Assembly line verification error prediction method and system based on LSTM neural network, and storage medium

Abstract

The invention discloses an assembly line verification error prediction method and system based on an LSTM neural network, and a storage medium. The method comprises the following steps: constructing a data set based on historical temperature data of assembly line verification equipment and the surrounding environment of the assembly line verification equipment and historical verification error data of the assembly line verification equipment, inputting the historical temperature data in the data set into an LSTM neural network model, and taking the historical verification error data in the data set as a label of the LSTM neural network model, the LSTM neural network model is trained; and inputting the actual temperature data of the assembly line verification equipment and the surrounding environment of the assembly line verification equipment into the trained LSTM neural network model to obtain a verification error of the assembly line verification equipment. According to the method, the monitoring data are analyzed and predicted through the neural network model, and the accuracy and efficiency of pipeline verification error prediction are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial automation, and in particular relates to a method and system for predicting assembly line calibration errors based on an LSTM neural network, a system and a storage medium. Background Art

[0002] In the industrial production process, the calibration accuracy and stability of assembly line equipment are crucial to ensuring product quality. Existing assembly line equipment calibration methods mainly rely on regular manual maintenance and traditional sensor data. However, these methods have some significant shortcomings. First, the frequency of manual maintenance is low, and the operating status of the equipment cannot be monitored in real time, which can easily lead to equipment failure during the detection cycle, thus affecting the continuity of the production line and product quality. Secondly, traditional sensors cannot fully cover the temperature changes of key parts of the equipment, and lack the ability to fully perceive the temperature field of the equipment, which makes it difficult to detect some local temperature anomalies in time and cannot warn of potential equipment failures in advance. These technical defects will affect the efficiency and accuracy of equipment calibration in the production process, leading to production stagnation and unstable product quality, which in turn affects the production efficiency and market competitiveness of the enterprise. Summary of the invention

[0003] The purpose of the present invention is to solve the deficiencies of the above-mentioned background technology and provide a method and system for predicting the error of pipeline calibration based on LSTM neural network, system and storage medium. By establishing an LSTM neural network model and using temperature field information to predict the error of equipment calibration, abnormal temperature conditions can be discovered in time, and real-time prediction of equipment calibration errors can be achieved, thereby improving the reliability and stability of equipment and ensuring the calibration quality of products on the production line. Compared with existing methods, LSTM can not only handle complex nonlinear relationships, but also has superior performance in long-term dependence on time series data.

[0004] The technical solution adopted by the present invention is: a pipeline verification error prediction method based on LSTM neural network,

[0005] A data set is constructed based on the historical temperature data of the assembly line verification equipment and the environment around the assembly line verification equipment and the historical verification error data of the assembly line verification equipment. The historical temperature data in the data set is input into the LSTM neural network model. The historical verification error data in the data set is used as the label of the LSTM neural network model to train the LSTM neural network model.

[0006] The actual temperature data of the assembly line calibration equipment and the environment around the assembly line calibration equipment are input into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

[0007] In the above technical solution, the assembly line calibration equipment includes a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

[0008] In the above technical solution, a distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

[0009] The above technical solution also includes preprocessing the acquired historical temperature data and historical calibration error data:

[0010] The data samples are cleaned, completed, smoothed and normalized. The normalization uses the Min-Max model, specifically:

[0011]

[0012] In the formula: x is the input data, y is the normalized output data, a is the minimum value after normalization, which is -1, b is the maximum value after normalization, which is 1, and the target range of a and b after normalization is [-1, 1]. min and x max are the minimum and maximum values ​​of the input data.

[0013] In the above technical solution, the LSTM neural network model is a time series neural network, and the prediction form is (x(kt), x(k-t+1), ..., x(k-1), x(k))→(y(k+1), y(k+2), ..., y(k+p), wherein k represents the current time step, t represents the number of backtracking time steps, that is, the number of past time steps, (x(kt), x(k-t+1), ..., x(k-1), x(k)) represents the input data sequence of t time steps forward from the current time step k, which are used as the input of LSTM to predict future values, p represents the predicted number of future time steps, (y(k+1), y(k+2), ..., y(k+p) represents the output data sequence of p time steps in the future predicted from the current time step k, which are the future values ​​predicted by the LSTM model through the input sequence.

