A method and device for monitoring abnormal key performance indicators in float glass production process

Abstract

The present invention relates to the technical field of chemical process quality monitoring, and in particular to a method and device for monitoring abnormalities of key performance indicators in a float glass production process. The method comprises: obtaining empirical knowledge and historical data of the float glass production process; screening out variables related to waviness using empirical knowledge and the maximum relevant minimum redundancy method; setting knowledge confidence according to the knowledge source, fusing knowledge and the results of data mining to obtain the final variables related to waviness; constructing a time series knowledge graph of float glass through data and knowledge fusion drive; converting the time series knowledge graph into an association graph, and extracting multi-level spatiotemporal features through differentiable pooling and long short-term memory networks; incorporating fault information into statistics through supervised comparative learning; calculating control limits through kernel density estimation; and inputting the data to be monitored into the monitoring model for fault monitoring. The present invention can effectively monitor key performance indicators of the float glass production process and ensure product quality and production efficiency.

Description

Technical Field

[0001] The invention relates to the technical field of chemical process quality monitoring, and in particular to a method and a device for monitoring abnormalities of key performance indicators in a float glass production process. Background Art

[0002] The production process of float glass mainly consists of raw materials and ingredients, glass melting and melting furnace, glass molding and tin bath, glass annealing and annealing furnace. It takes at least 48 hours for the raw materials to enter the melting furnace and flow into the tin bath for molding and cooling. This process is theoretically divided into five stages, and the glass solution will undergo a series of physical, chemical, and physicochemical reactions. The progress of these reactions is closely related to the output and quality of the glass. However, in the actual production process, the five stages are carried out simultaneously, which greatly increases the difficulty of monitoring the key performance indicators (waviness) of the float glass production process.

[0003] Failures occurred in the float glass production process. At the very least, the product was unqualified and had to be re-melted or the plate exploded or broke, causing economic losses, and at worst, it could cause casualties. At present, the measurement method of waviness, a key performance indicator of the quality of finished glass products, can only be destructive testing. This means that only sampling of finished glass products can tell whether the production is qualified. When unqualified products are detected, the production line has been producing and running unqualified glass for at least 48 hours, and these glasses can only be re-melted. This brings a huge blow and property losses to the production of high-quality glass. Therefore, the study of real-time monitoring methods for the key performance indicator waviness of the float glass manufacturing process is particularly important in terms of ensuring production safety, improving production quality and efficiency, and improving production economic benefits. It is an important content that can promote manufacturing transformation and upgrading. Summary of the invention

[0004] In order to solve the technical problems in the prior art that the float glass production process is complicated and has many steps, and the key performance indicator waviness cannot be measured in real time, and it is often difficult to find the fault when it occurs, the embodiment of the present invention provides a method and device for monitoring the abnormality of the key performance indicators in the float glass production process. The technical solution is as follows:

[0005] On the one hand, a method for monitoring abnormalities of key performance indicators in a float glass production process is provided, the method comprising:

[0006] S1. Acquire empirical knowledge and historical data of float glass production process;

[0007] S2. Filter the empirical variable set related to the waviness based on empirical knowledge; filter the actual variable set related to the waviness through the maximum correlation minimum redundancy algorithm based on historical data;

[0008] S3, setting different confidence levels for the empirical knowledge and presetting a confidence threshold; obtaining the union of the waviness-related variables in the empirical knowledge whose confidence levels are higher than the confidence threshold and the results of data mining;

[0009] S4. Construct a temporal knowledge graph of the float glass production process through knowledge and data fusion;

[0010] S5. Construct a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the time series knowledge graph and construct monitoring statistics;

[0011] S6. Monitor the abnormality of corrugation. Obtain real-time data. Input the real-time data into the monitoring model. Output the statistics of the real-time data through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred. Otherwise, no corrugation-related fault has occurred.

[0012] Optionally, in S2, the empirical variable set related to the waviness is screened according to the empirical knowledge; and the actual variable set related to the waviness is screened by the maximum correlation minimum redundancy algorithm according to the historical data, including:

[0013] By studying the glass production process, asking on-site engineers, and consulting papers, we can obtain variables related to waviness and obtain a knowledge variable set;

[0014] The maximum correlation and minimum redundancy (mRMR) algorithm is used to select the corrugation-related variables that have the greatest correlation with the corrugation but the smallest correlation with each other; the calculation results of the maximum correlation and the calculation results of the minimum redundancy are combined;

[0015] The optimal solution is obtained through the incremental search method, and the corrugation-related variables are screened by maximizing the combined results to obtain the actual variable set related to the corrugation based on the process data.

[0016] Optionally, in S3, different confidence levels are set for the empirical knowledge, and a confidence threshold is preset; and a union of the waviness-related variables with confidence levels higher than the confidence threshold in the empirical knowledge and the results of data mining is obtained, including:

[0017] Different confidence levels are set according to the reliability of the knowledge source; including: setting the confidence level of knowledge with high reliability from physical formulas, chemical equations, and experimental results of journal articles to 1; setting the confidence level of knowledge from engineers' production experience and intuition to 0.5-0.9 according to the engineers' recognition; when the same knowledge has multiple sources, the confidence level is averaged;

[0018] Presetting a confidence threshold; extracting a knowledge variable set related to the waviness after knowledge screening above the confidence threshold;

[0019] The knowledge variable set above the confidence threshold and the waviness-related variable set selected by maximum correlation and minimum redundancy are combined.

