Environment intelligent interaction cognition system and method based on data driving

Abstract

The invention relates to the technical field of environment intelligent interaction, in particular to an environment intelligent interaction cognition system and method based on data driving, and the method comprises the steps: in a qualitative cognition stage, firstly carrying out the dimension reduction and structural classification of high-dimensional data through employing a PAM clustering algorithm, determining the optimal clustering number through a contour coefficient method, and achieving the preliminary organization of the data; then, an FP-Growth algorithm is used for mining association rules among features, an interpretable knowledge rule base is constructed, in the quantitative cognition stage, a qualitative analysis result is mapped back to original data, a POS-BP neural network model is adopted, the initial weight of a neural network is automatically optimized through a particle swarm optimization (PSO) algorithm, the convergence speed and prediction precision of the model are remarkably improved, and the prediction accuracy of the model is improved. And finally, accurate prediction of key indexes (quality, energy consumption, cost and the like) in the manufacturing process is realized through the optimized BP neural network. According to the method, effective mining of massive, high-dimensional and low-quality data in the continuous manufacturing industry is realized through a divide-and-conquer strategy.

Description

Technical Field

[0001] The present invention relates to the field of environmental intelligent interaction technology, and in particular to a data-driven environmental intelligent interaction cognitive system and method. Background Art

[0002] As most manufacturing companies have transitioned from traditional manual management to information-based management, they have accumulated vast amounts of data on processing, quality, equipment, personnel, and other aspects. In this context, extracting valuable information from this vast and high-dimensional manufacturing data has become a key challenge. Effective data mining and analysis can provide precise guidance for daily manufacturing operations, helping to reduce quality issues, energy consumption, and operating costs during production, while also improving production efficiency and driving continuous optimization and development in the industry.

[0003] Continuous manufacturing is an efficient and automated production method with broad application prospects and development potential. Big data analysis and machine learning models enable more precise control and optimization of production processes, helping to proactively identify potential problems and implement preventative measures, thereby improving overall operational efficiency and competitiveness. Compared to discrete manufacturing, which focuses on the discontinuous movement of materials and diverse processing methods, continuous manufacturing emphasizes continuous production processes. The data generated by these processes is large, multi-dimensional, correlated, coupled, and noisy. Directly using deep learning methods to build neural network models for production factor prediction can lead to network node redundancy, increased computational costs, the curse of dimensionality, and model overfitting.

[0004] In summary, how to effectively process the multi-dimensional, correlated and coupled data generated by the continuous manufacturing industry has become a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The purpose of the present invention is to provide a data-driven environmental intelligent interactive cognition system and method to solve the problem in the prior art of lacking an effective data processing method for multi-dimensional, correlated coupling data of continuous manufacturing workshops.

[0006] To achieve the above objectives, the present invention provides a data-driven method for intelligent interactive cognition of an environment, which comprises the following steps:

[0007] Data cleaning: Select the data set generated during the production process, delete duplicate values ​​in the data, and fill in missing values;

[0008] Qualitative cognition: Use the PAM clustering algorithm to classify continuous data and establish labels, then use the FP-Growth association rule relationship algorithm to mine the relationship between data features and indicators, and build a knowledge rule base;

[0009] Quantitative cognition: Traverse the knowledge obtained in the qualitative cognition stage, return the data labels to the cleaned raw data, and establish a POS-BP neural network model. At the same time, the collected and cleaned data is input into the POS-BP neural network model as training samples, and the initial weights of the BP neural network are optimized using the POS algorithm;

[0010] Index prediction: Utilize the optimized BP neural network for training, update the weights through forward and back propagation, predict the performance indicators in the continuous manufacturing process, and complete the quantitative understanding of multi-dimensional data.

[0011] Among them, in the step "data cleaning", if the missing data does not affect the overall analysis, the entire row or column of data can be deleted; if the data is evenly distributed and there are few missing values, the corresponding statistical values ​​can be used to complete the data; if the data is complex and there are many missing values, regression models and random forest models can be used to predict missing values.