[0014] In the above technical solution, training the LSTM neural network model includes:

[0015] Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model;

[0016] Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value;

[0017] Use the back propagation algorithm to calculate the gradient of the loss value in the loss function with respect to the weight and bias parameters of the LSTM neural network model;

[0018] According to the calculated gradient, the stochastic gradient descent SGD optimization algorithm is used to update the weights and bias parameters of the LSTM neural network model;

[0019] The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

[0020] In the above technical solution, through continuous back propagation and parameter updating, the LSTM neural network model can gradually optimize the model parameters in the regression task, so that the model can better fit the training data and improve the generalization ability of new data.

[0021] In the above technical solution, the mean square error loss function MSE is:

[0022]

[0023] Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0024] In the above technical solution, the stochastic gradient descent SGD optimization algorithm is:

[0025]

[0026] Among them, θ t represents the model parameters at step t; θ t+1 represents the updated model parameters; α represents the learning rate; Represents the gradient of the loss function with respect to the parameters.

[0027] In the above scheme, through continuous iteration, the parameters of the model are gradually optimized, so that the loss function gradually decreases.

[0028] In the above technical scheme, the LSTM neural network model is constructed in the Matlab environment, including Sequencelnputlayer (sequence input layer), LSTMLayer (long short-term memory layer), Dropoutlayer (discard layer), LSTMLayer (long short-term memory layer), DropoutLayer (discard layer), LSTMLayer (long short-term memory layer), DropoutLayer (discard layer), ReLULayer (ReLU activation layer), FullyConnectedLayer (fully connected layer), and RegressionOutputlayer (regression output layer), a total of 10 layers of LSTM neural network model.

[0029] In the above technical solution, the LSTMLayer is composed of LSTM units, and the LSTM units include an input gate, a forget gate and an output gate;

[0030] The input gate is used to decide what information will be added to the cell state at the current time step;

[0031] The forget gate is used to determine which information will be forgotten from the cell state in the current time step;

[0032] The output gate is used to decide which parts of the cell state will be output at the current time step.

[0033] In the above technical solution, the initial learning rates of the three LSTMLayers are set to 0.005, and the numbers of hidden layers of the three LSTMLayers are 128, 64, and 32 respectively. After the setting, the LSTM neural network model is trained.

[0034] In the above technical solution, RMSE and loss evaluation indicators are used to evaluate the accuracy of the predicted calibration error of the assembly line calibration equipment.

[0035] The present invention also discloses a pipeline verification error prediction system based on LSTM neural network, including a data acquisition module, a network training module and a pipeline verification error prediction module.

[0036] A data acquisition module, used for constructing a data set based on historical temperature data of the assembly line verification equipment and the environment surrounding the assembly line verification equipment and historical verification error data of the assembly line verification equipment;

[0037] The network training block is used to input the historical temperature data in the data set into the LSTM neural network model, and use the historical verification error data in the data set as the label of the LSTM neural network model to train the LSTM neural network model;

[0038] The assembly line calibration error prediction module is used to input the actual temperature data of the assembly line calibration equipment and the surrounding environment of the assembly line calibration equipment into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

[0039] In the above technical solution, the assembly line calibration equipment includes a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

[0040] In the above technical solution, a distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

[0041] Training the LSTM neural network model includes:

[0042] Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model;

[0043] Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value;

[0044] Use the back propagation algorithm to calculate the gradient of the loss value in the loss function with respect to the weight and bias parameters of the LSTM neural network model;

[0045] According to the calculated gradient, the stochastic gradient descent SGD optimization algorithm is used to update the weights and bias parameters of the LSTM neural network model;

[0046] The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

[0047] In the above technical solution, the mean square error loss function MSE is:

[0048]

[0049] Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0050] In the above technical solution, the stochastic gradient descent SGD optimization algorithm is:

[0051]

[0052] Among them, θ t represents the model parameters at step t; θ t+1represents the updated model parameters; α represents the learning rate; Represents the gradient of the loss function with respect to the parameters.

