Industrial decision intelligent generation method based on neural network model

Abstract

The invention provides an industrial decision intelligent generation method based on a neural network model, and belongs to the technical field of intelligent decision, and the method comprises the steps: constructing a neural network model; the decision difference of each first decision sample based on the neural network model is extracted, and neurons are added to each hidden layer in the neural network model; determining an optimal variable of each first decision sample in the first industrial scene, and performing optimization processing on the corresponding first decision sample in combination with a historical optimal probability to obtain a second decision sample; acquiring a third decision sample in a second industrial scene related to the industrial task, and training the neural network model added with the neurons in combination with the second decision sample to obtain an industrial decision model; and obtaining the industrial anomaly of the new industrial task and inputting the industrial anomaly into the industrial decision-making model, and automatically generating an intelligent decision-making scheme. Intelligent decision-making schemes can be quickly and accurately generated for various production problems, and decision-making errors are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent decision-making, and particularly relates to an intelligent generation method for industrial decision-making based on a neural network model. Background Art

[0002] In the modern industrial production system, efficient and accurate industrial decision-making plays a crucial role in ensuring the stable operation of equipment, optimizing the production process flow, and ensuring the compliance of product quality. Traditional industrial decision-making methods mostly rely on manual experience judgment and simple data statistical analysis. For example, operators set equipment operation parameters based on past production experience and execute the production process according to fixed process standards. This method not only has low decision-making efficiency but is also greatly affected by human factors, easily leading to problems such as delayed equipment fault diagnosis, lagging production processes, and unstable product quality, thus resulting in low production efficiency.

[0003] With the development of artificial intelligence technology, neural networks have gradually been applied to the field of industrial decision-making due to their powerful non-linear fitting and data processing capabilities. By constructing a neural network model and training it with historical data, some decision-making tasks in industrial scenarios can be automatically processed, improving the decision-making efficiency and accuracy to a certain extent. However, the existing industrial decision-making schemes based on neural networks still have many limitations. On the one hand, during the training process of traditional neural network models, they only rely on limited samples, resulting in insufficient accuracy of the models in judging equipment operation, production process, and quality inspection in industrial scenarios, and it is difficult to cope with complex and changing industrial environments. On the other hand, the existing model structures are relatively fixed. When industrial scenarios change or new decision-making requirements emerge, the generalization ability of the models is poor and they cannot be adjusted in time to provide reliable decision-making results.

[0004] Therefore, the present invention proposes an intelligent generation method for industrial decision-making based on a neural network model. Summary of the Invention

[0005] The present invention provides an intelligent generation method for industrial decision-making based on a neural network model to solve the above-mentioned technical problems.

[0006] The present invention proposes an intelligent generation method for industrial decision-making based on a neural network model, including:

[0007] Step 1: Construct a neural network model, wherein the neural network model is trained based on first decision-making samples in a first industrial scenario related to an industrial task and is used for judging equipment operation, production process, and quality inspection in the first industrial scenario;

[0008] Step 2: Extract the decision differences of each first decision sample based on the neural network model, and add neurons to each hidden layer in the neural network model;

[0009] Step 3: Determine the optimal variables of each first decision sample in the first industrial scenario, and perform optimization processing on the corresponding first decision sample in combination with the historical optimal probability to obtain the second decision sample;

[0010] Step 4: Obtain the third decision sample in the second industrial scenario related to the industrial task, and train the neural network model with added neurons in combination with the second decision sample to obtain the industrial decision model;

[0011] Step 5: Obtain the industrial anomalies of the new industrial task and input them into the industrial decision model to automatically generate an intelligent decision-making plan.

[0012] Preferably, constructing a neural network model includes:

[0013] Obtain the first anomaly and the first decision for equipment operation, the second anomaly and the second decision for the production process, and the third anomaly and the third decision for quality inspection from the first decision samples, and construct an input-output vector, where the input-output vector is: the actual input vector composed of the first anomaly, the second anomaly, and the third anomaly, and the actual output vector composed of the first decision, the second decision, and the third decision;

[0014] Train all the input-output vectors in the first industrial scenario to obtain the neural network model.

[0015] Preferably, adding neurons to each hidden layer in the neural network model includes:

[0016] Extract the actual input vector from the first decision samples, and input it into the neural network model to obtain the predicted output vector. Compare the predicted output vector with the actual output vector in the input-output vector to construct the decision difference;

[0017] Determine the number of neurons to be supplemented in the corresponding hidden layer according to each decision difference to obtain the addition array of all hidden layers, and construct an addition matrix to obtain the addition feature vector and the variance of each hidden layer;

[0018] Extract the feature coefficient and variance of each hidden layer in the addition feature vector, and extract the average number of each hidden layer based on the addition matrix. Match the number of neurons to be added in the corresponding hidden layer from the average-coefficient-variance-number comparison table, and supplement the corresponding hidden layer.

