Industrial big data simulation method and system based on digital twin

Abstract

The present application provides an industrial big data simulation method and system based on digital twins. In the network debugging link, the adaptive integration of the estimated costs of each state of the multi-task state reasoning component in the debugging link is completed based on the cost integration component. At the same time, according to the standardized operation of the influence coefficient variable in the cost integration component of the equipment state classification network, a target equipment state classification network including a measurement integration component is obtained, and the prediction and debugging costs are separated. The adaptive coordination and integration of the prediction and debugging costs are completed respectively. It is effectively applied to the imbalance effect situation when identifying abnormal equipment states, so that the target equipment state classification network not only has high-quality recognition accuracy for state classifications with more debugging samples (such as high-frequency equipment abnormal state classification), but also has high-quality recognition accuracy for state classifications with fewer debugging samples (such as low-frequency equipment abnormal state classification).

Description

Technical Field

[0001] The present application relates to but is not limited to the fields of machine learning and data processing technology, and in particular to an industrial big data simulation method and system based on digital twins. Background Art

[0002] With the rapid development of the industrial Internet, more and more industrial equipment and systems are connected to the Internet, generating a large amount of industrial data. These data contain rich information and can be used to optimize production processes, improve equipment efficiency, predict faults, etc. However, due to the complexity and diversity of industrial data, traditional data analysis methods often fail to meet the needs of the industrial field. To solve this problem, digital twin technology has emerged. Digital twin is a digital model based on physical entities that can reflect the state and behavior of physical entities in real time. By applying digital twin technology to industrial big data, real-time monitoring and optimization of industrial systems can be achieved. However, existing methods still have some problems and challenges. For example, before synchronizing digital twins, artificial intelligence models can be used to identify the operating status of industrial equipment. In the model training phase, the sample data obtained often has a small number of samples occupying most of the data, while most samples only occupy a small amount of data. This distribution will cause the neural network model to over-focus on a small number of samples during training and ignore the information of most samples, resulting in a decrease in the accuracy of the model. Summary of the invention

[0003] In view of this, the embodiments of the present application at least provide an industrial big data simulation method and system based on digital twins.

[0004] The technical solution of the embodiment of the present application is implemented as follows:

[0005] On the one hand, an embodiment of the present application provides an industrial big data simulation method based on digital twins, the method comprising:

[0006] Obtaining industrial large data sets to be analyzed;

[0007] According to the target state classification representation vector of each state classification in the set number of state classifications, a multi-task state recognition component based on the target device state classification network performs device state reasoning on the industrial big data set to be analyzed to obtain first state reasoning information and second state reasoning information;

[0008] The first state reasoning information indicates a commonality coefficient between the industrial big data set to be analyzed and each of the state classifications, and the second state reasoning information indicates a confidence level of the industrial big data set to be analyzed corresponding to each of the state classifications;

[0009] Performing a metric integration operation on the first state reasoning information and the second state reasoning information according to the metric integration component of the target device state classification network to obtain integrated state reasoning information;

[0010] Determining target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information;

[0011] Performing corresponding state display in the industrial digital twin model based on the target state reasoning information;

[0012] The target state classification representation vector of each state classification is determined according to the target training representation vector corresponding to the target industrial big data debugging sample of each state classification in the target industrial big data debugging sample set, and the target training representation vector is obtained by extracting the representation vector of the corresponding target industrial big data debugging sample according to the target device state classification network; wherein, the target device state classification network is debugged according to the following steps:

[0013] Obtain an industrial big data debugging sample set and an industrial big data verification sample set; the industrial big data debugging sample set includes industrial big data debugging samples of a set number of status classifications, and the status classification of the industrial big data verification samples in the industrial big data verification sample set is the same as that of the industrial big data debugging sample set;

[0014] The industrial big data debugging sample set and the industrial big data verification sample set are loaded into the multi-task state recognition component of the state classification network of the device to be debugged to perform device state reasoning, and the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample are obtained; the first state reasoning information indicates the commonality coefficient between the corresponding industrial big data verification sample and each of the state classifications, and the second state reasoning information indicates the confidence of the corresponding industrial big data verification sample corresponding to each of the state classifications;

[0015] Determine a first state estimated cost and a second state estimated cost according to errors between the first state reasoning information and the second state reasoning information corresponding to each of the industrial big data verification samples and the state classification of the industrial big data verification samples;

[0016] Loading the first state estimated cost and the second state estimated cost into the cost integration component of the state classification network of the device to be debugged to perform a cost integration operation to obtain a first target state estimated cost;

[0017] Debugging the device state classification network to be debugged according to the first target state estimated cost until it meets the set debugging cutoff requirement, thereby obtaining a device state classification network;

[0018] A standardization operation is performed on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component to obtain a target device state classification network.

