Industrial data hybrid model construction method and system for discrete manufacturing

Abstract

An industrial data hybrid model construction method and system for discrete manufacturing. The method comprises: firstly, collecting discrete manufacturing industrial data in real time, and performing feature extraction to obtain a feature vector; then, separately inputting feature data into a first model and a second model to obtain a first output and a second output; and performing weighted summation on the first output and the second output to obtain an industrial data hybrid model. The first model is used for obtaining a fitting curve of the feature data in a discrete space, and the second model is used for using a decision tree as a weak learner to perform ensemble learning on the feature data. Regarding the characteristic of service scenarios in the discrete manufacturing industry being complex and variable, the flexibility of intelligent data inference is improved.

Description

A method and system for building an industrial data hybrid model for discrete manufacturing Technical Field

[0001] The present invention relates to the field of intelligent reasoning, and in particular to a method and system for constructing an industrial data hybrid model for discrete manufacturing. Background Art

[0002] As the discrete manufacturing industry enters the era of digital transformation, data-driven intelligent reasoning is key to improving production efficiency and quality while reducing energy consumption. Within discrete manufacturing, data can come from multiple sources, including sensors, equipment, and production lines. Furthermore, many decisions within the manufacturing environment require real-time responses, not just analysis of historical data. Conditions within the industrial environment are also constantly changing, requiring reasoning systems to continuously optimize performance in this ever-changing environment to adapt to new processes and requirements. Manufacturing often involves complex production processes involving multiple links and stages. Therefore, discrete manufacturing industrial data is multi-source, heterogeneous, massive, high-dimensional, multi-scale, multi-temporal, and requires high real-time performance. Building accurate models to reason about these complex processes is extremely challenging.

[0003] Traditional methods for intelligent data reasoning are generally static and simplistic, making them inflexible in adapting to dynamically changing environments and problems. For applications requiring high real-time performance, traditional methods can be inflexible. While they perform well for a single type of industrial data, they require constant manual updating of rules and logic, or selection of more appropriate models, when the data comes from multiple sources or multiple time, space, and scales. This makes data reasoning based on traditional models inflexible. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a hybrid model construction method and system based on reinforcement learning for intelligent reasoning of discrete manufacturing industrial data, so as to solve the multi-source, heterogeneous and multi-scale problems of discrete manufacturing industrial data.

[0005] Technical solution: The method for constructing an industrial data hybrid model for discrete manufacturing according to the present invention comprises the following steps:

[0006] Collect discrete manufacturing industrial data in real time and perform feature extraction to obtain feature vectors;

[0007] Passing the characteristic data through a first model and a second model respectively to obtain a first output and a second output;

[0008] Taking a weighted sum of the first output and the second output to obtain a hybrid model of industrial data;

[0009] The first model is used to obtain a fitting curve of the feature data in a discrete space, and the second model is used to perform ensemble learning on the feature data using a decision tree as a weak learner.

[0010] Furthermore, the first model calculates the output for each input feature vector Where l represents industrial loss, w t is the weight vector of the input vector, is a high-dimensional feature space, x i is the input feature vector, i = 1, 2, ... n;

[0011] By minimizing l, we can get l and w t , the constraints are met during the solution process:

[0012] Among them, B is the penalty term, y is the actual target value of the sample, w is the weight vector, ε is the threshold of tolerable error, and e i is the slack variable.

[0013] Furthermore, the second model calculates the output for each input feature vector f K (x i )∈R, where K is the number of weak learners, R is the original feature space of discrete manufacturing industry data, x i is the input feature vector, i = 1, 2, ... n; f K is a function in R, representing the model of the K-th tree trained;

[0014] The objective function of the second model is By iteratively adding tree models to minimize the objective function; where loss is the discrete manufacturing industry dataset loss, M and is the number of decision tree child nodes and their weights, and C is the penalty coefficient.

[0015] Furthermore, the hybrid model is Y=w1y1+w2y2, wherein y1 is the first output, w1 is the weight of the first output, y2 is the second output, and w2 is the weight of the second output.

[0016] Furthermore, the weight w1 of the first output and the weight w2 of the second output are calculated using a least squares weight fusion method.

[0017] Furthermore, the R2 index is used to evaluate the parameters of the hybrid model.

[0018] The discrete manufacturing-oriented industrial data hybrid model construction system of the present invention includes:

[0019] Feature extraction unit, used to collect discrete manufacturing industrial data in real time and perform feature extraction to obtain feature vectors;

[0020] a data processing unit, configured to obtain a first output and a second output by passing the feature data through a first model and a second model, respectively; the first model is configured to obtain a fitting curve for the feature data in a discrete space, and the second model is configured to perform ensemble learning on the feature data using a decision tree as a weak learner;

[0021] The hybrid model establishing unit is configured to perform a weighted summation of the first output and the second output to obtain a hybrid model of the industrial data.