[0053] In the above technical scheme, the LSTM neural network model is constructed in the Matlab environment, and the model consists of 10 layers: Sequencelnputlayer, LSTMLayer, Dropoutlayer, LSTMLayer, DropoutLayer, LSTMLayer, DropoutLayer, ReLULayer, FullyConnectedLayer, and RegressionOutputlayer.

[0054] In the above technical solution, the LSTMLayer is composed of LSTM units, and the LSTM units include an input gate, a forget gate and an output gate;

[0055] The input gate is used to decide what information will be added to the cell state at the current time step;

[0056] The forget gate is used to determine which information will be forgotten from the cell state in the current time step;

[0057] The output gate is used to decide which parts of the cell state will be output at the current time step.

[0058] The present invention also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the pipeline verification error prediction method based on the LSTM neural network are implemented.

[0059] The temperature data of the present invention covers the temperature data of each key part of the assembly line calibration equipment, such as the main control board, power module, load device, heat dissipation device, conveyor belt motor and other parts and the ambient temperature of the calibration equipment. By establishing an LSTM neural network model and using temperature field information to predict equipment calibration errors, temperature anomalies can be discovered in time, the calibration equipment can be repaired, the reliability and stability of the equipment can be improved, and the calibration quality of the production line for the product can be ensured. Secondly, the temperature field temperature information is used as the input of the LSTM neural network model to improve the stability of the production process and product quality, providing reliable support and guarantee for industrial production.

[0060] The optical fiber sensing technology used in the present invention can realize high-precision monitoring of the production environment, and has the advantages of real-time and remote monitoring. The optical fiber sensing technology used in the present invention refers to distributed optical fiber sensing technology. Distributed optical fiber sensing technology uses the optical fiber itself as a sensor to sense changes in environmental parameters by measuring changes in optical signals on the optical fiber. The simultaneous measurement of multiple temperature parameter points on a single optical fiber realizes the monitoring of each point on the entire optical fiber length, and monitors the temperature changes of the production environment in real time, thereby providing a more detailed and comprehensive monitoring capability for the production environment.

[0061] The long short-term memory (LSTM) neural network model adopted in the present invention has the advantages of strong memory ability and good adaptability, and is suitable for modeling and predicting time series data.

[0062] The present invention combines distributed optical fiber sensing technology and LSTM neural network model to propose a pipeline calibration error prediction method and system based on LSTM neural network. It can not only realize real-time monitoring and remote monitoring of pipeline equipment, but also analyze and predict monitoring data through neural network model, so as to timely discover potential problems and improve the accuracy and efficiency of calibration. This is of great significance for improving the stability of industrial production process and product quality.

[0063] The present invention combines the application of distributed optical fiber sensing technology and LSTM neural network model in production environment monitoring and equipment calibration error prediction, and can be used for equipment calibration and maintenance in manufacturing and industrial production processes. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a flow chart of the pipeline verification error prediction method based on LSTM neural network;

[0065] Figure 2 It is the LSTM unit structure diagram;

[0066] Figure 3 It is the training flow chart of the long short-term memory (LSTM) neural network model.

[0067] Figure 4 It is a system diagram of the module connection of the present invention. DETAILED DESCRIPTION

[0068] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but they do not constitute a limitation on the present invention.

[0069] Embodiment 1:

[0070] like Figure 1As shown, the present invention provides a pipeline verification error prediction method based on LSTM neural network, comprising the following steps:

[0071] A data set is constructed based on the historical temperature data of the assembly line verification equipment and the environment around the assembly line verification equipment and the historical verification error data of the assembly line verification equipment. The historical temperature data in the data set is input into the LSTM neural network model. The historical verification error data in the data set is used as the label of the LSTM neural network model to train the LSTM neural network model.

[0072] The actual temperature data of the assembly line calibration equipment and the environment around the assembly line calibration equipment are input into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

[0073] The assembly line verification equipment comprises a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

[0074] A distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

[0075] It also includes preprocessing the acquired historical temperature data and historical calibration error data:

[0076] The data samples are cleaned, completed, smoothed and normalized. The normalization uses the Min-Max model, specifically:

[0077]

[0078] In the formula: x is the input data, y is the normalized output data, a is the minimum value after normalization, which is -1, b is the maximum value after normalization, which is 1, and the target range of a and b after normalization is [-1, 1]. min and x max are the minimum and maximum values ​​of the input data.