[0019] In some embodiments, the influence coefficient variable of the cost integration component includes a first cost influence coefficient and a second cost influence coefficient; the step of loading the first state estimated cost and the first state estimated cost into the cost integration component of the state classification network of the device to be debugged to perform a cost integration operation to obtain the first target state estimated cost includes:

[0020] Loading the first state estimated cost into the cost integration component of the state classification network of the device to be debugged, outputting and calculating the first state estimated cost according to the first cost influence coefficient and the corresponding first cost offset, to obtain the first target sub-state estimated cost; the first cost influence coefficient is negatively correlated with the first cost offset;

[0021] The second state estimated cost is loaded into the cost integration component of the state classification network of the device to be debugged, and the second state estimated cost is output and calculated according to the second cost influence coefficient and the corresponding second cost offset to obtain the second target sub-state estimated cost; the second cost influence coefficient is negatively correlated with the second cost offset;

[0022] The first target sub-state estimated cost and the second target sub-state estimated cost are added together to obtain a first target state estimated cost.

[0023] In some embodiments, the method further comprises:

[0024] Determine a first debugging cost and a second debugging cost according to an error between the first state reasoning information and the second state reasoning information of each industrial big data verification sample; the first debugging cost is used to make the first state reasoning information close to the second state reasoning information, and the second debugging cost is used to make the second state reasoning information close to the first state reasoning information;

[0025] Adding the first debugging cost and the second debugging cost to obtain a second target state estimated cost;

[0026] The step of debugging the device state classification network to be debugged according to the first target state estimated cost until the debugging cutoff requirement is met to obtain the device state classification network includes:

[0027] Determining a total state estimated cost according to the first target state estimated cost and the second target state estimated cost;

[0028] The device state classification network to be debugged is debugged according to the total state estimated cost until it meets the set debugging cutoff requirement, thereby obtaining the device state classification network.

[0029] In some embodiments, the multi-task state recognition component includes a representation vector extraction component and a multi-task state reasoning component, and the multi-task state reasoning component includes a state reasoning evaluation component and a normalized mapping reasoning component; the multi-task state recognition component that loads the industrial big data debugging sample set and the industrial big data verification sample set into the state classification network of the device to be debugged performs device state reasoning, and obtains the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample, including:

[0030] Loading the industrial big data debugging sample set and the industrial big data verification sample set into the representation vector extraction component to extract the representation vector, and obtaining the training representation vector of each industrial big data debugging sample and the verification representation vector of each industrial big data verification sample;

[0031] Determine a state classification representation vector for each state classification according to the training representation vectors of each industrial big data debugging sample corresponding to each state classification;

[0032] Determine the commonality coefficient between each verification characterization vector and each of the state classification characterization vectors according to the state reasoning evaluation component, and obtain the first state reasoning information of each of the industrial big data verification samples;

[0033] Determine the confidence of each verification representation vector corresponding to each state classification according to the normalized mapping reasoning component, and obtain the second state reasoning information of each industrial big data verification sample;

[0034] The step of determining the state classification representation vector of each state classification according to the training representation vector of each industrial big data debugging sample corresponding to each state classification includes:

[0035] Averaging the training representation vectors of the industrial big data debugging samples corresponding to each state classification in the industrial big data debugging sample set to obtain an initial state centroid representation vector of each state classification;

[0036] Determine the similarity measure between each training representation vector corresponding to each state classification and the initial state centroid representation vector of the state classification, and obtain the influence coefficient of each training representation vector corresponding to each state classification;

[0037] Performing influence adjustment calculation on each training representation vector of the state classification according to the influence coefficient of each training representation vector corresponding to each state classification and completing the average to obtain the optimized state centroid representation vector of each state classification;

[0038] Performing a same-state grouping operation on each training representation vector corresponding to each state classification to obtain a same-state centroid representation vector of each state classification;

[0039] An integration operation is performed on the optimized state centroid representation vector of each state classification and the same-state centroid representation vector of the state classification to obtain a state classification representation vector of each state classification.