[0019] Preferably, determining the number of neurons to be supplemented in each hidden layer according to each decision difference includes:

[0020] Determine the additional neuron positions in the corresponding hidden layer according to the prediction paths of each differential decision in the corresponding hidden layer and the prediction errors of each neuron, where the additional neuron positions are the positions before the first neuron in the prediction path of the corresponding hidden layer or the positions after the last neuron in the prediction path of the corresponding hidden layer;

[0021] Meanwhile, determine the number of neurons to be supplemented in the corresponding hidden layer according to the prediction errors of each neuron;

[0022]

[0023] Among them, represents the number of neurons to be supplemented in the j-th hidden layer; represents the prediction error of the i1-th neuron in the prediction path of the i2-th first decision sample based on the j-th hidden layer; represents the maximum error of the sum of the prediction errors of all neurons in the prediction path of m1 first decision samples based on the j-th hidden layer; n represents the total number of neurons in the prediction path based on the j-th hidden layer; represents the empirical coefficient, with a value of 1.2; m1 represents the number of first decision samples; represents the ceiling symbol.

[0024] Preferably, when the additional position is before the first neuron, it is fully connected to the neurons in the previous layer; when the additional position is after the last neuron, it is probabilistically connected to the neurons in the next layer;

[0025] Among them, the activation probability of the probabilistic connection is dynamically adjusted through the following formula:

[0026]

[0027] Among them, is the activation probability; represents the average activation value of all neurons in the prediction path of the j-th hidden layer; represents the maximum activation value among all neurons in the prediction path of the j-th hidden layer; is the basic offset, with a value of 0.2.

[0028] Preferably, determining the optimal variables of each first decision sample in the first industrial scenario includes:

[0029] Extract the independent variables corresponding to each anomaly in the first decision sample and the decision for each anomaly based on the dependent variables of each independent variable corresponding to the anomaly, and establish an active influence set for each independent variable, where the active influence set contains a number of dependent variables that affect the corresponding independent variable, and the influence coefficient is determined based on the non-linear relationship between the independent variable and the dependent variable;

[0030] Input the active influence set into the set analysis model, set importance labels for the independent variables corresponding to the anomalies, and obtain a first important variable group for equipment operation, a second important variable group for production processes, and a third important variable group for quality inspection. Among them, the first important variable group, the second important variable group, and the third important variable group are used as the optimal variables for the first industrial scenario.

[0031] Preferably, before optimizing the first decision sample in combination with the historical optimal probability, it includes:

[0032] Obtain the actual participation effect of each independent variable in each first decision sample, screen the first variables whose actual participation effect is greater than the preset participation effect. If all the first variables belong to the optimal variables, at this time, it is determined that the first decision sample reaches the optimum;

[0033] Count the first quantity that reaches the optimum in all the first decision samples, and combine the total number of samples of all the first decision samples to obtain the optimal probability Yp;

[0034] Perform random probability screening on the first decision samples that do not reach the optimum according to rand(0, 1 - Yp);

[0035] When the screening result is 0, it is determined that there is no need to optimize the first decision samples that do not reach the optimum;

[0036] Otherwise, it is determined that it is necessary to optimize the first decision samples that do not reach the optimum.

[0037] Preferably, after determining that it is necessary to optimize the first decision samples that do not reach the optimum, it further includes:

[0038] Construct the participation effect difference sets of the first decision samples that do not reach the optimum respectively , where represents the difference effect conversion value between the actual participation effect and the preset participation effect of the h1-th variable; En represents the total number of variables of the first decision samples that do not reach the optimum;

[0039] Calculate the similarity degrees between the first decision samples that do not reach the optimum and each first decision sample that reaches the optimum respectively, and use the clustering algorithm to perform clustering analysis on the first decision samples that reach the optimum according to the similarity degrees to extract the clustering results under the largest cluster head;

[0040] Determine the variables in the participation effect difference set that are less than 0 and regard them as the second variables, and successively perform reverse variable statistics on each of the second variables with the first decision samples that reach the optimum in the clustering result to obtain a third quantity;

[0041] Lock the second variables in the first decision samples corresponding to the maximum quantity among the third quantities, and successively replace the second variables in the corresponding first decision samples that do not reach the optimum to obtain the second decision samples.

[0042] Compared with the prior art, the beneficial effects of the present application are as follows:

[0043] By gradually optimizing the neural network model, the industrial decision-making model can effectively learn the decision-making experience of different production links. From the construction of the initial model to the multi-scenario data fusion training, the model continuously improves its ability to identify and process industrial anomalies, and can quickly and accurately generate intelligent decision-making solutions for various production problems, which helps the intelligent management of industrial decision-making, reduces manual intervention and decision-making errors, and improves production efficiency and product quality.

[0044] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be achieved and obtained by the structures specifically pointed out in the written specification and the drawings.

[0045] The technical solutions of the present invention will be further described in detail below through the drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0047] Figure 1 is a flowchart of an industrial decision intelligent generation method based on a neural network model in an embodiment of the present invention;

[0048] Figure 2 is a structural diagram of a neural network model in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The following describes the preferred embodiments of the present invention with reference to the drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not used to limit the present invention.

[0050] The present invention provides an industrial decision intelligent generation method based on a neural network model, as Figure 1 shown, including:

[0051] Step 1: Construct a neural network model, where the neural network model is trained based on first decision-making samples in a first industrial scenario related to an industrial task and is used for equipment operation determination, production process determination, and quality inspection determination in the first industrial scenario;

[0052] Step 2: Extract the decision-making differences of each first decision-making sample based on the neural network model, and add neurons to each hidden layer in the neural network model;

[0053] Step 3: Determine the optimal variables of each first decision-making sample in the first industrial scenario, and perform optimization processing on the corresponding first decision-making sample in combination with the historical optimal probability to obtain second decision-making samples;

[0054] Step 4: Obtain third decision-making samples in a second industrial scenario related to the industrial task, and train the neural network model with added neurons in combination with the second decision-making samples to obtain an industrial decision-making model;

[0055] Step 5: Obtain the industrial anomalies of a new industrial task and input them into the industrial decision-making model to automatically generate an intelligent decision-making plan.