[0020] Optionally, in S4, a temporal knowledge graph of the float glass production process is constructed, including:

[0021] Define entities of the temporal knowledge graph of the float glass production process; the entities include waviness-related variables; define the relationships between entities based on expert knowledge and historical data;

[0022] Based on the relationship between entities, the following temporal knowledge graph is constructed:

[0023] TKG=<E,R,S,T>

[0024] Among them, E is the set of entities; R is the set of relations; S is the set of all static knowledge triples; T is the set of all connected time event quads;

[0025] Slice the temporal knowledge graph according to time and transform the quadruple into triple (es, t, r, eo, t), 0<t≤T; where e s,t is the head entity at time t, r is the association relationship, e o,t is the tail entity corresponding to time t.

[0026] Optionally, the relationships between entities are jointly defined based on expert knowledge and historical data, including:

[0027] When entities are obtained from text, equations, or engineering experience, the relationships between entities can be directly defined;

[0028] When entities can be determined to have interactions based on expert experience knowledge, but their size cannot be quantified or it is impossible to determine whether there is an interactive relationship, the distance correlation coefficient that takes time lag into account is used to calculate the relationship between the entities.

[0029] Optionally, a distance correlation coefficient, including:

[0030] The distance correlation coefficient between two entities u and v is denoted as DCOR(u,v). The two entities u and v are independent of each other if and only if DCOR(u,v)=0; the stronger the correlation between the two entities u and v, the larger the DCOR(u,v).

[0031] Optionally, in step S5, based on a monitoring model of differentiable pooling-long short-term memory network and supervised contrastive learning, multi-level spatiotemporal features in the time series knowledge graph are extracted and monitoring statistics are constructed, including:

[0032] Monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning;

[0033] Through the monitoring model's differentiable pooling-long short-term memory network, the multi-level spatiotemporal features of the temporal knowledge graph are extracted;

[0034] Incorporate fault data when constructing monitoring statistics through supervised comparative learning of monitoring models;

[0035] Calculation of control limits for waviness monitoring by kernel density estimation.

[0036] Optionally, extract multi-level spatiotemporal features in the temporal knowledge graph, including:

[0037] Spatial features are extracted by aggregating neighbor information through GCN, and hierarchical information is extracted by learning the allocation matrix through differentiable pooling. The float glass production process is divided into four levels from low to high: sensor level, equipment level, sub-process level, and process level. The nodes of the lower level are allocated to the higher level, and the node features of the lth level and the allocation matrix of the lth level are calculated.

[0038] Based on the node embedding and adjacency matrix of the node embedding layer l+1, the association graph of the l+1 layer is obtained, and the allocation matrix and association graph of each level are repeatedly calculated to obtain spatial features of multiple levels;

[0039] Through the long short-term memory network, the current state is connected with the long-term historical data to extract the long-term dependencies in the multi-level spatial features; the updated cell state and hidden state are repeatedly passed through the output gate of the long short-term memory network combined with the forgetting gate output, and passed to the next time series to extract the long-term associations in the time series data.

[0040] Optionally, the fault information is incorporated into the statistics through supervised comparative learning of the monitoring model; and the control limit of the waviness monitoring is calculated through kernel density estimation, including:

[0041] Obtain the linear model parameter matrix and multi-level spatiotemporal features, and calculate the control limits through kernel density estimation;

[0042] Construct positive and negative samples corresponding to the features to be predicted, where the positive samples are the features of the real samples at the next τ moments, and the negative samples are the features of the known fault samples in the historical data;

[0043] Obtain a fault sample data set and construct statistics by combining control limits and positive and negative samples;

[0044] The control limit of the statistic is calculated by kernel density estimation, wherein the kernel density estimation models the data distribution of the statistic when the corrugation meets the production requirements; the probability density function of the statistic is constructed, the probability density function of the statistic is integrated to obtain the cumulative distribution function, and the statistic with a cumulative distribution function value of 95% is selected as the control limit of the corrugation monitoring.

[0045] On the other hand, a device for monitoring abnormal key performance indicators in a float glass production process is provided, and the device is applied to a method for monitoring abnormal key performance indicators in a float glass production process. The device comprises:

[0046] Data acquisition module, used to obtain empirical knowledge and historical data of the float glass production process;

[0047] The variable screening module is used to screen the empirical variable set related to the corrugation based on empirical knowledge; based on historical data, the actual variable set related to the corrugation is screened by the maximum correlation minimum redundancy algorithm;

[0048] The knowledge and data fusion module is used to set different confidence levels for the empirical knowledge and preset the confidence threshold; obtain the union of the waviness-related variables with confidence levels higher than the confidence threshold in the empirical knowledge and the results of data mining;

[0049] The time series knowledge graph module is used to construct a time series knowledge graph of the float glass production process through knowledge and data fusion;

[0050] Monitoring model building module, used to build a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the time series knowledge graph and construct monitoring statistics;

[0051] The abnormality monitoring module is used for monitoring the abnormality of corrugation, acquiring real-time data, inputting the real-time data into the monitoring model, and outputting the statistics of the real-time data through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred; otherwise, no corrugation-related fault has occurred.

[0052] On the other hand, a device for monitoring abnormalities of key performance indicators in a float glass production process is provided, the device comprising: a processor; a memory, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, any one of the above-mentioned methods for monitoring abnormalities of key performance indicators in a float glass production process is implemented.

[0053] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement any one of the above-mentioned methods for monitoring abnormalities of key performance indicators in a float glass production process.

[0054] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0055] The method provided by the present invention 1) monitors the change of the corrugation of the float glass production process in real time through a monitoring model, and when the corrugation is abnormal, an alarm is given in time to ensure production safety and improve production efficiency;

[0056] 2) Based on the time series knowledge graph, the certainty of knowledge and the uncertainty of data can be effectively integrated to improve the interpretability and monitoring performance of the model;

[0057] 3) Based on supervised comparative learning, the monitoring model has good robustness, the model's ability to identify faults is improved, the credibility of the monitoring results is increased, and it can greatly improve the product waviness quality and production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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 creative work.

