4 (en) * 2020-09-10 2020-10-22 C, Vijayalakshmi Mrs An intelligent embedded tool recognition system for cnc machines using deep learning algorithms
CN112257341A (zh) * 2020-10-20 2021-01-22 浙江大学 一种基于异源数据差补融合的定制产品性能预测方法
CN114004346A (zh) * 2021-08-17 2022-02-01 中国地质大学（武汉） 基于门控堆叠同构自编码器的软测量建模方法及存储介质

Also Published As

Publication number Publication date
US20240094716A1 (en) 2024-03-21

Similar Documents

Publication Publication Date Title
CN110378799B (zh) 基于多尺度深度卷积网络的氧化铝综合生产指标决策方法
CN106371427B (zh) 基于层次分析法和模糊融合的工业过程故障分类方法
CN103676881B (zh) 一种半导体生产线动态瓶颈分析方法
WO2023142424A1 (zh) 基于gru-lstm神经网络的电力金融业务风控方法及系统
CN111126658A (zh) 一种基于深度学习的煤矿瓦斯预测方法
Li et al. Dynamic time features expanding and extracting method for prediction model of sintering process quality index
CN103593719B (zh) 一种基于板坯与合同优化匹配的轧制节能方法
CN108388762A (zh) 基于深度置信网络的烧结矿化学成分预测方法
CN103123483B (zh) 一种基于聚类分析的新钢种轧制变形抗力预测系统
US20230080871A1 (en) Method for estimating throat temperature of blast furnace based on multilayer ore-to-coke ratio distribution model
CN104898562A (zh) 数控机床热误差补偿的建模方法
CN114015825B (zh) 基于注意力机制的高炉热负荷异常状态监测方法
CN109102002A (zh) 结合卷积神经网络和概念机递归神经网络的图像分类方法
CN109902931A (zh) 一种基于运行数据融合的多态制造系统运行风险建模方法
CN103778466B (zh) 一种基于矢量误差的转炉炼钢温度建模预测方法及系统
CN116520772A (zh) 基于多任务迁移学习的薄板零件数控加工质量预测方法
CN112862196A (zh) 一种基于模糊逻辑理论的地铁站台乘客分布预测系统
WO2024059965A1 (zh) 基于双通道信息互补融合堆叠自编码器的产品质量预测方法
CN117150232B (zh) 一种大模型非时序训练数据质量评价方法
CN104537383A (zh) 一种基于粒子群的海量组织机构数据分类方法及系统
CN110298385A (zh) 一种基于*信息与增量svdd在线早期故障检测方法
CN113705897A (zh) 一种面向工业铜箔生产的产品质量预测方法及系统
Liu et al. Temporal hypergraph attention network for silicon content prediction in blast furnace
CN112163474A (zh) 一种基于模型融合的齿轮箱智能诊断方法
CN111079832B (zh) 具有抗特征噪声性能的钢板表面缺陷分类方法

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: 22958979

Country of ref document: EP

Kind code of ref document: A1