[0012] The specific steps of the step "qualitative cognition" include the following:

[0013] The PAM clustering algorithm is used to divide the cleaned data set into k clusters, and the silhouette coefficient method is used to determine the optimal number of clusters to complete the structured classification of the data;

[0014] Use FP-Growth algorithm to mine frequent item sets, establish association rules between features and indicators, and form a qualitative cognitive model;

[0015] Perform redundancy detection and cleaning on association rules, merge similar knowledge, and finally generate a refined knowledge rule base.

[0016] Among them, the specific steps of step "qualitative cognition" are:

[0017] Map the knowledge rules generated in the qualitative cognition stage back to the original data and reconstruct it into a labeled training dataset;

[0018] A POS-BP neural network model was constructed, and the particle swarm optimization algorithm (PSO) was used to optimize the initial weight parameters of the BP neural network. The reconstructed training data was used for model training and optimization.

[0019] The specific steps of the step "constructing a POS-BP neural network model, optimizing the initial weight parameters of the BP neural network using a particle swarm optimization algorithm (PSO), and using the reconstructed training data for model training and optimization" include the following:

[0020] Neural network structure design: Construct a BP neural network structure consisting of an input layer, a double hidden layer, and an output layer. The number of nodes in each layer is determined based on the characteristic elements and the number of prediction targets.

[0021] PSO parameter initialization: set PSO algorithm parameters, including particle swarm size, maximum number of iterations, inertia weight, learning factor and maximum speed;

[0022] Particle swarm optimization process: By iteratively updating the particle speed and position, the initial weights of the neural network are optimized until the termination condition is met.

[0023] In the step “neural network structure design”, the input layer is a layer with N neurons. m (∪P i The number of characteristic elements) is , and the number of neurons in the hidden layer is N. i and N k , the output layer is one layer, the number of nodes is N j (The number of characteristic elements of K)

[0024] Among them, in the step "particle swarm optimization process", the PSO algorithm searches the solution space by iteratively updating the speed and position of the particles. The speed update formula is specifically:

[0025] v i (t+1)=w*v i (t)+c1*r1*(p best,i -x i (t))+c2*r2*(g best -x i (t))

[0026] Where, v i (t): velocity of particle i at iteration t;

[0027] x i (t): the position of particle i at the tth iteration;

[0028] w: inertia weight, which controls the influence of particle historical velocity;

[0029] c1, c2: acceleration constants, controlling the particle's dependence on the individual optimum and the global optimum;

[0030] r1, r2: random numbers, ranging from [0,1], to increase randomness;

[0031] p best,i : the individual optimal position of particle i;

[0032] g best : global optimal position;

[0033] The particle position update formula is as follows:

[0034] x i (t+1)=xi (t)+v i (t+1).

[0035] The specific steps of the step "Indicator Prediction" include the following:

[0036] Forward propagation calculation: Calculate the input and output values ​​of hidden layer 1, hidden layer 2, and output layer in sequence, using ReLU and Softmax activation functions for nonlinear transformation respectively;

[0037] Error back propagation: Based on the difference between the actual output and the predicted output, the error value of the output layer and the hidden layer is calculated layer by layer;

[0038] Weight update: Based on the error calculation results, the gradient descent method is used to adjust the weight matrix and threshold parameters of each layer;

[0039] Model application: Apply the trained neural network model to the manufacturing process to predict key performance indicators such as quality, energy consumption, and cost;

[0040] Performance verification: The prediction accuracy of the model is evaluated through indicators such as mean square error and goodness of fit to verify the effectiveness of the model.

[0041] The present invention also provides a data-driven environmental intelligent interactive cognition system, which uses the data-driven environmental intelligent interactive cognition method as described above to complete data processing of multi-dimensional, correlated coupling data generated by continuous manufacturing.