[0059] Figure 1 This is a flow chart of a method for monitoring abnormalities of key performance indicators in a float glass production process provided by an embodiment of the present invention;

[0060] Figure 2 is a float glass production flow chart provided by an embodiment of the present invention;

[0061] Figure 3 A screening process for waviness-related monitoring variables provided for an example of the present invention;

[0062] Figure 4 A flowchart of the screening results of the fused waviness-related variables provided for the example of the present invention;

[0063] Figure 5 A schematic diagram of a float glass time series knowledge graph provided for an example of the present invention;

[0064] Figure 6 A schematic diagram of a monitoring model provided for an example of the present invention;

[0065] Figure 7 A schematic diagram of a multi-level feature extraction model provided by an example of the present invention;

[0066] Figure 8 This is a block diagram of a device for monitoring abnormal key performance indicators in a float glass production process provided by an embodiment of the present invention;

[0067] Fig. 9 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0069] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0070] In the embodiments of the present invention, "image" and "picture" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same. "of", "corresponding", and "corresponding" can sometimes be used interchangeably. It should be noted that when the difference between them is not emphasized, the meanings they intend to express are the same.

[0071] In the embodiments of the present invention, sometimes the subscripts such as W 1 It may be expressed in non-subscript form such as W1. When the difference is not emphasized, the meaning is the same.

[0072] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0073] Since the production process of float glass is complex and has many steps, and the key performance indicator waviness cannot be measured in real time, it is often difficult to detect when a fault occurs. In order to monitor the key performance indicator-related faults in the production process as early as possible, process monitoring technology is an extremely effective method. Considering the influence of dynamic factors such as equipment and raw materials in the production process, making full use of empirical knowledge and process data, studying how to integrate the results of knowledge and data mining to screen out monitoring variables related to key performance indicators has become a key issue in quality-related monitoring. Secondly, considering that the glass manufacturing process has a set of rigorous theoretical knowledge and the collected process data often has dynamic, time-varying, nonlinear and other characteristics, and the features mined by data are mostly low-interpretability or even black-box. Studying how to make full use of the respective advantages of knowledge and data and jointly drive process monitoring has become the core issue of quality-related monitoring. Finally, studying how to effectively use fault data to guide the training of the monitoring model. When the monitoring model learns the fault characteristics, each sample to be tested can not only perceive the normal sample characteristics in the monitoring model, but also judge its distance from the fault characteristics, thereby enhancing the model's ability to identify faults and improving the robustness of the model. There is a relative lack of research on the above-mentioned quality-related monitoring ideas. Therefore, real-time monitoring of the waviness of the key performance indicator of the float glass production process has not yet been achieved. In view of the above-mentioned problems, an embodiment of the present invention provides a method for monitoring the abnormality of the key performance indicators of the float glass production process. The method can be implemented by a monitoring device for the abnormality of the key performance indicators of the float glass production process. The monitoring device for the abnormality of the key performance indicators of the float glass production process can be a terminal or a server. Figure 1 The flowchart of the method for monitoring abnormal key performance indicators in the float glass production process is shown; the processing flow of the method may include the following steps:

[0074] S1. Acquire empirical knowledge and historical data of float glass production process;

[0075] S2. Filter the empirical variable set X related to waviness based on empirical knowledge k ; Based on historical data, the actual variable set X related to corrugation is screened by the maximum correlation minimum redundancy algorithm m ;

[0076] In a feasible implementation method, float glass production is a complex industrial process with a large scale and various processes. The production process is mainly divided into four parts: raw materials and ingredients, glass melting and melting furnace, glass forming and tin bath, glass annealing and annealing furnace, such as Figure 2 As shown, this example selects the process data of the three sub-processes of melting furnace, tin bath and annealing furnace and their related empirical knowledge as test data.

[0077] In a feasible implementation, the screening flow chart of the corrugation-related monitoring variables is as follows: Figure 3 As shown in S2, the empirical variable set X related to the corrugation is selected based on empirical knowledge k ; Based on historical data, the actual variable set X related to corrugation is screened by the maximum correlation minimum redundancy algorithm m ,include:

[0078] By studying the glass production process, asking on-site engineers, and consulting papers, we can obtain variables related to waviness and obtain a knowledge variable set;

[0079] The maximum correlation minimum redundancy (mRMR) algorithm is used to screen the corrugation-related variables that have the greatest correlation with the corrugation but the smallest correlation with each other; the maximum correlation D is calculated as follows:

[0080]

[0081] Among them, S represents the feature set; I(x i ; Y) represents x i Mutual information with Y.

[0082] The minimum redundancy is calculated as follows:

[0083]

[0084] Define the operator φ to combine the maximum relevance D with the minimum redundancy R:

[0085] φ(D,R)=DR (3)

[0086] Maximize the operator φ, obtain the optimal solution through the incremental search method, and screen the corrugation-related variables by maximizing φ to obtain the actual variable set X related to the corrugation based on the process data m .

[0087] S3, setting different confidence levels for the empirical knowledge and presetting a confidence threshold; obtaining the union of the waviness-related variables in the empirical knowledge whose confidence levels are higher than the confidence threshold and the results of data mining;

[0088] In a feasible implementation, the screening results of the waviness-related variables are integrated, and the flow chart is as follows: Figure 4 As shown, in S3, different confidence levels are set for the experience knowledge, and a confidence threshold is preset; the union of the waviness-related variables with confidence levels higher than the confidence threshold in the experience knowledge and the results of data mining are obtained, including:

[0089] Different confidence levels are set according to the reliability of the knowledge source; including: setting the confidence level of knowledge with high reliability from physical formulas, chemical equations, and experimental results of journal articles to 1; setting the confidence level of knowledge from engineers' production experience and intuition to 0.5-0.9 according to the engineers' recognition; when the same knowledge has multiple sources, the confidence level is averaged;

[0090] Preset the confidence threshold; extract the knowledge variable set X related to the corrugation after knowledge screening above the confidence threshold y ;

[0091] The knowledge variable set above the confidence threshold and the corrugation-related variable set X selected by maximum correlation and minimum redundancy are combined. m Take the union: X = X y ∪X m .