[0022] Furthermore, the first model calculates the output for each input feature vector Where l represents industrial loss, w t is the weight vector of the input vector, is a high-dimensional feature space, x i is the input feature vector, i = 1, 2, ... n;

[0023] By minimizing l, we can get l and w t , the constraints are met during the solution process:

[0024] Among them, B is the penalty term, y is the actual target value of the sample, w is the weight vector, ε is the threshold of tolerable error, and e i is the slack variable;

[0025] The second model calculates the output for each input feature vector f K (x i )∈R, where K is the number of weak learners, R is the original feature space of discrete manufacturing industry data, x i is the input feature vector, i = 1, 2, ... n; f K is a function in R, representing the model of the K-th tree trained;

[0026] The objective function of the second model is By iteratively adding tree models to minimize the objective function; where loss is the discrete manufacturing industry dataset loss, M and is the number of decision tree child nodes and their weights, and C is the penalty coefficient.

[0027] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. It is characterized in that when the computer program is loaded into the processor, it implements the method for constructing an industrial data hybrid model for discrete manufacturing.

[0028] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the method for constructing an industrial data hybrid model for discrete manufacturing is implemented.

[0029] Beneficial effect: Compared with the existing technology, the advantages of the present invention are: the present invention proposes to pass discrete manufacturing industrial data through two models respectively, apply the least squares weight fusion method, and output a hybrid model of intelligent reasoning of discrete manufacturing industrial data. It intersects with traditional data intelligent reasoning methods, is more adaptable to dynamically changing environments and problems, and better meets the complex and changeable business scenario requirements of the discrete manufacturing industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] FIG1 is a flow chart of a hybrid model construction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0032] As shown in FIG1 , the method for constructing an industrial data hybrid model for discrete manufacturing includes the following steps.

[0033] S1 collects industrial data from discrete manufacturing in real time (including real-time equipment status data, energy consumption data, lifecycle data, sensor data, production process data, and quality data). The collected data is mapped from the original feature space to a high-dimensional feature space for easy fitting.

[0034] S2, the discrete manufacturing industrial data are respectively processed by the discrete data vector regression (DSVR) algorithm and the discrete data gradient boosting (DXGB) algorithm.

[0035] (1) DSVR algorithm:

[0036] By finding a curve in the discrete space and making all data points as close as possible to the curve, the loss is minimized for subsequent data reasoning.

[0037] Where l represents industrial loss, w t is the weight vector of the input vector, is a high-dimensional feature space, x i is the input vector.

[0038] The purpose of DSVR is to optimize the model by minimizing the loss l and solving w t and l, but in the solution process, certain constraints must be observed ε cannot be set too large or too small, so the variable e is set for each sample of DSVR i, to avoid loss of rationality:

[0039] Among them, B is the penalty coefficient, which is generally set to 1, y is the actual target value of the sample, w is the weight vector, ε is the threshold of the tolerable error, and e i is the slack variable.

[0040] (2) DXGB algorithm:

[0041] Taking the decision tree as a weak learner, the ensemble mode of the learner is: f K (x i )∈R,x i is the input vector of the i-th discrete manufacturing industry data, K is the number of weak learners, R is the original feature space of discrete manufacturing industry data, f K is a function in R, representing the model of the Kth tree trained. The DXGB objective function is By iteratively adding tree models to minimize the objective function. loss is the loss of discrete manufacturing industry dataset, M and is the number of child nodes and their weights, and C is the penalty coefficient.

[0042] S3, after the discrete manufacturing industry data passes through the two models, the least squares weight fusion method is applied to obtain the combined weight, and the output is the hybrid model: Y = w1y1 + w2y2, where w1 is the weight of DSVR in the hybrid model, and w2 is the weight of DXGB in the hybrid model. The loss function is Satisfying the constraint w1 + w2 = 1, the optimal solution for the comprehensive weight is performed to minimize the loss function H, i.e. minH, and w1 and w2 are obtained at the same time. Where w1 is the weight of DSVR in the hybrid model, and w2 is the weight of DXGB in the hybrid model.

[0043] S4, the R2 indicator is used to evaluate the model parameters.

[0044] Among them, λ 2 is the variance, and MSE is the mean square error. The closer the R2 score is to 1, the better the hybrid model is.