[0079] The LSTM neural network model is a time series neural network. The prediction form is (x(kt), x(k-t+1), ..., x(k-1), x(k)) → (y(k+1), y(k+2), ..., y(k+p), where k represents the current time step, t represents the number of backtracking time steps, (x(kt), x(k-t+1), ..., x(k-1), x(k)) represents the input data sequence starting from the current time step k and counting t time steps forward, p represents the predicted future time steps, (y(k+1), y(k+2), ..., y(k+p) +p) represents the output data sequence of the next p time steps predicted from the current time step k. In this embodiment, the time window t is set to 7 days, and the prediction length p is set to one day, that is, the historical temperature data values ​​of 7 days are used to predict the calibration error of the next day. Therefore, the sample data are divided and combined, and a continuous time series is extracted from the original data set. Assuming that the original data set has N groups of data {x(1), x(2), ..., x(N)}{y(1), y(2), ..., y(N)}, then Nt input-output pairs can be generated, and the division form is as follows:

[0080] (X1, Y1)=((x(1),x(2),...,x(7)),y(8))

[0081] (X2, Y2)=((x(2),x(3),...,x(8)),y(9))

[0082]

[0083] (X N-t , Y N-t )=((x(Nt),x(N-t+1),...,x(N-1)),y(N)).

[0084] like Figure 3 As shown, training the LSTM neural network model includes:

[0085] Forward propagation: Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model.

[0086] Loss calculation: Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value.

[0087] The mean square error loss function MSE is:

[0088]

[0089] Where N is the number of samples, y iis the true value of the i-th sample, is the predicted value of the ith sample.

[0090] Back propagation: The back propagation algorithm is used to calculate the gradient of the loss value in the loss function to the weight and bias parameters of the LSTM neural network model. Starting from the loss function, the gradient is calculated layer by layer through the chain rule and propagated back to each parameter.

[0091] Parameter update: Based on the calculated gradient, the stochastic gradient descent (SGD) optimization algorithm is used to update the weight and bias parameters of the LSTM neural network model. The update process is performed according to the following formula:

[0092]

[0093] Among them, θ t represents the model parameters at step t; θ t+1 Represents the updated model parameters; α represents the learning rate (i.e., the step size of the optimization algorithm); Represents the gradient of the loss function with respect to the parameters. Through continuous iteration, the parameters of the model are gradually optimized, making the loss function gradually decrease.

[0094] Repeated training: The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

[0095] When training a neural network, the number of iterations refers to the number of times the model performs a complete forward propagation and back propagation on the entire training dataset. Each iteration includes the steps of forward propagation, loss calculation, back propagation, and parameter update. The number of iterations is a very important parameter in the training process, which directly affects the convergence and final performance of the model.

[0096] Through continuous back-propagation and parameter updating, the LSTM neural network model can gradually optimize the model parameters in the regression task, so that the model can better fit the training data and improve the generalization ability of new data.

[0097] The LSTM neural network model is constructed in the Matlab environment, with a total of 10 layers: Sequencelnputlayer, LSTMLayer, Dropoutlayer, LSTMLayer, DropoutLayer, LSTMLayer, DropoutLayer, ReLULayer, FullyConnectedLayer, and RegressionOutputlayer.

[0098] LSTMLayer is composed of LSTM units. Figure 2This is a schematic diagram of the LSTM unit structure. The composition of the LSTM unit is as follows:

[0099] Assume that at time step t, the input is x t , the hidden state of the previous time step is h t-1 , the cell state at the previous time step is C t-1 In LSTM, there are three main gates (input gate, forget gate, output gate) to control the flow of information:

[0100] 1. Input Gate:

[0101] The input gate determines what information will be added to the cell state at the current time step.

[0102] i t =σ(W i ·[h t-1 , x t ]+b i )

[0103] i t : The output vector of the input gate, indicating the state of the input gate at the current time t;

[0104] σ: Sigmoid activation function, which limits the result to [0, 1];

[0105] W i : weight matrix of input gate;

[0106] h t-1 : The hidden state vector of the previous moment;

[0107] x t : The input vector at the current moment;

[0108] [h t-1 , x t ]:h t-1 and x t To splice;

[0109] b i : Bias vector for input gate.