[0092] S4. Construct a temporal knowledge graph of the float glass production process through knowledge and data fusion;

[0093] In a feasible implementation, a time series knowledge graph is constructed by fusion of knowledge and data. The schematic diagram of the constructed time series knowledge graph is as follows: Figure 5 As shown, in S4, a temporal knowledge graph of the float glass production process is constructed, including:

[0094] Define entities of the temporal knowledge graph of the float glass production process; the entities include waviness-related variables; define the relationships between entities based on expert knowledge and historical data;

[0095] Based on the relationship between entities, the following temporal knowledge graph is constructed:

[0096] TKG=<E,R,S,T>

[0097] Among them, E is the set of entities; R is the set of relations; S is the set of all static knowledge triples; T is the set of all connected time event quads;

[0098] Among them, the knowledge graph can be represented by several knowledge triples (head (H-node), relationship (r), tail (T-node)). However, knowledge is not static, but changes dynamically over time, so the knowledge triple needs to be added with the time to form a quadruple (head (H-node), relationship (r), tail (T-node), time (Time)) to represent the knowledge and its effective time. A temporal knowledge graph composed of knowledge quadruple is obtained.

[0099] In order to extract graph features conveniently, the temporal knowledge graph is sliced ​​according to time, and the quadruple is converted into a triple (es, t, r, eo, t), 0<t≤T; where e s,tis the head entity at time t, r is the association relationship, e o,t is the tail entity corresponding to time t.

[0100] In a feasible implementation, the relationship between entities is defined based on expert knowledge and historical data, including:

[0101] When entities are obtained from text, equations, or engineering experience, the relationships between entities can be directly defined;

[0102] When entities can be determined to have interactions based on expert experience knowledge, but their size cannot be quantified or it is impossible to determine whether there is an interactive relationship, the distance correlation coefficient considering time lag is used to calculate the relationship between entities;

[0103] The calculation formula is as follows:

[0104]

[0105] Among them, τ is the time lag, and u and v are two entities. In order to avoid being affected by the results of data mining with lower credibility, a threshold is set to finally determine the association relationship between entities:

[0106]

[0107] Among them, Thr is the correlation threshold, which is defined by expert experience knowledge.

[0108] In a feasible implementation manner, the distance correlation coefficient includes:

[0109] The distance correlation coefficient overcomes the shortcomings of the Pearson correlation coefficient. The distance correlation coefficient of two entities u and v is denoted as DCOR(u,v). The two entities u and v are independent of each other if and only if DCOR(u,v)=0; the stronger the correlation between the two entities u and v, the greater the DCOR(u,v).

[0110] The distance correlation coefficient calculation formula is as follows:

[0111]

[0112] in, They are:

[0113]

[0114] S5. Construct a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the temporal knowledge graph and construct monitoring statistics.

[0115] In a feasible implementation method, the monitoring model diagram is as follows: Figure 6As shown, including:

[0116] Monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning;

[0117] Through the monitoring model's differentiable pooling-long short-term memory network, the multi-level spatiotemporal features of the temporal knowledge graph are extracted;

[0118] Incorporate fault data when constructing monitoring statistics through supervised comparative learning of monitoring models;

[0119] Calculation of control limits for waviness monitoring by kernel density estimation.

[0120] In a feasible implementation method, the schematic diagram of the multi-level feature extraction model is as follows: Figure 7 As shown, including:

[0121] Spatial features are extracted by aggregating neighbor information through GCN, and hierarchical information is extracted by learning the allocation matrix through differentiable pooling. The float glass production process is divided into four levels: sensor level, equipment level, sub-process level, and process level. The nodes of the lower level are allocated to the higher level, and the node features of the lth level and the allocation matrix of the lth level are calculated.

[0122] Based on the node embedding and adjacency matrix of the node embedding layer l+1, the association graph of the l+1 layer is obtained, and the allocation matrix and association graph of each level are repeatedly calculated to obtain spatial features of multiple levels;

[0123] The node features of the lth layer and the allocation matrix of this layer can be calculated using the following formula:

[0124]

[0125] Among them, GCN l,emb and GCN l,pool The model structure is the same but the model parameters are different. F (l) is the adjacency matrix of the lth level, X (l) is the node embedding of the lth layer. (l) and node feature Z (l) The nodes in the lth layer can be assigned to a higher layer (the lth + 1th layer). The nodes in the lth + 1th layer are embedded in X (l+1) and the adjacency matrix A F (l+1) The calculation formula is as follows:

[0126]

[0127] Get node embedding X (l+1) and the adjacency matrix A F (l+1)The association graph of the l+1th layer is obtained. At this time, the above steps can be repeated to extract spatial features of multiple levels.

[0128] In a feasible implementation, the current state is linked to the long-term historical data through a long short-term memory network to extract the long-term dependencies in the multi-level spatial features; the long short-term memory network is mainly composed of three gating mechanisms: the forget gate F t : Control whether to forget the hidden cell state h of the previous layer t-1 , the formula is as follows:

[0129] F t =σ(W f ·[h t-1 ,x t ]+b f ) (10)

[0130] Among them, x t is the input, W f is the weight matrix of the forget gate, b f is the bias and σ is the activation function.