[0042] The present invention proposes a data-driven, intelligent interactive cognitive system and method for environmental data. In the qualitative cognitive phase, the PAM clustering algorithm is first used to reduce the dimensionality and perform structured classification on high-dimensional data. The silhouette coefficient method is then used to determine the optimal number of clusters, achieving preliminary data organization. Subsequently, the FP-Growth algorithm is used to mine association rules between features and construct an interpretable knowledge rule base. In the quantitative cognitive phase, the qualitative analysis results are mapped back to the original data. A POS-BP neural network model is employed, and the particle swarm optimization (PSO) algorithm is used to automatically optimize the initial neural network weights, significantly improving the model's convergence speed and prediction accuracy. Ultimately, the optimized BP neural network enables accurate prediction of key manufacturing process indicators (quality, energy consumption, cost, etc.). This method, through a "divide and conquer" strategy, effectively mines massive, high-dimensional, low-quality data in the continuous manufacturing industry, providing reliable technical support for the transformation of manufacturing enterprises from "informatization" to "intelligence." BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is a flowchart of the steps of the data-driven environmental intelligent interactive cognition method provided by the present invention.

[0045] Figure 2 This is the particle swarm optimization algorithm + neural network flow chart provided by the present invention.

[0046] Figure 3 This is the BP neural network design diagram provided by the present invention.

[0047] Figure 4 This is an overview of the overall process of the data-driven environmental intelligent interactive cognition method provided by the present invention.

[0048] Figure 5 It is a schematic diagram of the cognitive method provided by the present invention.

[0049] Figure 6 It is a quality and energy consumption index data fitting diagram provided by the present invention.

[0050] Figure 7 It is a schematic diagram of the evaluation results provided by the present invention. DETAILED DESCRIPTION

[0051] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0052] See also Figures 1 to 7 The present invention provides a data-driven environment intelligent interactive cognition method, which includes the following steps:

[0053] Data cleaning: Select the data set generated during the production process, delete duplicate values ​​in the data, and fill in missing values;

[0054] Qualitative cognition: Use the PAM clustering algorithm to classify continuous data and establish labels, then use the FP-Growth association rule relationship algorithm to mine the relationship between data features and indicators, and build a knowledge rule base;

[0055] Quantitative cognition: Traverse the knowledge obtained in the qualitative cognition stage, return the data labels to the cleaned raw data, and establish a POS-BP neural network model. At the same time, the collected and cleaned data is input into the POS-BP neural network model as training samples, and the initial weights of the BP neural network are optimized using the POS algorithm;

[0056] Index prediction: Utilize the optimized BP neural network for training, update the weights through forward and back propagation, predict the performance indicators in the continuous manufacturing process, and complete the quantitative understanding of multi-dimensional data.

[0057] In this embodiment, in the qualitative cognition stage, the PAM clustering algorithm is first used to reduce the dimensionality and perform structured classification on high-dimensional data, and the optimal number of clusters is determined by the silhouette coefficient method to achieve preliminary organization of the data; then the FP-Growth algorithm is used to mine the association rules between features and construct an interpretable knowledge rule base. In the quantitative cognition stage, the qualitative analysis results are mapped back to the original data, and the POS-BP neural network model is used to automatically optimize the initial weights of the neural network through the particle swarm optimization algorithm (PSO), which significantly improves the model convergence speed and prediction accuracy. Finally, the optimized BP neural network is used to achieve accurate prediction of key indicators of the manufacturing process (quality, energy consumption, cost, etc.). This method realizes the effective mining of massive, high-dimensional, low-quality data in the continuous manufacturing industry through a "divide and conquer" strategy, providing reliable technical support for the transformation of manufacturing enterprises from "informatization" to "intelligence".

[0058] Furthermore, in the step "data cleaning", if the missing data does not affect the overall analysis, the entire row or column of data can be deleted; if the data is evenly distributed and there are few missing values, the corresponding statistical values ​​can be used to complete the data; if the data is complex and there are many missing values, regression models and random forest models can be used to predict the missing values.

[0059] Furthermore, the specific steps of the step “qualitative cognition” include the following:

[0060] The PAM clustering algorithm is used to divide the cleaned data set into k clusters, and the silhouette coefficient method is used to determine the optimal number of clusters to complete the structured classification of the data;

[0061] Use FP-Growth algorithm to mine frequent item sets, establish association rules between features and indicators, and form a qualitative cognitive model;

[0062] Perform redundancy detection and cleaning on association rules, merge similar knowledge, and finally generate a refined knowledge rule base.