[0110] 2. Forget Gate:

[0111] 3. The forget gate determines which information will be forgotten from the cell state in the current time step.

[0112] f t =σ(W f ·[h t-1 , x t ]+b f )

[0113] f t : The output vector of the forget gate, indicating the state of the input gate at the current time t;

[0114] σ: Sigmoid activation function, which limits the result to [0, 1];

[0115] W f : weight matrix of input gate;

[0116] h t-1 : The hidden state vector of the previous moment;

[0117] x t : The input vector at the current moment;

[0118] [h t-1 , x t ]:h t-1 and x t To splice;

[0119] b f : Bias vector of forget gate.

[0120] 4. Update Cell State:

[0121] Candidate memory cell status:

[0122]

[0123] Candidate memory cell state vector;

[0124] tanh: Hyperbolic tangent activation function, which limits the result to [-1, 1];

[0125] W C : The weight matrix of the candidate memory cell state;

[0126] h t-1 : The hidden state vector of the previous moment;

[0127] x t : The input vector at the current moment;

[0128] [h t-1 , x t ]:h t-1 and x t To splice;

[0129] b C : bias vector of candidate memory cell state;

[0130] The memory cell status is updated as follows:

[0131]

[0132] C t : The state vector of the memory unit at the current time t;

[0133] f t : Output vector of the forget gate;

[0134] C t-1 : The memory unit state vector at the previous moment;

[0135] i t : Output vector of input gate;

[0136] Candidate memory cell state vector;

[0137] ⊙: Successive multiplication of elements (Hadamard product).

[0138] 4. Output Gate:

[0139] The output gate determines which parts of the cell state at the current time step will be output.

[0140] o t =σ(W o ·[h t-1 , x t ]+b o )

[0141] O t : The output vector of the output gate, representing the state of the input gate at the current time t;

[0142] σ: Sigmoid activation function, which limits the result to [0, 1];

[0143] W o : The weight matrix of the output gate;

[0144] h t-1 : The hidden state vector of the previous moment;

[0145] x t : The input vector at the current moment;

[0146] [h t-1 , x t ]:h t-1 and x t To splice;

[0147] b o : Bias vector for the output gate.

[0148] 5. Hide status update:

[0149] Hidden state h t The update is given by:

[0150] h t =o t ⊙tanh(C t )

[0151] h t : The hidden state vector at the current time t;

[0152] O t : Output vector of the output gate;

[0153] ⊙: element-by-element multiplication (Hadamard product);

[0154] tanh(C t ): The value of the memory cell state at the current moment after passing through the tanh activation function.

[0155] Set three LSTMLayers, and the initial learning rate is 0.005. The number of hidden layers of the three LSTMLayers are 128, 64, and 32 respectively. After setting, the LSTM neural network model is trained.

[0156] Embodiment 2:

[0157] like Figure 4 As shown, the present invention provides a pipeline verification error prediction system based on LSTM neural network, including a data acquisition module, a network training module and a pipeline verification error prediction module.

[0158] A data acquisition module, used for constructing a data set based on historical temperature data of the assembly line verification equipment and the environment surrounding the assembly line verification equipment and historical verification error data of the assembly line verification equipment;

[0159] The network training block is used to input the historical temperature data in the data set into the LSTM neural network model, and use the historical verification error data in the data set as the label of the LSTM neural network model to train the LSTM neural network model;

[0160] The assembly line calibration error prediction module is used to input the actual temperature data of the assembly line calibration equipment and the surrounding environment of the assembly line calibration equipment into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

[0161] The assembly line verification equipment comprises a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

[0162] A distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

[0163] It also includes preprocessing the acquired historical temperature data and historical calibration error data:

[0164] The data samples are cleaned, completed, smoothed and normalized. The normalization uses the Min-Max model, specifically:

[0165]

[0166] In the formula: x is the input data, y is the normalized output data, a is the minimum value after normalization, which is -1, b is the maximum value after normalization, which is 1, and the target range of a and b after normalization is [-1, 1]. min and x max are the minimum and maximum values ​​of the input data.