[0056] In this embodiment, the new industrial task refers to the industrial task being executed at the current time.

[0057] In this embodiment, the first industrial scenario is, for example, the stamping workshop in automobile production, which is responsible for stamping steel plates into various parts required for the vehicle body, such as car doors, hoods, etc.

[0058] The first decision-making samples, for example, involve: when stamping a certain batch of car doors, equipment operation parameters such as the pressure, stamping speed, and die temperature of the stamping machine, production process parameters such as the material and thickness of the steel plate, and quality inspection data such as the dimensional accuracy and surface flatness of the car door after stamping. This facilitates the construction of input-output vectors for training the first decision-making samples.

[0059] Equipment operation determination is to judge whether the equipment operation parameters meet the set standards. For example, whether the pressure of the stamping machine is stable. If the pressure fluctuates too much, it may affect the forming quality of the car door, and the pressure parameters need to be adjusted.

[0060] Production process determination is the production process flow for the execution of the industrial task. For example, determining whether the current die temperature is appropriate. If the temperature is too high or too low, it may cause the steel plate to deform or wear the die, and the temperature control strategy needs to be adjusted.

[0061] Quality inspection determination is to judge whether the quality of the execution result of the industrial task is qualified. For example, based on data such as the dimensional deviation and surface scratches of the car door, determine whether the car door meets the production standards.

[0062] The decision-making difference is the difference between the actual output vector and the predicted output vector of the neural network model. The optimal variables are the variables that have been in a stable state under equipment operation determination, production process determination, and quality inspection determination.

[0063] The second decision-making sample is a sample obtained by adjusting the first decision-making sample in combination with the optimal variables and historical optimal probabilities. For example, the stamping machine pressure and stamping speed in the original decision-making sample are adjusted to the optimal variable values to form a new second decision-making sample.

[0064] The second industrial scenario is different from the first industrial scenario. For example, in the welding workshop of automobile production, it is responsible for welding the stamped body parts into a complete body frame.

[0065] The third decision-making sample, for example, is related to the equipment operation parameters such as the welding current, welding speed, and weld spacing of the welding robot when welding the body frame, the production process parameters such as the type and specification of the welding material, and the quality inspection data such as the strength and tightness of the body frame after welding.

[0066] The beneficial effects of the above technical solution are as follows: By gradually optimizing the neural network model, the industrial decision-making model can effectively learn the decision-making experience of different production links. From the initial model construction to the multi-scenario data fusion training, the model continuously improves its ability to identify and handle industrial anomalies, and can quickly and accurately generate intelligent decision-making solutions for various production problems, which helps the intelligent management of industrial decision-making, reduces manual intervention and decision-making errors, and improves production efficiency and product quality.

[0067] The present invention proposes an intelligent generation method for industrial decision-making based on a neural network model, and constructs a neural network model, including:

[0068] Obtain the first anomaly and the first decision for equipment operation, the second anomaly and the second decision for production process, and the third anomaly and the third decision for quality inspection from the first decision-making sample, and construct an input-output vector, where the input-output vector is: the actual input vector composed of the first anomaly, the second anomaly, and the third anomaly, and the actual output vector composed of the first decision, the second decision, and the third decision;

[0069] Train all the input-output vectors in the first industrial scenario to obtain a neural network model.

[0070] In this embodiment, for example, in the automobile engine production workshop, when a certain numerical control machine tool processes a piston, the vibration frequency of the equipment exceeds the normal range, which is the first anomaly for equipment operation.

[0071] The first decision. For example, when the vibration frequency of a numerically controlled machine tool is detected to be abnormal, the decision made by the technician is to reduce the machine tool speed and check the tool wear condition. This decision is the first decision.

[0072] The second anomaly. For example, in terms of the production process, during the piston machining process, the cutting temperature continuously exceeds the standard temperature range required by the process. This belongs to the second anomaly.

[0073] The second decision. For example, in response to the situation of excessive cutting temperature, the technician decides to increase the coolant flow rate and adjust the cutting parameters. This decision is the second decision.

[0074] The third anomaly. For example, during quality inspection, it is found that the diameter size deviation of the produced piston exceeds the allowable tolerance range. This is the third anomaly.

[0075] The third decision. For example, regarding the problem of unqualified piston size, the decision is to repair the unqualified pistons and readjust the machine tool processing parameters. This is the third decision.

[0076] The actual input vector is a vector composed of the first anomaly (abnormal vibration frequency data of the numerically controlled machine tool), the second anomaly (excessive cutting temperature data), and the third anomaly (piston diameter size deviation data) in a certain order. For example, [vibration frequency value, cutting temperature value, piston diameter deviation value].

[0077] The actual output vector is a vector formed by combining the first decision (decision encoding for reducing speed and checking the tool), the second decision (decision encoding for increasing coolant flow rate and adjusting cutting parameters), and the third decision (decision encoding for repair processing and adjusting machine tool parameters), such as [decision encoding 1, decision encoding 2, decision encoding 3].