[0045] The discrete manufacturing-oriented industrial data hybrid model construction system of the present invention includes:

[0046] Feature extraction unit, used to collect discrete manufacturing industrial data in real time and perform feature extraction to obtain feature vectors;

[0047] a data processing unit, configured to obtain a first output and a second output by passing the feature data through a first model and a second model, respectively; the first model is configured to obtain a fitting curve for the feature data in a discrete space, and the second model is configured to perform ensemble learning on the feature data using a decision tree as a weak learner;

[0048] The hybrid model establishing unit is configured to perform a weighted summation of the first output and the second output to obtain a hybrid model of the industrial data.

[0049] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. It is characterized in that when the computer program is loaded into the processor, it implements the method for constructing an industrial data hybrid model for discrete manufacturing.

[0050] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the method for constructing an industrial data hybrid model for discrete manufacturing is implemented.

[0051] The computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store desired program code in the form of instructions or data structures and that can be accessed by a computer.

[0052] The processor is configured to execute the computer program stored in the memory to implement the various steps in the method involved in the above embodiment.

Claims

1. A method for constructing an industrial data hybrid model for discrete manufacturing, characterized in that: The steps include: Collect discrete manufacturing industrial data in real time and perform feature extraction to obtain feature vectors; Passing the feature data through the first model and the second model respectively to obtain a first output and a second output; Taking a weighted sum of the first output and the second output to obtain a hybrid model of industrial data; The first model is used to obtain a fitting curve of the feature data in a discrete space, and the second model is used to perform ensemble learning on the feature data using a decision tree as a weak learner; The first model calculates the output for each input feature vector Where l represents industrial loss, w t is the weight vector of the input vector, is a high-dimensional feature space, x i is the input feature vector, i = 1, 2, ... n; By minimizing l, we can get l and w t , the constraints are met during the solution process: Among them, B is the penalty term, y is the actual target value of the sample, w is the weight vector, ε is the threshold of tolerable error, and e i is the slack variable; The second model calculates the output for each input feature vector Where K is the number of weak learners, R is the original feature space of discrete manufacturing industry data, x i is the input feature vector, i = 1, 2, ... n; f K is a function in R, representing the model of the K-th tree trained; The objective function of the second model is By iteratively adding tree models to minimize the objective function; where loss is the discrete manufacturing industry dataset loss, M and is the number of decision tree child nodes and their weights, and C is the penalty coefficient.

2. The method for constructing an industrial data hybrid model for discrete manufacturing according to claim 1, characterized in that: The hybrid model is Y=w1y1+w2y2, where y1 is the first output, w1 is the weight of the first output, y2 is the second output, and w2 is the weight of the second output.

3. The method for constructing an industrial data hybrid model for discrete manufacturing according to claim 2, characterized in that: The weight w1 of the first output and the weight w2 of the second output are calculated using the least squares weight fusion method.

4. The method for constructing an industrial data hybrid model for discrete manufacturing according to claim 1, characterized in that: The R2 index is used to evaluate the parameters of the hybrid model.

5. An industrial data hybrid model construction system for discrete manufacturing, characterized by: include: Feature extraction unit, used to collect discrete manufacturing industrial data in real time and perform feature extraction to obtain feature vectors; a data processing unit, configured to obtain a first output and a second output by passing the feature data through a first model and a second model, respectively; the first model is configured to obtain a fitting curve for the feature data in a discrete space, and the second model is configured to perform ensemble learning on the feature data using a decision tree as a weak learner; a hybrid model building unit, configured to perform a weighted summation of the first output and the second output to obtain a hybrid model of the industrial data; The first model calculates the output for each input feature vector Where l represents industrial loss, w t is the weight vector of the input vector, is a high-dimensional feature space, x i is the input feature vector, i = 1, 2, ... n; By minimizing l, we can get l and w t , the constraints are met during the solution process: Among them, B is the penalty term, y is the actual target value of the sample, w is the weight vector, ε is the threshold of tolerable error, and e i is the slack variable; The second model calculates the output for each input feature vector Where K is the number of weak learners, R is the original feature space of discrete manufacturing industry data, x i is the input feature vector, i = 1, 2, ... n; f K is a function in R, representing the model of the K-th tree trained; The objective function of the second model is By iteratively adding tree models to minimize the objective function; where loss is the discrete manufacturing industry dataset loss, M and is the number of decision tree child nodes and their weights, and C is the penalty coefficient.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the computer program is loaded into a processor, the method for constructing an industrial data hybrid model for discrete manufacturing according to any one of claims 1 to 4 is implemented.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for constructing an industrial data hybrid model for discrete manufacturing according to any one of claims 1 to 4 is implemented.

Images