[0040] In some embodiments, obtaining the industrial big data debugging sample set and the industrial big data verification sample set includes:

[0041] Acquire an industrial big data database; the industrial big data database includes the industrial big data set samples of the set number of state classifications;

[0042] The industrial big data set is divided into a support data set and a development data set; the support data set and the development data set both have the industrial big data set samples of the set number of state classifications;

[0043] Performing multiple debugging data sampling on the support data set and the development data set, each time arbitrarily sampling a first set number of industrial big data set samples in each state classification of the support data set to obtain an industrial big data debugging sample set of corresponding debugging data, and arbitrarily sampling a second set number of industrial big data set samples in each state classification of the development data set to obtain an industrial big data verification sample set of the debugging data;

[0044] According to any one of the multiple debugging data, the industrial big data debugging sample set and the industrial big data verification sample set are obtained.

[0045] In some embodiments, the step of performing a normalization operation on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component to obtain the target device state classification network includes:

[0046] Acquire a device status classification network obtained by debugging respectively according to a plurality of the debugging data;

[0047] Determining a state reasoning effect parameter of each of the device state classification networks based on the development data set;

[0048] According to the state reasoning effect parameter of each of the device state classification networks, a candidate device state classification network is determined from the plurality of the device state classification networks; the state reasoning effect parameter of the candidate device state classification network is better than the state reasoning effect parameter of any of the remaining device state classification networks;

[0049] A standardization operation is performed on the influence coefficient variable of the cost integration component in the candidate device state classification network to transform the cost integration component in the candidate device state classification network into a metric integration component to obtain a target device state classification network.

[0050] In some embodiments, the multi-task state recognition component based on the target device state classification network performs device state reasoning on the industrial big data set to be analyzed based on the target state classification representation vector of each state classification in the set number of state classifications to obtain first state reasoning information and second state reasoning information, including:

[0051] Extracting a representation vector of the industrial big data set to be analyzed according to the representation vector extraction component of the multi-task state recognition component in the target device state classification network to obtain a representation vector of the industrial big data set to be analyzed;

[0052] Determine the commonality coefficient between the characterization vector of the industrial big data set to be analyzed and each of the target state classification characterization vectors according to the state reasoning evaluation component of the multi-task state recognition component in the target device state classification network, and obtain first state reasoning information;

[0053] According to the normalized mapping reasoning component of the multi-task state recognition component in the target device state classification network, the confidence of the characterization vector of the industrial big data set to be analyzed corresponding to each of the state classifications is determined to obtain second state reasoning information.

[0054] In some embodiments, the target industrial big data debugging sample set is obtained based on a supporting data set including the industrial big data set samples of the set number of state classifications; the metric integration component based on the target device state classification network performs a metric integration operation on the first state reasoning information and the second state reasoning information, and obtaining the integrated state reasoning information includes:

[0055] According to the number of industrial big data set samples corresponding to each state classification in the support data set, determine the first state reasoning information influence coefficient and the second state reasoning information influence coefficient; the first state reasoning information influence coefficient is negatively correlated with the number of industrial big data set samples corresponding to each state classification, and the second state reasoning information influence coefficient is positively correlated with the number of industrial big data set samples corresponding to each state classification, and the sum of the first state reasoning information influence coefficient and the second state reasoning information influence coefficient is the set value;

[0056] multiplying the first state reasoning information influence coefficient by the first state reasoning information to obtain target first state reasoning information;

[0057] and multiplying the second state reasoning information influence coefficient by the second state reasoning information to obtain target second state reasoning information;

[0058] The target first state reasoning information and the target second state reasoning information are loaded into the metric integration component of the target device state classification network to perform a metric integration operation to obtain integrated state reasoning information.