[0078] In this embodiment, all input-output vectors are obtained by combining the input-output vectors of each first decision sample, and a model is built using a deep learning framework (such as TensorFlow or PyTorch). All input-output vectors are divided into a training set, a validation set, and a test set. The training set is used to train the model, the validation set is used to adjust the model hyperparameters, and the test set is used to evaluate the model performance. Through the backpropagation algorithm and an optimizer (such as stochastic gradient descent), the actual input vector is input into the model, and according to the difference between the model output and the actual output vector, the parameters of the neural network are continuously adjusted, and training is repeated until the model reaches better performance indicators on the validation set, and finally a neural network model suitable for the first industrial scenario is obtained.

[0079] The beneficial effects of the above technical solution are as follows: By accurately extracting industrial anomalies and corresponding decisions from the first decision sample and constructing an input-output vector pair neural network model for training, the model can learn the mapping relationship between anomalies and decisions in the industrial scenario. The trained neural network model can, in actual production, quickly predict reasonable decision-making schemes based on real-time detected industrial anomalies, assist industrial production decision-making, improve the automation and intelligence levels of the production process, promptly respond to various problems in production, and ensure the efficiency and stability of production.

[0080] The present invention proposes an intelligent industrial decision-making generation method based on a neural network model, and adds neurons to each hidden layer in the neural network model, including:

[0081] Extract the actual input vector from the first decision sample and input it into the neural network model to obtain a predicted output vector. Compare the predicted output vector with the actual output vector in the input-output vector to construct a decision difference.

[0082] Determine the number of neurons to be supplemented for the neurons in the corresponding hidden layer according to each decision difference to obtain an addition array for all hidden layers, and construct an addition matrix to obtain an addition feature vector and the variance of each hidden layer.

[0083] Extract the feature coefficients and variances of each hidden layer in the addition feature vector, and extract the average number of each hidden layer based on the addition matrix. Match the number of added neurons for the neurons in the corresponding hidden layer from the average-coefficient-variance-number comparison table, and supplement the corresponding hidden layer.

[0084] In this embodiment, the actual input vector for the first decision sample is: {First anomaly, Second anomaly, Third anomaly}, the predicted output vector is: {Predicted decision for the first anomaly, Predicted decision for the second anomaly, Predicted decision for the third anomaly}, and the actual output vector is: {Actual decision for the first anomaly, Actual decision for the second anomaly, Actual decision for the third anomaly};

[0085] At this time, the decision difference is: {Difference between the actual decision and the predicted decision for the first anomaly, Difference between the actual decision and the predicted decision for the second anomaly, Difference between the actual decision and the predicted decision for the third anomaly}.

[0086] In this embodiment, the number of neurons that need to be supplemented for each hidden layer can be determined by analyzing the decision difference.

[0087] In this embodiment, such as Figure 2As shown, the neural network model consists of an input layer (the number of nodes is related to the number of features of the industrial task, for example, 20 nodes are set), a hidden layer with 3 hidden layers, and each layer has 30 neurons, and an output layer (outputting decision results).

[0088] In this embodiment, since there are 3 hidden layers, an additional array = {the data to be supplemented in hidden layer 1, the quantity to be supplemented in hidden layer 2, the quantity to be supplemented in hidden layer 3} is added, and the obtained additional matrix = .

[0089] In this embodiment, the variance of each hidden layer is the variance of the column vectors in the corresponding matrix, reflecting the degree of dispersion of the neuron outputs.

[0090] In this embodiment, the additional feature vector is a vector obtained by performing feature extraction on the additional matrix. For example, [b1, b2, b3] represents the feature enhancement coefficients of each hidden layer.

[0091] In this embodiment, the average quantity is the average of the quantities to be supplemented involved in each column vector of the corresponding additional matrix, that is, it serves as the average quantity of neurons added to the corresponding hidden layer.

[0092] In this embodiment, the average - coefficient - variance - quantity comparison table contains different quantities to be supplemented, feature coefficients, variances, and the corresponding added quantities that are combined and matched to ensure the reliability of the neuron settings in the hidden layer. For example, the feature coefficient of hidden layer 1 is: 0.78, the variance is 0.3, and the average quantity is 3. At this time, the added quantity matched according to the comparison table is 4.

[0093] The beneficial effects of the above - mentioned technical solution are: supplementing neurons in the hidden layer according to the difference between the actual output and the preset output of the neural network model can dynamically optimize the model structure and improve the decision - making accuracy.

[0094] The present invention proposes an industrial decision intelligent generation method based on a neural network model, which determines the quantity of neurons to be supplemented in each hidden layer according to each decision difference, including:

[0095] Determine the adding position of neurons in the corresponding hidden layer according to the prediction path of each differential decision in the corresponding hidden layer and the prediction error of each neuron, where the adding position of the neuron is the position before the first neuron in the prediction path of the corresponding hidden layer or the position after the last neuron in the prediction path of the corresponding hidden layer;

[0096] At the same time, determine the quantity of neurons to be supplemented in the corresponding hidden layer according to the prediction error of each neuron;

[0097]

[0098] Among them, represents the quantity to be supplemented for the j-th hidden layer; represents the prediction error of the i1-th neuron in the prediction path of the i2-th first decision sample based on the j-th hidden layer; represents the maximum error among the sum of prediction errors of all neurons in the prediction path of m1 first decision samples based on the j-th hidden layer; n represents the total number of neurons in the prediction path based on the j-th hidden layer; represents the empirical coefficient, with a value of 1.2; m1 represents the number of first decision samples; represents the ceiling symbol.