[0167] The LSTM neural network model is a time series neural network, and the prediction form is (x(kt), x(k-t+1), ..., x(k-1), x(k))→(y(k+1), y(k+2), ..., y(k+p), where k represents the current time step, t represents the number of backtracking time steps, (x(kt), x(k-t+1), ..., x(k-1), x(k)) represents the input data sequence of t time steps forward from the current time step k, p represents the predicted number of future time steps, and (y(k+1), y(k+2), ..., y(k+p) represents the output data sequence of p time steps in the future predicted from the current time step k.

[0168] Training the LSTM neural network model includes:

[0169] Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model;

[0170] Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value;

[0171] Use the back propagation algorithm to calculate the gradient of the loss value in the loss function with respect to the weight and bias parameters of the LSTM neural network model;

[0172] According to the calculated gradient, the stochastic gradient descent SGD optimization algorithm is used to update the weights and bias parameters of the LSTM neural network model;

[0173] The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

[0174] The mean square error loss function MSE is:

[0175]

[0176] Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

[0177] The stochastic gradient descent SGD optimization algorithm is:

[0178]

[0179] Among them, θ t represents the model parameters at step t; θ t+1 represents the updated model parameters; α represents the learning rate; Represents the gradient of the loss function with respect to the parameters.

[0180] The LSTM neural network model is constructed in the Matlab environment, with a total of 10 layers of LSTM neural network model, including Sequencelnputlayer, LSTMLayer, Dropoutlayer, LSTMLayer, DropoutLayer, LSTMLayer, DropoutLayer, ReLULayer, FullyConnectedLayer, and RegressionOutputlayer.

[0181] The LSTMLayer is composed of LSTM units, and the LSTM units include an input gate, a forget gate and an output gate;

[0182] The input gate is used to decide what information will be added to the cell state at the current time step;

[0183] The forget gate is used to determine which information will be forgotten from the cell state in the current time step;

[0184] The output gate is used to decide which parts of the cell state will be output at the current time step.

[0185] Embodiment 3:

[0186] The present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the pipeline verification error prediction method based on the LSTM neural network are implemented.

[0187] The present invention can achieve high-precision monitoring of multiple points in the production environment, while the LSTM neural network model can accurately predict time series data. Combining the two can improve the accuracy and efficiency of equipment verification error prediction.

[0188] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0189] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0190] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0191] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0192] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0193] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the invention, but these changes, modifications or equivalent substitutions are all within the protection scope of the pending claims of the invention.

Claims

1. A pipeline verification error prediction method based on LSTM neural network, Features: A data set is constructed based on the historical temperature data of the assembly line verification equipment and the environment around the assembly line verification equipment and the historical verification error data of the assembly line verification equipment. The historical temperature data in the data set is input into the LSTM neural network model. The historical verification error data in the data set is used as the label of the LSTM neural network model to train the LSTM neural network model. The actual temperature data of the assembly line calibration equipment and the environment around the assembly line calibration equipment are input into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

2. According to the LSTM neural network-based pipeline calibration error prediction method of claim 1, It is characterized in that The assembly line verification equipment comprises a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

3. According to the LSTM neural network-based pipeline calibration error prediction method of claim 2, It is characterized in that A distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

4. The pipeline verification error prediction method based on LSTM neural network according to claim 1, It is characterized in that Training the LSTM neural network model includes: Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model; Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value; Use the back propagation algorithm to calculate the gradient of the loss value in the loss function with respect to the weight and bias parameters of the LSTM neural network model; According to the calculated gradient, the stochastic gradient descent SGD optimization algorithm is used to update the weights and bias parameters of the LSTM neural network model; The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

5. According to the LSTM neural network-based pipeline calibration error prediction method of claim 4, Features: The mean square error loss function MSE is: Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

6. The pipeline verification error prediction method based on LSTM neural network according to claim 4, Features: The stochastic gradient descent SGD optimization algorithm is: Among them, θ t represents the model parameters at step t; θ t+1 represents the updated model parameters; α represents the learning rate; Represents the gradient of the loss function with respect to the parameters.