[0059] In some embodiments, obtaining the industrial big data set to be analyzed includes:

[0060] Acquire an initial industrial big data set, perform region of interest recognition processing on the initial industrial big data set, and obtain a region of interest marking window and region of interest recognition state classification information corresponding to the region of interest marking window;

[0061] Acquire the industrial data cluster corresponding to the region of interest annotation window from the initial industrial big data set to obtain the industrial big data set to be analyzed;

[0062] Determining target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information includes:

[0063] The target state reasoning information is determined according to the ROI recognition state classification information corresponding to the ROI annotation window and the integrated state reasoning information.

[0064] On the other hand, the present application provides an industrial big data simulation system, including a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and the processor implements the steps in the above-described method when executing the program.

[0065] The beneficial effects of this application include at least:

[0066] The industrial big data simulation method and system based on digital twins provided in the embodiment of the present application, in the debugging process of the equipment state classification network, the multi-task state recognition component determines the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample, the first state reasoning information indicates the commonality coefficient between the corresponding industrial big data verification sample and each state classification, and the second state reasoning information indicates the confidence of the state classification corresponding to the corresponding industrial big data verification sample, and then according to the error between the first state reasoning information and the second state reasoning information of each industrial big data verification sample and the state classification of the industrial big data verification sample, respectively, the first state estimated cost and the second state estimated cost are determined, and the first state estimated cost and the second state estimated cost are loaded into the cost integration component of the equipment state classification network to be debugged to perform a cost integration operation to obtain a first target state estimated cost, and the equipment state classification network to be debugged is debugged according to the first target state estimated cost until it meets the set debugging cutoff requirement to obtain the equipment state classification network, and then the influence coefficient variable of the cost integration component in the equipment state classification network is standardized to transform the cost integration component in the equipment state classification network into a metric integration component to obtain the target equipment state classification network. In the above process, the cost integration component is used to complete the adaptive integration of the estimated costs of each state of the multi-task state reasoning component in the debugging link. At the same time, the target device state classification network including the measurement integration component is obtained according to the standardized operation of the influence coefficient variable in the cost integration component of the device state classification network, and the prediction and debugging costs are separated to complete the adaptive coordination and integration of the prediction and debugging costs respectively. It is effectively applied in the imbalance effect situation such as the abnormal state identification of the device, so that the target device state classification network not only has high-quality recognition accuracy for the state classification with more debugging samples (such as high-frequency device abnormal state classification), but also has high-quality recognition accuracy for the state classification with fewer debugging samples (such as low-frequency device abnormal state classification).

[0067] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0069] Figure 1 A schematic diagram of the debugging process of a device status classification network provided in an embodiment of the present application.

[0070] Figure 2 A schematic diagram of the composition structure of an industrial big data simulation device provided in an embodiment of the present application.

[0071] Figure 3 A hardware entity schematic diagram of an industrial big data simulation system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present application.

[0073] In the following description, reference is made to "some embodiments", which describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or different subsets of all possible embodiments, and may be combined with each other without conflict. The terms "first / second / third" involved are merely used to distinguish similar objects and do not represent a specific order for the objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0074] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing this application and are not intended to limit this application.

[0075] The embodiment of the present application provides an industrial big data simulation method based on digital twins, which can be executed by a processor of an industrial big data simulation system. The industrial big data simulation system can refer to a server, a laptop, a tablet computer, a desktop computer, or other device with data processing capabilities.

[0076] In the industrial big data simulation method based on digital twins provided in the embodiment of the present application, it is necessary to use the device status classification network to implement it. For the sake of ease of description and easy understanding, the embodiment of the present application first introduces the debugging process of the device status classification network, and then introduces the industrial big data simulation method based on digital twins based on the device status classification network. First, the debugging process of the device status classification network can include the following steps:

[0077] Step S110, obtaining an industrial big data debugging sample set and an industrial big data verification sample set.