[0099] In this embodiment, the value of j is 3.

[0100] In this embodiment, the number of hidden layer neurons reflects the current structure scale, and the logarithm of the training set sample number reflects the demand of the data scale for the model complexity. The combination of the three can scientifically calculate the appropriate supplementary quantity under different data scales and model structures, avoiding overfitting or underfitting caused by blindly adding neurons. The maximum error value represents the situation where the deviation between the model prediction and the actual situation is the largest. As the denominator, it makes the calculated dynamic redundancy coefficient more sensitive to such extreme deviation situations. In industrial decision-making scenarios, such as equipment fault diagnosis, once a misjudgment occurs, it may lead to serious consequences. By participating in the calculation with the maximum error value, the model can pay more attention to such extreme errors, adjust the number of neurons in a timely manner, enhance the learning ability for complex or special situations, and improve the robustness of the model.

[0101] In this embodiment, information transmission is along a specific neuron connection as the prediction path. Assuming that in the j-th hidden layer, information is transmitted and participates in the calculation from neuron A → neuron B → neuron C → neuron D. This transmission path is the prediction path of the corresponding hidden layer. Each neuron undertakes different feature operation tasks in the path. For example, neuron A is responsible for initially integrating the correlation features of welding current and time, and neuron B is responsible for strengthening the influence feature of this correlation on welding strength, etc.

[0102] In this embodiment, each neuron in the prediction path will generate a prediction error. For example, for neuron B, after it operates based on the input information, it expects the output to enable the subsequent neurons to more accurately predict the welding quality, but there is a deviation between the actual output and the ideal output. Assuming that the contribution of the ideal output value of neuron B to the welding strength feature is 0.8 (a value after standardization, etc.), and the contribution of the actual output is 0.6, the difference of 0.2 between the two is the prediction error of this neuron (the i1-th neuron, where i1 corresponds to the serial number of neuron B in the path). .

[0103] In this embodiment, taking the prediction path neurons A → neuron B → neuron C → neuron D in the j-th hidden layer as an example, either a new neuron is added in front of neuron A to preprocess information such as welding process parameters input to this hidden layer earlier; or a new neuron is added behind neuron D to further optimize the information processed by this hidden layer and then transmit it. This newly added position is the added position of the neuron.

[0104] In this embodiment, the cumulative error contribution A1 of all prediction errors under the first neuron is determined, and the cumulative error contribution A2 of all prediction errors of the last neuron is determined. If A1 is greater than A2, it indicates that the "front-end influence" of the first neuron on the error is more significant; otherwise, the "back-end influence" of the last neuron is more prominent. At this time, a new neuron is added before the first neuron; otherwise, a new neuron is added after the last neuron.

[0105] The beneficial effects of the above technical solution are as follows: By accurately analyzing the error situation of differential decision-making in the prediction path of the hidden layer, scientifically calculating and determining the added positions and quantities of hidden layer neurons, the neural network structure can be effectively optimized, enabling the model to better learn the complex relationships between industrial decisions, improving the prediction accuracy, and reducing the deviation between actual production and model prediction.

[0106] The present invention proposes an intelligent generation method for industrial decisions based on a neural network model. When the added position is before the first neuron, it is fully connected to the neurons in the previous layer; when the added position is after the last neuron, it is probabilistically connected to the neurons in the next layer.

[0107] Among them, the activation probability of the probabilistic connection is dynamically adjusted through the following formula:

[0108]

[0109] Among them, is the activation probability; represents the average activation value of all neurons in the prediction path of the j-th hidden layer; represents the maximum activation value among all neurons in the prediction path of the j-th hidden layer; is the basic offset, with a value of 0.2.

[0110] In this embodiment, since the value range of The range becomes from 0 to 0.8, and then adding 0.2 makes the value range of the connection probability P between 0.2 and 1, avoiding the situation that the connection probability is too low (close to 0) resulting in almost no connection between the new neuron and the subsequent layer, and also preventing the probability from being too high (close to 1) causing over-connection, ensuring an appropriate connection relationship between the new neuron and the neurons in the subsequent layer. 0.2, as a basic offset, ensures that even when the average activation value is 0, there is still a connection probability of 0.2 between the new neuron and the neurons in the subsequent layer, so as not to completely disconnect, maintaining the basic connectivity of the network structure.

[0111] The beneficial effects of the above technical solution are as follows: The differential design of full connection and probability connection enables the neural network to not only grasp the basic process features but also flexibly respond to complex decisions, improving the decision-making accuracy of process optimization and quality inspection. The dynamic probability adjustment ensures that the network can still stably output decisions during production line fluctuations and process switching, reducing production line stops and quality accidents caused by model failures.

[0112] The present invention proposes an intelligent industrial decision-making generation method based on a neural network model, which determines the optimal variables of each first decision sample in the first industrial scenario, including:

[0113] Extract each abnormal independent variable and each abnormal decision corresponding to the first decision sample, and the dependent variables corresponding to each independent variable under the corresponding abnormality, and establish an active influence set for each independent variable, where the active influence set contains several dependent variables that affect the corresponding independent variable, and the influence coefficient is determined based on the non-linear relationship between the independent variable and the dependent variable;

[0114] Input the active influence set into the set analysis model to set importance labels for the independent variables under the corresponding abnormality, and obtain a first important variable group for equipment operation, a second important variable group for production process, and a third important variable group for quality inspection, where the first important variable group, the second important variable group, and the third important variable group are used as the optimal variables of the first industrial scenario.