7. The pipeline verification error prediction method based on LSTM neural network according to claim 1, It is characterized in that The LSTM neural network model is constructed in the Matlab environment, with a total of 10 layers of LSTM neural network model, including Sequencelnputlayer, LSTMLayer, Dropoutlayer, LSTMLayer, DropoutLayer, LSTMLayer, DropoutLayer, ReLULayer, FullyConnectedLayer, and RegressionOutputlayer.

8. The pipeline verification error prediction method based on LSTM neural network according to claim 7, Features: The LSTMLayer is composed of LSTM units, and the LSTM units include an input gate, a forget gate and an output gate; The input gate is used to decide what information will be added to the cell state at the current time step; The forget gate is used to determine which information will be forgotten from the cell state in the current time step; The output gate is used to decide which parts of the cell state will be output at the current time step.

9. A pipeline verification error prediction system based on LSTM neural network, Features: It includes data acquisition module, network training module and pipeline verification error prediction module. A data acquisition module, used for constructing a data set based on historical temperature data of the assembly line verification equipment and the environment surrounding the assembly line verification equipment and historical verification error data of the assembly line verification equipment; The network training block is used to input the historical temperature data in the data set into the LSTM neural network model, and use the historical verification error data in the data set as the label of the LSTM neural network model to train the LSTM neural network model; The assembly line calibration error prediction module is used to input the actual temperature data of the assembly line calibration equipment and the surrounding environment of the assembly line calibration equipment into the trained LSTM neural network model to obtain the calibration error of the assembly line calibration equipment.

10. The pipeline verification error prediction system based on LSTM neural network according to claim 9, Features: The assembly line verification equipment comprises a main control board, a power module, a load device, a heat dissipation device and a conveyor belt motor.

11. The pipeline verification error prediction system based on LSTM neural network according to claim 10, Features: A distributed optical fiber sensor is installed at any point on each component of the assembly line calibration equipment or in a space environment within 0.8 to 1.2 m around the assembly line calibration equipment, and temperature data is obtained using the distributed optical fiber sensor.

12. The pipeline verification error prediction system based on LSTM neural network according to claim 9, Features: Training the LSTM neural network model includes: Input sequence data and calculate the predicted value of the LSTM neural network model through each layer of the LSTM neural network model; Use the selected mean square error loss function to calculate the difference between the predicted value of the LSTM neural network model and the true label of the LSTM neural network model to obtain the loss value; Use the back propagation algorithm to calculate the gradient of the loss value in the loss function with respect to the weight and bias parameters of the LSTM neural network model; According to the calculated gradient, the stochastic gradient descent SGD optimization algorithm is used to update the weights and bias parameters of the LSTM neural network model; The above steps are repeated for multiple iterations on the entire training dataset until the calibrated number of iterations is reached and the loss function converges.

13. The pipeline verification error prediction system based on LSTM neural network according to claim 12, Features: The mean square error loss function MSE is: Where N is the number of samples, y i is the true value of the i-th sample, is the predicted value of the ith sample.

14. The pipeline verification error prediction system based on LSTM neural network according to claim 12, Features: The stochastic gradient descent SGD optimization algorithm is: Among them, θ t represents the model parameters at step t; θ t+1 represents the updated model parameters; α represents the learning rate; Represents the gradient of the loss function with respect to the parameters.

15. The pipeline verification error prediction system based on LSTM neural network according to claim 9, Features: The LSTM neural network model is constructed in the Matlab environment, with a total of 10 layers of LSTM neural network model, including Sequencelnputlayer, LSTMLayer, Dropoutlayer, LSTMLayer, DropoutLayer, LSTMLayer, DropoutLayer, ReLULayer, FullyConnectedLayer, and RegressionOutputlayer.

16. The pipeline verification error prediction system based on LSTM neural network according to claim 15, Features: The LSTMLayer is composed of LSTM units, and the LSTM units include an input gate, a forget gate and an output gate; The input gate is used to decide what information will be added to the cell state at the current time step; The forget gate is used to determine which information will be forgotten from the cell state in the current time step; The output gate is used to decide which parts of the cell state will be output at the current time step.

17. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images