[0063] In this embodiment, the step of "using the PAM clustering algorithm to divide the cleaned data set into k clusters, and determining the optimal number of clusters by the silhouette coefficient method to complete the structured classification of the data" specifically refers to:

[0064] Using the PAM clustering algorithm, m data objects are divided into k clusters. In order to make the classification results clearer, the average silhouette coefficient of different k values ​​is calculated by the silhouette coefficient method, and the k value with the largest silhouette coefficient is selected. When the silhouette coefficient is closer to 1, the clustering effect is better. The basic principle of the PAM clustering algorithm is to first randomly select k center points as the cluster center of each cluster, and then calculate the similarity or distance between the remaining data points and these cluster centers to perform cluster assignment and divide them into the nearest cluster. Finally, the cost function is used for evaluation, and the non-center points are continuously used to replace the center points to improve the clustering effect, thereby giving the best clustering analysis and constructing cluster 1: Medoid is A(x1, x2

[0065] ,…,x n ), containing similar labels such as data points A, B, and C;

[0066] The specific contents of the step "Using the PAM clustering algorithm to divide the cleaned data set into k clusters, and determining the optimal number of clusters through the silhouette coefficient method to complete the structured classification of the data" include:

[0067] Using the FP-Growth association rule relationship algorithm, the product data knowledge obtained through data preprocessing is scanned into a data set to find the association relationship between different features and indicators, accumulate the count of each item, and mine the items that meet the minimum support. When selecting the minimum support, you can first count the frequency distribution of the item set, and then select the inflection point of the frequency distribution as the minimum support; you can also set a smaller empirical initial value first, and adjust it according to the number of rules. If there are too many rules, the support will be increased, and if there are too few rules, it will be reduced. When the data noise is large, the support should be set higher to reduce the impact of the noise; when the amount of data is large, the support can be set lower.

[0068] Obtain frequent item sets from the data set, create an empty root node, and add the results of each item's support in descending order to the root node; then build an FP-tree, and add items to the FP-tree layer by layer according to the size of the items' support calculated earlier, according to the sorting results, until all items are added to the FP-tree; finally, mine frequent item sets with association rules based on the FP-tree until all frequent item sets are mined. Establish qualitative cognition and establish a mathematical model of knowledge: Where, P i and K represent the set of conditions and conclusions of the corresponding knowledge, respectively, for example Indicates that conclusion K can be derived from conditions P1 and P2;

[0069] The specific contents of the step "detecting and cleaning redundancy in association rules, merging similar knowledge, and finally generating a refined knowledge rule base" include:

[0070] Extract information from the obtained knowledge, extract the correlation between each piece of knowledge and merge similar knowledge to form a refined knowledge rule base, thus completing the qualitative cognition of the data. ① If there are two pieces of knowledge and and satisfy Knowledge It is knowledge Redundancy can be removed during cleaning ②If there are two pieces of knowledge and and satisfy Knowledge It is knowledge Redundancy can be removed during cleaning

[0071] Furthermore, the specific steps of the step "qualitative cognition" are:

[0072] Map the knowledge rules generated in the qualitative cognition stage back to the original data and reconstruct it into a labeled training dataset;

[0073] A POS-BP neural network model was constructed, and the particle swarm optimization algorithm (PSO) was used to optimize the initial weight parameters of the BP neural network. The reconstructed training data was used for model training and optimization.

[0074] Furthermore, the specific steps of the step “constructing a POS-BP neural network model, optimizing the initial weight parameters of the BP neural network using a particle swarm optimization algorithm (PSO), and using the reconstructed training data for model training and optimization” include the following:

[0075] Neural network structure design: Construct a BP neural network structure consisting of an input layer, a double hidden layer, and an output layer. The number of nodes in each layer is determined based on the characteristic elements and the number of prediction targets.

[0076] PSO parameter initialization: set PSO algorithm parameters, including particle swarm size, maximum number of iterations, inertia weight, learning factor and maximum speed;

[0077] Particle swarm optimization process: By iteratively updating the particle speed and position, the initial weights of the neural network are optimized until the termination condition is met.