[0115] In this embodiment, variables such as, in an automobile final assembly workshop, parameters that can be directly measured and controlled, such as torque value, tightening speed, and temperature during the bolt tightening process.

[0116] The dependent variable is the result directly or indirectly caused by the independent variable, such as quality indicators such as bolt pre-tightening force, fatigue life, and loosening probability.

[0117] Taking the torque value as an example, the active influence set may contain {pre-tightening force, fatigue life}, indicating that these two dependent variables are directly affected by the torque value.

[0118] The influence coefficient is an index that measures the strength of the non-linear relationship between the independent variable and the dependent variable. Non-linear mapping relationships between the independent variable and the dependent variable are established using algorithms such as random forests and gradient boosting trees. For example, for the torque value T and the pre-tightening force F, through fitting, we get: , when T = 50 N·m, the influence coefficient is approximately 0.78, indicating that for every 1% change in the torque value, the pre-tightening force changes by approximately 0.78%.

[0119] In this embodiment, the set analysis model is a feature evaluation model based on Shapley values or Gini importance, and is used to calculate the contribution degree of each independent variable to the decision result.

[0120] The importance label is an importance score assigned to each independent variable. For example, the importance label of the torque value is 0.85, indicating its degree of influence on the decision result.

[0121] The first important variable group is the key variables for equipment operation, such as the motor current and rotational speed fluctuation of the tightening gun. The second important variable group is the key variables for production processes, such as the hardness of the bolt material and the lubrication state. The third important variable group is the key variables for quality inspection, such as the pre-tightening force fluctuation range and the torque decay rate.

[0122] Variable grouping divides variables into three groups: equipment operation, production process, and quality inspection according to their business attributes. For example: Equipment operation group: {motor current, rotational speed fluctuation, battery voltage}; Production process group: {bolt material hardness, lubrication state, surface roughness}; Quality inspection group: {pre-tightening force fluctuation range, torque decay rate, axial force deviation}.

[0123] Threshold screening sets an importance threshold (such as 0.5), and retains the variables that exceed the threshold as the optimal variables. For example, the finally determined optimal variable groups are: First important variable group: {motor current, rotational speed fluctuation}; Second important variable group: {bolt material hardness, lubrication state}; Third important variable group: {pre-tightening force fluctuation range, torque decay rate}.

[0124] The beneficial effects of the above technical solution are: In industrial processes, the relationships between variables are usually non-linear, which can accurately describe the influence differences of torque on pre-tightening force in different intervals, avoiding the limitations of linear models, focusing on key variables, optimizing sensor deployment and data acquisition frequency, clarifying the importance levels of each variable, and providing a quantitative basis for process improvement.

[0125] The present invention proposes an intelligent industrial decision-making generation method based on a neural network model. Before optimizing the first decision sample in combination with the historical optimal probability, it includes:

[0126] Obtain the actual participation effect of each independent variable in each first decision sample, and screen the first variables whose actual participation effect is greater than the preset participation effect. If all the first variables belong to the optimal variables, at this time, it is determined that the first decision sample reaches the optimal;

[0127] Count the first quantity that reaches the optimal among all the first decision samples, and combine it with the total sample quantity of all the first decision samples to obtain the optimal probability Yp;

[0128] Perform random probability screening on the first decision samples that do not reach the optimal according to rand(0, 1 - Yp);

[0129] When the screening result is 0, it is determined that there is no need to optimize the first decision samples that do not reach the optimal;

[0130] Otherwise, it is determined that it is necessary to optimize the first decision samples that do not reach the optimal.

[0131] In this embodiment, the actual participation effect is the actual effect measurement value of the corresponding independent variable on the quality, and the preset participation effect is the threshold set in advance for the corresponding independent variable according to historical experience and process standards. For example, the actual effect measurement value of the independent variable k1 is 0.7, and the corresponding threshold is 0.6. At this time, since 0.7 is greater than 0.6, the corresponding independent variable k1 is regarded as the first variable.

[0132] For example, the optimal variables are: {k1, k2, k3, k8, k9}. At this time, the first variables for the first decision sample are k1 and k2. At this time, since the first variables k1 and k2 belong to the optimal variables, the corresponding first decision sample is regarded as reaching the optimal. If the first variables for the first decision sample are k1 and k4, and there is a k4 that does not belong to the optimal variables, it is regarded as a first decision sample that does not reach the optimal.

[0133] In this embodiment, the optimal probability Yp = the first quantity / the total sample quantity. Based on this probability, the overall high-quality level of current industrial production can be quickly understood, providing a basic basis for subsequent decisions, which conforms to the basic idea of frequency estimating probability in statistics.

[0134] In this embodiment, when determining whether to optimize samples that do not reach the optimal state, a random probability screening mechanism is introduced instead of optimizing all non-compliant samples. There are mainly two considerations. On the one hand, over-optimization is avoided because in industrial production, some non-compliant samples may be caused by accidental factors (such as a short-term fluctuation of a certain piece of equipment or a slight difference in raw materials), rather than fundamental problems in the process or variable settings. Blind optimization may increase production costs and production complexity. On the other hand, by setting a random range (0, 1−Yp), the optimization probability of non-compliant samples is associated with the optimal probability. When the optimal probability is higher (the production process is more stable and of better quality), the range of probabilities for non-compliant samples to be optimized is smaller, which meets the requirement of "giving priority to ensuring stable production and moderately optimizing abnormal samples" in actual production and is an effective means to balance production stability and optimization needs.