[0131] The input gate is used to control the information to be input and update the cell state in combination with the output of the forget gate. The formula is as follows:

[0132]

[0133] Among them, W i , W C is the weight matrix, b i 、b C is the deviation, C t′ is the candidate cell state, σ and tanh are activation functions, C t is the updated cell state.

[0134] The output gate can get the hidden state, and the formula is as follows:

[0135]

[0136] Among them, W o is the weight matrix, b o It's a deviation.

[0137] LSTM networks extract long-term associations in time series data by continuously passing updated cell states and hidden states to the next time series.

[0138] In a feasible implementation, GCN extracts spatial features by aggregating neighbor information: the association graph converted from the temporal knowledge graph is defined as follows: G = (X, A, F). Where X is the node feature, A is the adjacency matrix, and F = ruv is the connection strength. The elements in the adjacency matrix A are only 0 and 1, and the formula is as follows:

[0139]

[0140] Among them, r uv It is the quantified size of the relationship between two entities.

[0141] The aggregation operation on the association graph aims to aggregate the information of each node with its neighboring nodes to obtain a richer representation, which can be expressed as follows:

[0142]

[0143] Among them, AGG n is the graph feature after aggregating neighbor information, k is the current layer number, n is the node in the graph, N(n) is the neighbor node of node n, Θ is the parameter to be learned, is the aggregated information of the neighboring node u in the (k-1)th layer. The update operation is to fuse the aggregated feature representation with the original feature of the current node to obtain the updated feature representation of the node in the next layer. The calculation formula is:

[0144]

[0145] Among them, f(·) is the activation function, W is the parameter to be learned, and x n is the current feature vector of node n. is the new feature vector of node n in the kth layer.

[0146] In a feasible implementation, fault information is incorporated into statistics through supervised contrast learning; and control limits for waviness monitoring are calculated through kernel density estimation, including:

[0147] Obtain the linear model parameter matrix and multi-level spatiotemporal features, and calculate the control limits through kernel density estimation;

[0148] Construct positive and negative samples corresponding to the features to be predicted, where the positive samples are the features of the real samples at the next τ moments, and the negative samples are the features of the known fault samples in the historical data;

[0149] Obtain a fault sample data set and construct statistics by combining control limits and positive and negative samples;

[0150] The control limit of the statistic is calculated by kernel density estimation, wherein the kernel density estimation models the data distribution of the statistic when the corrugation meets the production requirements; the probability density function of the statistic is constructed, the probability density function of the statistic is integrated to obtain the cumulative distribution function, and the statistic with a cumulative distribution function value of 95% is selected as the control limit of the corrugation monitoring.

[0151] In a feasible implementation, the extracted multi-level spatiotemporal features are used to construct a prediction agent task of the monitoring model, and a linear model is used to predict the features of the next τ moments.

[0152]

[0153] Among them, W τ is the linear model parameter matrix, C t It is the multi-level spatiotemporal feature described in S5.

[0154] For the predicted features, construct the corresponding positive and negative samples. The positive samples are the features of the real samples at the next τ moments. The negative samples are the features of the known fault samples in the historical data. It is hoped that the predicted features are close enough to the positive samples and far enough from the fault samples, so as to construct the statistics as follows:

[0155]

[0156] in Z t+τ is a positive sample, F is a fault sample set, and n is the number of fault samples.

[0157] Calculate statistics via kernel density estimation (KDE) The control limit of KDE modeling is the statistic when the corrugation meets the production requirements. The data distribution F(Δt) is selected, and the control limit is selected as the 95% probability density percentile. Integrating the probability density function F(Δt) gives the cumulative distribution function (CDF):

[0158]

[0159] Find the 95% mode statistic for the value of the cumulative distribution function This is the control limit for waviness monitoring.

[0160] S6. Abnormal corrugation monitoring: obtain real-time data and input the real-time data into the monitoring model. If the statistic exceeds the control limit, it means that a corrugation-related fault has occurred in the float glass production process.

[0161] In a feasible implementation, in step S6, the waviness abnormality is monitored, real-time data is obtained, and the data is input into the monitoring model. If the statistic exceeds the control limit, it means that a waviness-related fault has occurred in the float glass production process, including:

[0162] Real-time data is acquired and input into the monitoring model. The statistics of the real-time data are output through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred; otherwise, no corrugation-related fault has occurred.

[0163] In a feasible implementation, the algorithm is not only applicable to abnormal monitoring of waviness in the float glass production process, but also to abnormal monitoring of other key performance indicators, such as thickness, bubbles, etc.

[0164] In the embodiment of the present invention, the change of the corrugation of the float glass production process is monitored in real time through the monitoring model. When the corrugation is abnormal, an alarm is given in time to ensure production safety and improve production efficiency. Based on the time series knowledge graph, the certainty of knowledge and the uncertainty of data are effectively integrated to improve the interpretability and monitoring performance of the model. Based on supervised comparative learning, the monitoring model has good robustness, the model's ability to identify faults is improved, the credibility of the monitoring results is increased, and the product corrugation quality and production efficiency can be greatly improved.