[0078] In this embodiment, the BP neural network structure is determined: the input layer is a layer with N neurons.m (∪P i The number of characteristic elements) is , and the number of neurons in the two hidden layers is N i and N k , the output layer is one layer, the number of nodes is N j (Number of characteristic elements of K)

[0079] The number of particles in the group determines the coverage of the search space, and the number of particles is selected according to the empirical formula D in the formula is the dimension of the problem; the maximum number of iterations T is generally set between 100 and 1000 according to the problem scale and the expected convergence accuracy; the inertia weight w controls the influence of the particle's previous velocity on the current velocity and is usually set between [0.4, 1.2]; the individual learning factor c1 (which measures the ability of the particle to approach its own historical optimal position) and the social learning factor c2 (which measures the ability of the particle to approach the global optimal position) are usually set to c1 = c2 = 2.0, which can ensure that the particles have both exploration ability and convergence; the maximum speed Vmax is the maximum value that limits the particle speed to prevent particles from moving too fast and missing the global optimal solution. The selection method is usually set to 10%-20% of the problem search space range;

[0080] The PSO algorithm simulates the search process of a particle group in the solution space and gradually approaches the optimal solution by adjusting the speed and position of the particles. The particles achieve a balance between the individual optimal position and the global optimal position, combining global and local search, and finally converge to the optimal solution.

[0081] N m is the number of nodes in the output layer of the BP neural network; d k ,y k are the expected output and actual output of the kth node in the output layer respectively;

[0082] The PSO algorithm searches the solution space by iteratively updating the velocity and position of particles. The velocity update formula is:

[0083] v i (t+1)=w*v i (t)+c1*r1*(p best,i -x i (t))+c2*r2*(g best -x i (t))

[0084] v i (t): velocity of particle i at iteration t;

[0085] x i (t): the position of particle i at the tth iteration;

[0086] w: inertia weight, which controls the influence of particle historical velocity;

[0087] c1, c2: acceleration constants, controlling the particle's dependence on the individual optimum and the global optimum;

[0088] r1, r2: random numbers, ranging from [0,1], to increase randomness;

[0089] p best,i : the individual optimal position of particle i;

[0090] g best : global optimal position;

[0091] Particle position update formula:

[0092] x i (t+1)=x i (t)+v i (t+1);

[0093] Update the individual best and global best, and for each particle, calculate the objective function value of the current position:

[0094] F(x i (t+1))

[0095] The objective function value generally uses the loss of the BP neural network on the training set or the accuracy on the validation set;

[0096] If the objective function value of the current position is better than the individual historical best (that is, the value of the current position is smaller, indicating a better solution), then the individual best position p of the particle is updated. best,i , compare the individual best positions of all particles and update the global best position g best , that is, find the best solution among all current particles;

[0097] PSO terminates when any of the following conditions are met, reaching the maximum number of iterations; convergence conditions are met: for example, the global optimal solution changes very little over multiple consecutive iterations, and the algorithm is considered to be close to the optimal solution; the objective function value reaches the set threshold;

[0098] The particle swarm search process finds the optimal solution through individual search, social learning, and information exchange;

[0099] Individual search: Each particle performs a local search based on its own historical best position, that is, it tends to continue along the path it has previously taken;

[0100] Social learning: Each particle will perform a global search based on the global optimal position, that is, it will be influenced by other particles and move closer to the global optimal solution;

[0101] Information exchange: Particles guide the search direction through individual optimal positions and global optimal positions, which is equivalent to collaboration between particles;

[0102] Use the global optimal weight g output by PSO best As the initial weight of the BP neural network.

[0103] Furthermore, the specific steps of the step "indicator prediction" include the following:

[0104] Forward propagation calculation: Calculate the input and output values ​​of hidden layer 1, hidden layer 2, and output layer in sequence, using ReLU and Softmax activation functions for nonlinear transformation respectively;

[0105] Error back propagation: Based on the difference between the actual output and the predicted output, the error value of the output layer and the hidden layer is calculated layer by layer;

[0106] Weight update: Based on the error calculation results, the gradient descent method is used to adjust the weight matrix and threshold parameters of each layer;

[0107] Model application: Apply the trained neural network model to the manufacturing process to predict key performance indicators such as quality, energy consumption, and cost;

[0108] Performance verification: The prediction accuracy of the model is evaluated through indicators such as mean square error and goodness of fit to verify the effectiveness of the model.