[0135] The beneficial effects of the above technical solution are as follows: By comparing the first variable and the optimal variable, samples that meet the optimal process requirements and have high product quality can be accurately screened out. The calculation of the optimal probability Yp reflects the stability and high-quality output ability of the current production system from a macro perspective. The random probability screening mechanism not only avoids the "one-size-fits-all" optimization of non-compliant samples but also can targetedly handle samples that may indeed have room for optimization. Through scientific sample analysis, probability calculation, and screening mechanism, a set of intelligent and efficient decision-making and optimization methods are provided for the industrial production process.

[0136] The present invention proposes an intelligent industrial decision-making generation method based on a neural network model. After determining that the first decision sample that does not reach the optimal needs to be optimized, it further includes:

[0137] Construct the difference set of the participation effects of the first decision samples that do not reach the optimal respectively , where represents the difference effect conversion value between the actual participation effect and the preset participation effect of the h1-th variable; En represents the total number of variables of the first decision samples that do not reach the optimal;

[0138] Calculate the similarity degree between the first decision samples that do not reach the optimal and each first decision sample that reaches the optimal respectively, and use the clustering algorithm to perform clustering analysis on the first decision samples that reach the optimal according to the similarity degree to extract the clustering result under the largest cluster head;

[0139] Determine the variables in the participation effect difference set that are less than 0 and regard them as the second variables, and perform reverse variable statistics on each first decision sample that reaches the optimal in the clustering result in sequence with the second variables to obtain the third quantity;

[0140] Lock the second variable in the first decision sample corresponding to the largest quantity among the third quantity to reach the optimal, and sequentially replace the second variable in the corresponding first decision sample that has not reached the optimal to obtain the second decision sample.

[0141] In this embodiment, the differential effect conversion value is used to standardize and normalize the corresponding difference, eliminate the dimension problem, limit the result between -1 and 1, and regard the variable whose existence value after standardization and normalization in the differential effect set is less than 0 as the second variable.

[0142] In this embodiment, the similarity degree is calculated by using the Euclidean distance calculation formula to calculate the similarity degree between the corresponding first decision sample that has not reached the optimal and each first decision sample that has reached the optimal, that is: , where N0 represents the total number of variables included in the corresponding decision sample; represents the value of the j0th variable in the corresponding first decision sample that has not reached the optimal; represents the value of the j0th variable in the corresponding first decision sample that has reached the optimal. The Euclidean distance can effectively measure the difference of the parameter vector and meets the requirement of "finding similar high-quality cases" in process optimization.

[0143] In this embodiment, the clustering algorithm is implemented by the K-means algorithm, and the group with the largest number of samples is extracted as the largest cluster head for subsequent basis to avoid the contingency of individual samples.

[0144] For example, if the value of the second variable p1 is less than 0, and the value of the variable p1 in the first decision sample L1 that has reached the optimal in the corresponding clustering result is greater than 0, at this time, the variable p1 is statistically incremented by one, otherwise, it is not statistically incremented by one, and then the third quantity can be obtained.

[0145] In this embodiment, if the largest quantity is 1, at this time, the second variable in the corresponding first decision sample that has reached the optimal sequentially replaces the second variable in the corresponding first decision sample that has not reached the optimal;

[0146] If the largest quantity is multiple, at this time, randomly select a second variable in the corresponding first decision sample that has reached the optimal from the samples under the largest quantity and sequentially replace the second variable in the corresponding first decision sample that has not reached the optimal. For example, replace the second variable p2 in the first decision sample that has not reached the optimal with the second variable p1 in the first decision sample that has reached the optimal corresponding to the largest quantity obtained.

[0147] The beneficial effects of the above technical solutions are: through the closed-loop logic of deviation quantification, similarity clustering, direction screening, and variable replacement, industrial experience is transformed into computable and reusable optimization strategies.

[0148] Obviously, those skilled in the art can make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention also intends to include these modifications and variations therein.

Claims

1. An intelligent generation method for industrial decision-making based on a neural network model, characterized in that, Including: Step 1: Construct a neural network model, where the neural network model is trained based on first decision-making samples in a first industrial scenario related to an industrial task and is used for equipment operation determination, production process determination, and quality inspection determination in the first industrial scenario; Step 2: Extract the decision-making differences of each first decision-making sample based on the neural network model, and add neurons to each hidden layer in the neural network model; Step 3: Determine the optimal variables of each first decision-making sample in the first industrial scenario, and perform optimization processing on the corresponding first decision-making sample in combination with the historical optimal probability to obtain second decision-making samples; Step 4: Obtain third decision-making samples in a second industrial scenario related to the industrial task and train the neural network model with added neurons in combination with the second decision-making samples to obtain an industrial decision-making model; Step 5: Obtain industrial anomalies of a new industrial task and input them into the industrial decision-making model to automatically generate an intelligent decision-making plan.

2. The intelligent generation method for industrial decision-making based on a neural network model according to claim 1, wherein Constructing a neural network model includes: Obtain the first anomaly and the first decision for equipment operation, the second anomaly and the second decision for the production process, and the third anomaly and the third decision for quality inspection from the first decision-making samples, and construct an input-output vector, where the input-output vector is: an actual input vector composed of the first anomaly, the second anomaly, and the third anomaly, and an actual output vector composed of the first decision, the second decision, and the third decision; Train all input-output vectors in the first industrial scenario to obtain a neural network model.