[0078] Among them, the industrial big data debugging sample set includes industrial big data debugging samples of a set number of state classifications, and the state classification of the industrial big data verification samples in the industrial big data verification sample set is the same as the industrial big data debugging sample set. It can be understood that the industrial big data debugging sample set and the industrial big data verification sample set are both training sample sets used in the network debugging process, the industrial big data debugging sample is a support sample for debugging the network, and the industrial big data verification sample is a verification sample for evaluating the network performance in the network debugging process. Industrial big data is a data set containing target industrial production areas (such as production lines, warehouses, logistics, and energy facilities rooms). Specifically, the industrial big data debugging sample set and the industrial big data verification sample set essentially contain the operating data of industrial equipment in the target industrial production area, such as sensor data, equipment operating data, etc. Among them, the operating data corresponding to different industrial equipment may be different. For example, sensor data can represent the environmental data of industrial equipment, such as temperature data, humidity data, pressure data, etc., and equipment operating data can include fault data, operation data (such as speed, parameters, working hours, switch status, energy consumption, etc.), location data, etc. It can be understood that the above operating data can come from different industrial equipment in the target industrial production area. Each industrial equipment contains multiple address bits. For discrete data (such as fault status, switch status), it can be represented by valves (such as 0 and 1), 0 for fault, 1 for normal. For example, if an industrial equipment contains 5 address bits, according to the order of the address bits, the following binary data string "00110" is obtained, indicating that the equipment points corresponding to the 1st, 2nd and 5th address bits have faults. In this way, the binary record of discrete data is completed, or, for discrete data, data encoding can be used to complete the conversion from discrete to continuous values, such as performing unique hot encoding, etc. By obtaining the operating data of each industrial equipment, the corresponding industrial big data set can be constructed. The industrial big data set can be an M×N data matrix, each row represents an industrial equipment, and each column represents the operating data of the corresponding industrial equipment, for example, refer to the matrix shown in the following table.

[0079]

[0080]

[0081] Among them, the industrial big data debugging sample and the industrial big data verification sample are both industrial big data set samples that include the target industrial production area. The target industrial production area can correspond to an area composed of at least one industrial equipment, such as a cleaning area, which may include multiple industrial equipment such as rinsing equipment, air drying equipment, and polishing equipment. The state classification of the industrial big data set sample (industrial big data debugging sample / industrial big data verification sample) represents the labeled state classification (i.e., the actual state classification) of the target industrial production area in the industrial big data set sample, such as the equipment state classification of the industrial production area, such as the overall state of the industrial production area or the independent state combination of each industrial equipment therein. In the overall state of the industrial production area, it can be a two-class equipment state classification, such as normal, abnormal, or a multi-class equipment state classification, such as divided into level 0, level 1, level 2, level 3, etc. according to the abnormality level, where level 0 represents no abnormality, level 1 represents slight abnormality, level 2 represents general abnormality, and level 3 represents severe abnormality, etc. The specific classification standard is not limited. In the independent state combination of each industrial equipment, the operating state of each industrial equipment can be displayed and then combined, such as including industrial equipment 1, industrial equipment 2, and industrial equipment 3, and the respective equipment states are normal, abnormal, and normal, then the obtained equipment state classification is

[010] .

[0082] The state classification of the industrial big data debugging sample set and the industrial big data verification sample set is the same. For example, the industrial big data debugging sample set includes industrial big data debugging samples of P state classifications, and the industrial big data verification sample set also includes industrial big data verification samples of P state classifications. The difference is that the number of industrial big data debugging samples corresponding to each state classification in the industrial big data debugging sample set is different from the number of industrial big data verification samples corresponding to each state classification in the industrial big data verification sample set. For example, each state classification in the industrial big data debugging sample set includes H industrial big data debugging samples, and the number of industrial big data debugging sample sets is P×H industrial big data set samples; each state classification in the industrial big data verification sample set includes J industrial big data verification samples, and the data volume of the industrial big data verification sample set is a total of P×J industrial big data set samples. The following is an example of the industrial big data debugging sample set and the industrial big data verification sample set both having P state classifications (i.e., a set number of state classifications).

[0083] For example, the industrial big data debugging sample set is obtained based on arbitrary sampling of the support data set, and the industrial big data verification sample set can be obtained based on arbitrary sampling of the development data set, wherein the support data set and the development data set are obtained based on the division of the industrial big data set, and the industrial big data set includes industrial big data set samples of P state classifications, and the support data set and the development data set also have P state classifications. For example, 70% of the industrial big data set samples in the industrial big data set are randomly divided into the support data set, and 20% of the industrial big data set samples are divided into the development data set. Then, P state classifications are randomly sampled in the support data set, and the number of labeled data for each state classification is H, as the industrial big data debugging sample set (the number is P×H), and J industrial big data set samples are sampled for each state classification in the P state classifications of the development data set Development Set as the industrial big data verification sample set (the number is P×J).