[0109] In this embodiment, the calculation of the forward propagation step specifically includes:

[0110] Input value of hidden layer 1:

[0111] Where, Represents the data matrix corresponding to a sample of the training set of performance indicators (quality, energy consumption, cost) in the manufacturing process, N is the distance from the input layer to the hidden layer 1 i ×N m Weight matrix;

[0112] Since it is set to two hidden layers, the output value of hidden layer 1 is:

[0113] In the formula, g(z) represents the activation function, and g(z) uses the ReLU function; Represents the input value of hidden layer 1;

[0114] Input value of hidden layer 2:

[0115] Where, Represents N k ×N iThe weight matrix, where N i Represents the number of nodes in hidden layer 1, N m represents the number of nodes in hidden layer 2;

[0116] Output value of hidden layer 2:

[0117] Output layer input value:

[0118] Where, Represents N j ×N k The weight matrix, where N j Represents the number of output layer nodes, that is, the number of label categories of predicted factors (quality, energy consumption, cost), N k represents the number of nodes in hidden layer 2; Represents the hidden layer output value;

[0119] Output layer output value:

[0120] In the formula, g(z) represents the activation function, and g(z) uses the softmax function; Represents the output layer input value.

[0121] The specific contents of back propagation calculation error include:

[0122] Output layer error:

[0123] Hidden layer error:

[0124] Where, Represents the output value of the output layer; Represents the actual output value matrix; Represents the transpose of the weight matrix between the hidden layer and the output layer.

[0125] The specific content of updating the threshold and weight of the network according to the error is:

[0126] The updated weights are calculated according to the following formula:

[0127]

[0128] Where, is the expected output matrix, is the actual output matrix, is the input matrix.

[0129] According to the trained data recognition model based on POS-BP neural network, the corresponding performance indicators (such as quality, energy consumption, cost, etc.) in the continuous manufacturing process are predicted, and the cognition of the multi-dimensional data of the entire manufacturing workshop is improved.

[0130] like Figure 6 As shown in the figure, the performance of the model is mainly reflected in the degree of fit between the model prediction value and the actual result during prediction. The main evaluation index, Mean Square Error (MSE), is as small as possible, and close to 0 indicates a perfect prediction; the goodness of fit (R-squared, R 2 ) is closer to 1, indicating that the model fits the data perfectly.

[0131] The present invention also provides a data-driven environmental intelligent interactive cognition system, which uses the data-driven environmental intelligent interactive cognition method as described above to complete data processing of multi-dimensional, correlated coupling data generated by continuous manufacturing.

[0132] The above disclosure is only a preferred embodiment of the present invention, and certainly cannot be used to limit the scope of the rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made in accordance with the claims of the present invention are still within the scope of the invention.

Claims

1. A data-driven environmental intelligent interactive cognition method, characterized in that: The steps include: Data cleaning: Select the data set generated during the production process, delete duplicate values ​​in the data, and fill in missing values; Qualitative cognition: Use the PAM clustering algorithm to classify continuous data and establish labels, then use the FP-Growth association rule relationship algorithm to mine the relationship between data features and indicators, and build a knowledge rule base; Quantitative cognition: Traverse the knowledge obtained in the qualitative cognition stage, return the data labels to the cleaned raw data, and establish a POS-BP neural network model. At the same time, the collected and cleaned data is input into the POS-BP neural network model as training samples, and the initial weights of the BP neural network are optimized using the POS algorithm; Index prediction: Utilize the optimized BP neural network for training, update the weights through forward and back propagation, predict the performance indicators in the continuous manufacturing process, and complete the quantitative understanding of multi-dimensional data.

2. The data-driven environment intelligent interactive cognition method according to claim 1, characterized in that: In the "Data Cleaning" step, if missing data does not affect the overall analysis, you can delete the entire row or column of data. If the data is evenly distributed and there are few missing values, you can use the corresponding statistical values ​​to complete the data. If the data is complex and there are many missing values, you can use regression models or random forest models to predict the missing values.