[0165] Figure 8 The present invention is a block diagram of a device for monitoring abnormal key performance indicators in a float glass production process according to an exemplary embodiment. The device is used in a method for monitoring abnormal key performance indicators in a float glass production process. Figure 8 The device includes a data acquisition module 310, a variable screening module 320, a knowledge and data fusion module 330, a time series knowledge graph module 340, a monitoring model construction module 350, and an abnormality monitoring module 360. Among them:

[0166] Data acquisition module 310, used to obtain empirical knowledge and historical data of the float glass production process;

[0167] The variable screening module 320 is used to screen the empirical variable set related to the waviness according to the empirical knowledge; and screen the actual variable set related to the waviness by the maximum correlation minimum redundancy algorithm according to the historical data;

[0168] The knowledge and data fusion module 330 is used to set different confidence levels for the empirical knowledge and preset a confidence threshold; obtain the union of the waviness-related variables with confidence levels higher than the confidence threshold in the empirical knowledge and the results of data mining;

[0169] A time series knowledge graph module 340 is used to construct a time series knowledge graph of the float glass production process by fusion of knowledge and data;

[0170] A monitoring model building module 350 is used to build a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the time series knowledge graph and build monitoring statistics;

[0171] The abnormality monitoring module 360 ​​is used for monitoring the abnormality of corrugation, obtaining real-time data, inputting the real-time data into the monitoring model, and outputting the statistics of the real-time data through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred; otherwise, no corrugation-related fault has occurred.

[0172] Optionally, a variable screening module 320 is used to obtain waviness-related variables by learning glass production processes, asking field engineers, and searching papers to obtain a knowledge variable set;

[0173] The maximum correlation and minimum redundancy (mRMR) algorithm is used to select the corrugation-related variables that have the greatest correlation with the corrugation but the smallest correlation with each other; the calculation results of the maximum correlation and the calculation results of the minimum redundancy are combined;

[0174] The optimal solution is obtained through the incremental search method, and the corrugation-related variables are screened by maximizing the combined results to obtain the actual variable set related to the corrugation based on the process data.

[0175] Optionally, the knowledge and data fusion module 330 is used to set different confidence levels according to the reliability of the knowledge source; including: setting the confidence level of knowledge with high reliability from physical formulas, chemical equations, and experimental results of journal articles to 1; setting the confidence level of knowledge from engineers' production experience and intuition to 0.5 to 0.9 according to the engineers' recognition; when the same knowledge has multiple sources, the confidence level is averaged;

[0176] Presetting a confidence threshold; extracting a knowledge variable set related to the waviness after knowledge screening above the confidence threshold;

[0177] The knowledge variable set above the confidence threshold and the waviness-related variable set selected by maximum correlation and minimum redundancy are combined.

[0178] Optionally, the temporal knowledge graph module 340 is used to define entities of the temporal knowledge graph of the float glass production process; wherein the entities include waviness-related variables; and the relationships between the entities are jointly defined based on expert knowledge and historical data;

[0179] Based on the relationship between entities, the following temporal knowledge graph is constructed:

[0180] TKG=<E,R,S,T>

[0181] Among them, E is the set of entities; R is the set of relations; S is the set of all static knowledge triples; T is the set of all connected time event quads;

[0182] Slice the temporal knowledge graph according to time and transform the quadruple into triple (es, t, r, eo, t), 0<t≤T; where es,t is the head entity at time t, r is the association relationship, e o,t is the tail entity corresponding to time t.

[0183] Optionally, the relationships between entities are jointly defined based on expert knowledge and historical data, including:

[0184] When entities are obtained from text, equations, or engineering experience, the relationships between entities can be directly defined;

[0185] When entities can be determined to have interactions based on expert experience knowledge, but their size cannot be quantified or it is impossible to determine whether there is an interactive relationship, the distance correlation coefficient that takes time lag into account is used to calculate the relationship between the entities.

[0186] Optionally, a distance correlation coefficient, including:

[0187] The distance correlation coefficient between two entities u and v is denoted as DCOR(u,v). The two entities u and v are independent of each other if and only if DCOR(u,v)=0; the stronger the correlation between the two entities u and v, the larger the DCOR(u,v).

[0188] Optionally, a monitoring model building module 350 is used to extract multi-level spatiotemporal features of the temporal knowledge graph through a differentiable pooling-long short-term memory network;

[0189] Incorporate failure data when constructing monitoring statistics through supervised contrastive learning;

[0190] Calculate control limits for waviness monitoring by kernel density estimation;

[0191] Optionally, multi-level spatiotemporal feature extraction, including:

[0192] It is used to extract spatial features by aggregating neighbor information through GCN, and extract hierarchical information by learning the allocation matrix through differentiable pooling; divide the float glass production process into four levels: sensor level, equipment level, sub-process level and process level, allocate low-level nodes to high-level levels, and calculate the node features of the lth layer and the allocation matrix of the lth layer;

[0193] Based on the node embedding and adjacency matrix of the node embedding layer l+1, the association graph of the l+1 layer is obtained, and the allocation matrix and association graph of each level are repeatedly calculated to obtain spatial features of multiple levels;

[0194] Through the long short-term memory network, the current state is connected with the long-term historical data to extract the long-term dependencies in the multi-level spatial features; the updated cell state and hidden state are repeatedly passed through the output gate of the long short-term memory network combined with the forgetting gate output, and passed to the next time series to extract the long-term associations in the time series data.

[0195] Optionally, statistics are constructed by supervised contrastive learning, and fault information is incorporated into the statistics; control limits for waviness monitoring are calculated by kernel density estimation, including;

[0196] Construct positive and negative samples corresponding to the features to be predicted, where the positive samples are the features of the real samples at the next τ moments, and the negative samples are the features of the known fault samples in the historical data;

[0197] Obtain a fault sample data set and construct statistics by combining control limits and positive and negative samples;

[0198] The control limit of the statistic is calculated by kernel density estimation, wherein the kernel density estimation models the data distribution of the statistic when the corrugation meets the production requirements; the probability density function of the statistic is constructed, the probability density function of the statistic is integrated to obtain the cumulative distribution function, and the statistic with a cumulative distribution function value of 95% is selected as the control limit of the corrugation monitoring.