3. The industrial decision intelligent generation method based on a neural network model according to claim 2, wherein Adding neurons to each hidden layer in the neural network model includes: Extract the actual input vector from the first decision-making samples and input it into the neural network model to obtain a predicted output vector, compare the predicted output vector with the actual output vector in the input-output vector, and construct a decision-making difference; Determine the number of neurons to be supplemented in the corresponding hidden layer according to each decision-making difference to obtain an addition array for all hidden layers, and construct an addition matrix to obtain an addition feature vector and the variance of each hidden layer; Extract the feature coefficients and variances of each hidden layer in the addition feature vector, and extract the average number of each hidden layer based on the addition matrix. Match the number of neurons added to the corresponding hidden layer from the average-coefficient-variance-number comparison table, and supplement the corresponding hidden layer.

4. The intelligent generation method for industrial decision-making based on a neural network model according to claim 3, wherein Determining the number of neurons to be supplemented in each hidden layer according to each decision-making difference includes: Determine the neuron addition position in the corresponding hidden layer according to the prediction path of each difference decision in the corresponding hidden layer and the prediction error of each neuron, where the neuron addition position is the position before the first neuron in the prediction path of the corresponding hidden layer or the position after the last neuron in the prediction path of the corresponding hidden layer; At the same time, determine the number of neurons to be supplemented in the corresponding hidden layer according to the prediction error of each neuron; ； Among them, represents the quantity to be supplemented in the j-th hidden layer; represents the prediction error of the i1-th neuron in the prediction path of the i2-th first decision sample based on the j-th hidden layer; represents the maximum error of the sum of the prediction errors of all neurons in the prediction path of m1 first decision samples based on the j-th hidden layer; n represents the total number of neurons in the prediction path based on the j-th hidden layer; represents the empirical coefficient, with a value of 1.2; m1 represents the number of first decision samples; represents the ceiling symbol.

5. The intelligent generation method for industrial decision-making based on a neural network model according to claim 4, wherein When the addition position is before the first neuron, perform full connection with the previous layer neurons; when the addition position is after the last neuron, perform probabilistic connection with the next layer neurons; Among them, the activation probability of the probabilistic connection is dynamically adjusted through the following formula: ； Among them, is the activation probability; represents the average activation value of all neurons in the prediction path of the j-th hidden layer; represents the maximum activation value among all neurons in the prediction path of the j-th hidden layer; is the base offset, with a value of 0.

2.

6. The intelligent generation method for industrial decision-making based on a neural network model according to claim 2, wherein Determine the optimal variables of each first decision sample in the first industrial scenario, including: Extract the independent variables of each anomaly corresponding to the first decision sample and the dependent variables of each decision corresponding to each anomaly based on the independent variables of each anomaly, and establish an active influence set for each independent variable. Among them, the active influence set contains several dependent variables that affect the corresponding independent variable, and the influence coefficient is determined based on the non-linear relationship between the independent variable and the dependent variable; Input the active influence set into the set analysis model, set importance labels for the independent variables corresponding to the anomalies, and obtain a first important variable group for equipment operation, a second important variable group for production processes, and a third important variable group for quality inspection. Among them, the first important variable group, the second important variable group, and the third important variable group are used as the optimal variables of the first industrial scenario.

7. The intelligent generation method for industrial decision-making based on a neural network model according to claim 1, wherein Before optimizing the first decision sample in combination with the historical optimal probability, it includes: Obtain the actual participation effect of each independent variable in each first decision sample, screen the first variables whose actual participation effect is greater than the preset participation effect. If all the first variables belong to the optimal variables, at this time, it is determined that the first decision sample reaches the optimum; Count the first quantity of the first decision samples that reach the optimum among all the first decision samples, and combine the total number of samples of all the first decision samples to obtain the optimal probability Yp; Perform random probability screening on the first decision samples that do not reach the optimum according to rand(0, 1 - Yp); When the screening result is 0, it is determined that there is no need to optimize the first decision samples that do not reach the optimum; Otherwise, it is determined that the first decision samples that do not reach the optimum need to be optimized.

8. The intelligent generation method for industrial decision-making based on a neural network model according to claim 7, characterized in that After determining that the first decision samples that do not reach the optimum need to be optimized, it also includes: Construct a difference set of the participation effects of the first decision samples that do not reach the optimum respectively , where represents the difference effect conversion value between the actual participation effect and the preset participation effect of the h1-th variable; En represents the total number of variables of the first decision samples that do not reach the optimum Calculate the similarity between the corresponding first decision samples that do not reach the optimum and each first decision sample that reaches the optimum respectively, and use the clustering algorithm to perform clustering analysis on the first decision samples that reach the optimum according to the similarity to extract the clustering results under the largest cluster head; Determine the variables in the participation effect difference set that are less than 0 and regard them as second variables, and perform reverse variable statistics on the second variables and each first decision sample that reaches the optimum in the clustering results in turn to obtain the third quantity; Lock the second variables in the first decision samples that reach the optimum corresponding to the largest quantity among the third quantities, and replace the second variables in the corresponding first decision samples that do not reach the optimum in turn to obtain the second decision samples.

Images