[0084] Step S120, load the industrial big data debugging sample set and the industrial big data verification sample set into the multi-task state recognition component of the device state classification network to be debugged to perform device state reasoning, and obtain the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample.

[0085] The first state reasoning information is used to indicate the commonality coefficient between the corresponding industrial big data verification sample and each state classification, and the second state reasoning information is used to indicate the confidence of the corresponding industrial big data verification sample corresponding to each state classification. In other words, the reasoning result indicated by the first state reasoning information is the similarity between the industrial big data set to be analyzed and each state classification (that is, the equipment operation state category), which can be compared from the feature level of the industrial big data set to be analyzed and the feature level of the state classification (that is, the state classification representation vector) to obtain the corresponding commonality coefficient. The reasoning result of the second state reasoning information is the possibility that the industrial big data set to be analyzed belongs to each state classification.

[0086] The device state classification network to be debugged includes multiple components, each component is a network layer, wherein the multi-task state recognition component of the device state classification network to be debugged includes a representation vector extraction component and a multi-task state reasoning component, wherein the multi-task state reasoning component includes a state reasoning evaluation component and a normalized mapping reasoning component, the state reasoning evaluation component is a classifier based on distance measurement, that is, a metric classifier, such as a neighbor classifier (k-Nearest Neighbors, k-NN), a nearest center classifier (Nearest Centroid Classifier) ​​and a support vector machine (Support Vector Machine, SVM), etc., and the normalized mapping reasoning component is a fully connected neural network classifier, also known as a multi-layer perceptron (MLP) classifier.

[0087] The multi-task state recognition component of the device state classification network to be debugged for device state reasoning may include the following steps:

[0088] Step S210, loading the industrial big data debugging sample set and the industrial big data verification sample set into the representation vector extraction component to extract the representation vector, and obtaining the training representation vector of each industrial big data debugging sample and the verification representation vector of each industrial big data verification sample;

[0089] The characterization vector is the feature representation result of the corresponding sample, which is a feature vector. The characterization vector extraction component is, for example, a convolutional neural network (CNN). The characterization vector extraction component extracts the characterization vector of each input industrial big data debugging sample, completes feature extraction, and obtains the feature of each industrial big data debugging sample, namely, the training characterization vector. The characterization vector extraction component extracts the characterization vector of each input industrial big data verification sample, and obtains the feature of each industrial big data verification sample, namely, the verification characterization vector.

[0090] Step S220, determining a state classification representation vector for each state classification based on the training representation vectors of each industrial big data debugging sample corresponding to each state classification; wherein the state classification representation vector for each state classification is used to represent the characteristics of the corresponding state classification.

[0091] For example, each state classification can be characterized according to the average value of the training representation vector of the industrial big data debugging sample corresponding to each state classification in the industrial big data debugging sample set, that is, the state classification representation vector of each state classification can be determined based on the average value of the training representation vector of the state classification. The average value of the training representation vector based on each state classification is used as the state classification representation vector of the corresponding state classification. When an imbalance effect (i.e., long tail effect) occurs, redundant samples will appear, resulting in clustering imbalance, and the samples with outstanding characteristics will have insufficient influence, so that the state reasoning information accuracy of the network obtained by debugging is poor under the imbalance effect situation. Based on this, in order to make the debugged network more suitable for the imbalance effect situation and increase the network's recognition accuracy for all state classifications, step S220 can optionally include the following steps:

[0092] Step S410, averaging the training representation vectors of the industrial big data debugging samples corresponding to each state classification in the industrial big data debugging sample set to obtain the initial state centroid representation vector of each state classification.

[0093] The state centroid representation vector is a centroid (Class Center, also known as class center) feature. Suppose the H industrial big data debugging samples of the Hth state classification in the industrial big data debugging sample set are Among them, the value of n ranges from 1 to H, and the corresponding training representation vectors are Among them, the value of n is 1 to H, then the initial state centroid representation vector C of the Hth state classification H for:

[0094]

[0095] Step S420, determining the similarity measure between each training representation vector corresponding to each state classification and the initial state centroid representation vector of the state classification, and obtaining the influence coefficient of each training representation vector corresponding to each state classification.