[0199] In the embodiment of the present invention, the change of the corrugation of the float glass production process is monitored in real time through the monitoring model. When the corrugation is abnormal, an alarm is given in time to ensure production safety and improve production efficiency. Based on the time series knowledge graph, the certainty of knowledge and the uncertainty of data are effectively integrated to improve the interpretability and monitoring performance of the model. Based on supervised comparative learning, the monitoring model has good robustness, the model's ability to identify faults is improved, the credibility of the monitoring results is increased, and the product corrugation quality and production efficiency can be greatly improved.

[0200] Fig. 9 is a schematic diagram of the structure of a device for monitoring abnormal key performance indicators in a float glass production process provided by an embodiment of the present invention. Fig. 9 As shown, the abnormal monitoring equipment for key performance indicators of the float glass production process may include the above Figure 8 The device for monitoring abnormal key performance indicators in the production process of float glass as shown. Optionally, the device for monitoring abnormal key performance indicators in the production process of float glass 410 may include a first processor 2001 .

[0201] Optionally, the float glass production process key performance indicator abnormality monitoring device 410 may further include a memory 2002 and a transceiver 2003 .

[0202] The first processor 2001, the memory 2002 and the transceiver 2003 may be connected via a communication bus.

[0203] Combine the following Fig. 9 The various components of the float glass production process key performance indicator abnormality monitoring device 410 are specifically introduced:

[0204] The first processor 2001 is the control center of the float glass production process key performance indicator abnormality monitoring device 410, which can be a processor or a general term for multiple processing elements. For example, the first processor 2001 is one or more central processing units (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiment of the present invention, such as one or more microprocessors (DSP), or one or more field programmable gate arrays (FPGA).

[0205] Optionally, the first processor 2001 can execute various functions of the float glass production process key performance indicator abnormality monitoring device 410 by running or executing a software program stored in the memory 2002 and calling data stored in the memory 2002 .

[0206] In a specific implementation, as an embodiment, the first processor 2001 may include one or more CPUs, such as Fig. 9 CPU0 and CPU1 are shown in FIG.

[0207] In a specific implementation, as an embodiment, the float glass production process key performance indicator abnormality monitoring device 410 may also include multiple processors, such as Fig. 9 The first processor 2001 and the second processor 2004 are shown in FIG. Each of these processors can be a single-core processor (single-CPU) or a multi-core processor (multi-CPU). The processor here can refer to one or more devices, circuits, and / or processing cores for processing data (such as computer program instructions).

[0208] The memory 2002 is used to store the software program for executing the solution of the present invention, and is controlled to be executed by the first processor 2001. The specific implementation method can refer to the above method embodiment, which will not be repeated here.

[0209] Optionally, the memory 2002 may be a read-only memory (ROM) or other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 2002 may be integrated with the first processor 2001, or may exist independently, and access the memory 2002 through the interface circuit ( Fig. 9 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0210] The transceiver 2003 is used to communicate with a network device or a terminal device.

[0211] Optionally, the transceiver 2003 may include a receiver and a transmitter ( Fig. 9 The receiver is used to implement a receiving function, and the transmitter is used to implement a sending function.

[0212] Optionally, the transceiver 2003 may be integrated with the first processor 2001, or may exist independently, and may be connected to the first processor 2001 through the interface circuit ( Fig. 9 (not shown) is coupled to the first processor 2001, which is not specifically limited in this embodiment of the present invention.

[0213] It should be noted that Fig. 9 The structure of the float glass production process key performance indicator abnormality monitoring device 410 shown in the figure does not constitute a limitation on the router, and the actual knowledge structure recognition device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0214] In addition, the technical effects of the float glass production process key performance indicator abnormality monitoring device 410 can refer to the technical effects of the float glass production process key performance indicator abnormality monitoring method described in the above method embodiment, which will not be repeated here.

[0215] It should be understood that the first processor 2001 in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0216] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0217] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0218] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0219] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0220] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0221] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0222] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0223] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0224] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0225] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0226] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0227] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for monitoring abnormal key performance indicators in a float glass production process, characterized in that: The method comprises: S1. Acquire empirical knowledge and historical data of float glass production process; S2, screening the empirical variable set related to the waviness according to the empirical knowledge; screening the actual variable set related to the waviness by the maximum correlation minimum redundancy algorithm according to the historical data; S3, setting different confidence levels for the empirical knowledge and presetting a confidence threshold; obtaining a union of the waviness-related variables in the empirical knowledge whose confidence levels are higher than the confidence threshold and the results of data mining; S4. Construct a temporal knowledge graph of the float glass production process through knowledge and data fusion; S5. Construct a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the time series knowledge graph and construct monitoring statistics; In S5, based on the monitoring model of differentiable pooling-long short-term memory network and supervised contrastive learning, multi-level spatiotemporal features in the time series knowledge graph are extracted and monitoring statistics are constructed, including: Monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; Through the differentiable pooling-long short-term memory network of the monitoring model, multi-level spatiotemporal features of the temporal knowledge graph are extracted; Incorporating fault data into the construction of monitoring statistics through supervised comparative learning of the monitoring model; Calculate control limits for waviness monitoring by kernel density estimation; The step of extracting multi-level spatiotemporal features from the temporal knowledge graph includes: Spatial features are extracted by aggregating neighbor information through GCN, and hierarchical information is extracted by learning the allocation matrix through differentiable pooling. The float glass production process is divided into four levels from low to high: sensor level, equipment level, sub-process level, and process level. The nodes of the lower level are allocated to the higher level, and the node features of the lth level and the allocation matrix of the lth level are calculated. Based on the node embedding and adjacency matrix of the node embedding layer l+1, the association graph of the l+1 layer is obtained, and the allocation matrix and association graph of each level are repeatedly calculated to obtain spatial features of multiple levels; The current state is connected with the long-term historical data through the long short-term memory network to extract the long-term dependency in the multi-level spatial features; the cell state and hidden state updated by the output gate of the long short-term memory network combined with the forget gate output are repeatedly passed to the next time series to extract the long-term association in the time series data; S6. Monitoring of abnormal corrugation, obtaining real-time data, inputting the real-time data into a monitoring model, and outputting statistics of the real-time data through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred; otherwise, no corrugation-related fault has occurred.

2. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 1, characterized in that: In S2, the empirical variable set related to the waviness is selected according to the empirical knowledge; and the actual variable set related to the waviness is selected by the maximum correlation minimum redundancy algorithm according to the historical data, including: By studying the glass production process, asking on-site engineers, and consulting papers, we can obtain variables related to waviness and obtain a knowledge variable set; The maximum correlation and minimum redundancy (mRMR) algorithm is used to select the corrugation-related variables that have the greatest correlation with the corrugation but the smallest correlation with each other; the calculation results of the maximum correlation and the calculation results of the minimum redundancy are combined; The optimal solution is obtained through the incremental search method, and the corrugation-related variables are screened by maximizing the combined results to obtain the actual variable set related to the corrugation based on the process data.

3. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 2, characterized in that: In S3, different confidence levels are set for the empirical knowledge, and confidence thresholds are preset; Obtaining a union of the waviness-related variables with a confidence level higher than the confidence level threshold in the empirical knowledge and the results of data mining, including: Different confidence levels are set according to the reliability of the knowledge source; including: setting the confidence level of knowledge with high reliability from physical formulas, chemical equations, and experimental results of journal articles to 1; setting the confidence level of knowledge from engineers' production experience and intuition to 0.5-0.9 according to the engineers' recognition; when the same knowledge has multiple sources, the confidence level is averaged; Preset a confidence threshold; extract the knowledge variable set X related to the corrugation after knowledge screening above the confidence threshold y ; The knowledge variable set above the confidence threshold and the waviness-related variable set screened by maximum correlation and minimum redundancy are combined.

4. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 3, characterized in that: In S4, a time-series knowledge graph of the float glass production process is constructed, including: Defining entities of a temporal knowledge graph of a float glass production process; wherein the entities include waviness-related variables; and jointly defining relationships between the entities based on expert knowledge and historical data; Based on the relationship between the entities, the following temporal knowledge graph is constructed: TKG=<E,R,S,T> Among them, E is the set of entities; R is the set of relations; S is the set of all static knowledge triples; T is the set of all connected time event quads; Slice the temporal knowledge graph according to time and transform the quadruple into triple (es, t, r, eo, t), 0<t≤T; where e s,t is the head entity at time t, r is the association relationship, e o,t is the tail entity corresponding to time t.

5. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 4, characterized in that: Defining the relationship between the entities based on expert knowledge and historical data includes: When the entities are obtained from text, equations, or engineering experience, the association relationship between the entities can be directly defined; When it is determined from expert experience that the entities have an interaction, but the size of the interaction cannot be quantified or it cannot be determined whether there is an interaction relationship, the distance correlation coefficient considering time lag is used to calculate the relationship between the entities.

6. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 5, characterized in that: The distance correlation coefficient includes: The distance correlation coefficient between two entities u and v is denoted as DCOR(u,v). The two entities u and v are independent of each other if and only if DCOR(u,v)=0; the stronger the correlation between the two entities u and v, the larger the DCOR(u,v).

7. The method for monitoring abnormal key performance indicators in the float glass production process according to claim 6, characterized in that: Through supervised contrast learning, fault data is included in the construction of monitoring statistics; the control limits of waviness monitoring are calculated through kernel density estimation, including: Obtain the linear model parameter matrix and multi-level spatiotemporal features, and calculate the control limits through kernel density estimation; Construct positive and negative samples corresponding to the features to be predicted, where the positive samples are the features of the real samples at the next τ moments, and the negative samples are the features of the known fault samples in the historical data; Obtain a fault sample data set and construct statistics by combining control limits and positive and negative samples; The control limit of the statistic is calculated by kernel density estimation, wherein the kernel density estimation models the data distribution of the statistic when the corrugation meets the production requirements; a probability density function of the statistic is constructed, the probability density function of the statistic is integrated to obtain the cumulative distribution function, and the statistic whose value of the cumulative distribution function is 95% is selected as the control limit of the corrugation monitoring.

8. A device for monitoring abnormalities of key performance indicators in a float glass production process, wherein the device is used to implement the method for monitoring abnormalities of key performance indicators in a float glass production process according to any one of claims 1 to 7, characterized in that: The device comprises: Data acquisition module, used to obtain empirical knowledge and historical data of the float glass production process; A variable screening module is used to screen the empirical variable set related to the waviness according to the empirical knowledge; and screen the actual variable set related to the waviness by the maximum correlation minimum redundancy algorithm according to the historical data; The knowledge and data fusion module is used to set different confidence levels for the empirical knowledge and preset a confidence threshold; obtain the union of the waviness-related variables in the empirical knowledge whose confidence level is higher than the confidence threshold and the results of data mining; The time series knowledge graph module is used to construct the time series knowledge graph of the float glass production process through knowledge and data fusion; Monitoring model building module, used to build a monitoring model based on differentiable pooling-long short-term memory network and supervised contrastive learning; extract multi-level spatiotemporal features in the time series knowledge graph and construct monitoring statistics; The abnormality monitoring module is used for monitoring the abnormality of corrugation, acquiring real-time data, inputting the real-time data into the monitoring model, and outputting the statistics of the real-time data through the monitoring model. If the statistics are greater than the control limit, it is considered that a corrugation-related fault has occurred; otherwise, no corrugation-related fault has occurred.

Images