3. The data-driven environment intelligent interactive cognition method according to claim 2, characterized in that: The specific steps of the "qualitative cognition" step include the following: The PAM clustering algorithm is used to divide the cleaned data set into k clusters, and the silhouette coefficient method is used to determine the optimal number of clusters to complete the structured classification of the data; Use FP-Growth algorithm to mine frequent item sets, establish association rules between features and indicators, and form a qualitative cognitive model; Perform redundancy detection and cleaning on association rules, merge similar knowledge, and finally generate a refined knowledge rule base.

4. The data-driven environment intelligent interactive cognition method according to claim 3, characterized in that: The specific steps of the "qualitative cognition" step are: Map the knowledge rules generated in the qualitative cognition stage back to the original data and reconstruct it into a labeled training dataset; A POS-BP neural network model was constructed, and the particle swarm optimization algorithm (PSO) was used to optimize the initial weight parameters of the BP neural network. The reconstructed training data was used for model training and optimization.

5. The data-driven environment intelligent interactive cognition method according to claim 4, characterized in that: The specific steps of the step "constructing a POS-BP neural network model, optimizing the initial weight parameters of the BP neural network using the particle swarm optimization algorithm (PSO), and using the reconstructed training data for model training and optimization" include the following: Neural network structure design: Construct a BP neural network structure consisting of an input layer, a double hidden layer, and an output layer. The number of nodes in each layer is determined based on the characteristic elements and the number of prediction targets. PSO parameter initialization: set PSO algorithm parameters, including particle swarm size, maximum number of iterations, inertia weight, learning factor and maximum speed; Particle swarm optimization process: By iteratively updating the particle speed and position, the initial weights of the neural network are optimized until the termination condition is met.

6. The data-driven environment intelligent interactive cognition method according to claim 5, characterized in that: In the step "neural network structure design", the input layer is a layer with N neurons. m ( The number of characteristic elements) is , and the number of neurons in the hidden layer is N. i and N k , the output layer is one layer, the number of nodes is N j (The number of characteristic elements of K) 7. The data-driven environment intelligent interactive cognition method according to claim 6, characterized in that: In the step "Particle Swarm Optimization Process", the PSO algorithm searches the solution space by iteratively updating the speed and position of the particles. The speed update formula is as follows: v i (t+1)=w*v i (t)+c1*r1*(p best,i -x i (t))+c2*r2*(g best -x i (t)) Where, v i (t): velocity of particle i at iteration t; x i (t): the position of particle i at the tth iteration; w: inertia weight, which controls the influence of particle historical velocity; c1, c2: acceleration constants, controlling the particle's dependence on the individual optimum and the global optimum; r1, r2: random numbers, ranging from [0,1], to increase randomness; p best,i : the individual optimal position of particle i; g best : global optimal position; The particle position update formula is as follows: x i (t+1)=x i (t)+v i (t+1)。 8. The data-driven environment intelligent interactive cognition method according to claim 7, characterized in that: The specific steps of "Indicator Forecasting" are as follows: Forward propagation calculation: Calculate the input and output values ​​of hidden layer 1, hidden layer 2, and output layer in sequence, using ReLU and Softmax activation functions for nonlinear transformation respectively; Error back propagation: Based on the difference between the actual output and the predicted output, the error value of the output layer and the hidden layer is calculated layer by layer; Weight update: Based on the error calculation results, the gradient descent method is used to adjust the weight matrix and threshold parameters of each layer; Model application: Apply the trained neural network model to the manufacturing process to predict key performance indicators such as quality, energy consumption, and cost; Performance verification: The prediction accuracy of the model is evaluated through indicators such as mean square error and goodness of fit to verify the effectiveness of the model.

9. A data-driven environment intelligent interactive cognitive system, characterized by: The data-driven environmental intelligent interactive cognition method as claimed in claim 1 is applied to complete data processing of multi-dimensional, correlated coupling data generated by continuous manufacturing.

Images