[0096] The similarity measurement between vectors, i.e., the similarity measurement result, can be represented by spatial similarity, and can be obtained by calculating the vector distance using distance calculation methods such as Euclidean distance and cosine distance. The influence coefficient of the training representation vector represents the importance of the training representation vector and can be a weight.

[0097] Step S430, performing influence adjustment calculation on each training representation vector of each state classification according to the influence coefficient of each training representation vector corresponding to each state classification and completing the average to obtain the optimized state centroid representation vector of each state classification;

[0098] For example, for each training representation vector corresponding to each state classification, influence adjustment calculation is performed on each training representation vector according to the influence coefficient of each training representation vector and the average is completed to obtain the optimized state centroid representation vector of the state classification. The influence adjustment calculation is to perform weighted calculation on each training representation vector according to the influence coefficient. After that, after averaging the calculation results, the optimized state centroid representation vector is obtained, which can be understood as the updated state centroid representation vector.

[0099] Step S440, performing a same-state grouping operation on each training representation vector corresponding to each state classification to obtain a same-state centroid representation vector of each state classification;

[0100] The same-state grouping operation means clustering all sample representation vectors in the same state classification, also known as intra-class clustering. For example, K-Means can be used to perform the same-state grouping operation on each training representation vector corresponding to each state classification.

[0101] Step S450, integrating the optimized state centroid representation vector of each state classification with the same-state centroid representation vector of the state classification to obtain the state classification representation vector of each state classification.

[0102] For each state classification, first calculate the average value of the same-state centroid representation vectors corresponding to the state classification to obtain the same-state average state centroid representation vector, then add the optimized state centroid representation vector of the state classification and its same-state average state centroid representation vector, and determine the addition result as the state classification representation vector of the state classification.

[0103] In the above implementation method, the similarity measure between the training representation vector and the initial state centroid representation vector is used as the influence coefficient to optimize the state centroid representation vector, so that the influence coefficient with greater commonality with the centroid is smaller. In this way, the influence of similar industrial large data set samples is reduced, which can reduce the clustering imbalance problem caused by redundant samples and prevent prominent samples from being submerged. At the same time, the influence coefficient with less commonality with the centroid is larger, so that prominent samples are valued. Based on the grouping of the same state into multiple prominent centroids, the clustering imbalance caused by redundant samples can also be reduced. Furthermore, integrating the above optimized state centroid representation vector and the same state centroid representation vector can obtain a relatively fixed state centroid representation vector, so that the debugged network can be better applied in the imbalance effect situation and increase the network's recognition accuracy for all state classifications.

[0104] Step S230, determining the commonality coefficient between each verification representation vector and each state classification representation vector according to the state reasoning evaluation component, and obtaining the first state reasoning information of each industrial big data verification sample;

[0105] For example, for each verification representation vector, the corresponding verification representation vector can be combined with each state classification representation vector to obtain a representation vector group, and each representation vector group is loaded into the state reasoning evaluation component. The state reasoning evaluation component performs commonality measurement based on each representation vector group, such as commonality measurement based on Euclidean distance, and ignores the commonality coefficient between the verification representation vector corresponding to the industrial big data verification sample and each state classification.

[0106] Step S240, determining the confidence of each verification representation vector corresponding to each state classification based on the normalized mapping reasoning component, and obtaining the second state reasoning information of each industrial big data verification sample.

[0107] For example, for each verification representation vector, the verification representation vector is loaded into the normalized mapping reasoning component, and the normalized mapping reasoning component uses a dense layer to predict the confidence of each state classification according to the verification representation vector.

[0108] In the above steps, the state reasoning evaluation component in the multi-task state reasoning component uses the same-state features of each state classification to represent the state classification representation vector of the state classification, ignoring the connection between the state classifications, and has a better effect when there are fewer samples. The normalized mapping reasoning component in the multi-task state reasoning component takes into account the connection between the state classifications based on the learned feature information of all state classifications, so it has a better effect when there are more samples. That is to say, when there are more samples, the normalized mapping reasoning component can better learn the unique information of each state classification, and at the same time take into account the connection between the state classifications. The adaptive integration based on the embodiment of the present application can make the debugged network more applicable in application scenarios with imbalance effects such as abnormal device state identification, thereby increasing the network's recognition accuracy for all state classifications.

[0109] Step S130, determining the first state estimated cost and the second state estimated cost according to the errors between the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample and the state classification of the industrial big data verification sample.

[0110] According to steps S110 and S120, the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample in the industrial big data verification sample set can be obtained. Then in this step S130, for each industrial big data verification sample in the industrial big data verification sample set, the error (called the first error) between the first state reasoning information corresponding to the industrial big data verification sample and the state classification (specifically a state classification mark) of the industrial big data verification sample can be determined, and the error (called the second error) between the second state reasoning information corresponding to the industrial big data verification sample and the state classification (specifically a state classification mark) of the industrial big data verification sample can be determined. According to the first error corresponding to each industrial big data verification sample, a preset cost function is used to obtain the first state estimated cost, and according to the second error corresponding to each industrial big data verification sample, the second state estimated cost is obtained based on the preset cost function.

[0111] The first state estimated cost is used to make the inference result of the state inference evaluation component that outputs the first state inference information more consistent with the actual state classification, and the second state estimated cost is used to make the inference result of the normalized mapping inference component that outputs the second state inference information more consistent with the actual state classification. The preset cost function for obtaining the first state estimated cost and the preset cost function for obtaining the second state estimated cost can be the same or different, for example, both are cross entropy cost functions.

[0112] Step S140, loading the first state estimated cost and the second state estimated cost into the cost integration component of the device state classification network to be debugged to perform a cost integration operation, and obtaining a first target state estimated cost; debugging the device state classification network to be debugged according to the first target state estimated cost until it meets the set debugging cutoff requirement, and obtaining the device state classification network.

[0113] The cost integration component includes an adjustable influence coefficient variable, that is, a weight parameter. The first state estimated cost and the second state estimated cost are weighted and summed through the influence coefficient variable to obtain the first target state estimated cost. As a feasible design, the influence coefficient variable of the cost integration component includes the first cost influence coefficient (that is, the cost weights) and the second cost influence coefficient. The first state estimated cost and the second state estimated cost are loaded into the cost integration component of the state classification network of the device to be debugged for cost integration operation. The process of obtaining the first target state estimated cost includes: loading the first state estimated cost into the cost integration component, and the cost integration component calculates the first target state estimated cost based on the first cost influence coefficient and the corresponding first cost offset (that is, the cost offset cost). biases) outputs and calculates the estimated cost of the first state to obtain the estimated cost of the first target sub-state; wherein, the first cost influence coefficient is negatively correlated with the first cost offset, in other words, the first cost offset decreases as the first cost influence coefficient increases, and increases as the first cost influence coefficient decreases, and the first cost influence coefficient and the first cost offset constrain each other; the second state estimated cost is loaded into the cost integration component, and the cost integration component outputs and calculates the second state estimated cost according to the second cost influence coefficient and the corresponding second cost offset to obtain the estimated cost of the second target sub-state; the second cost influence coefficient is negatively correlated with the second cost offset, in other words, as the second cost influence coefficient increases, the second cost offset decreases, and as the second cost influence coefficient decreases, the second cost offset increases, that is, the second cost influence coefficient and the second cost offset constrain each other; the first target sub-state estimated cost is added to the second target sub-state estimated cost to obtain the first target state estimated cost.

[0114] For example, the estimated cost of the first state is C 1a , the estimated cost of the second state is C 2a , then the estimated cost of the first target state is C a for:

[0115] C a =(α 1a C 1a +b 1a )+(α 2a C 2a +b 2a )

[0116] Among them, (α 1a C 1a +b1a) is the estimated cost of the first target sub-state; (α 2a C 2a +b 2a ) is the estimated cost of the second target sub-state; α 1a C 1a is the first cost influence coefficient, b1a is the first cost offset; α 2a C 2a is the second cost influence coefficient, b 2a is the second cost offset.

[0117] For example, the mutual constraint between the cost influence coefficient and the cost offset can be completed by using an intermediate variable s, based on which the cost influence coefficient α and the cost offset b constrained by the cost influence coefficient α are obtained. For example, α=1 / (2s 2 ); b = logs 2 .

[0118] Based on this, the mutual constraints between the cost impact coefficient and the cost offset of the cost integration component can avoid the cost of one component being too prominent in the cost integration structure, so as to increase the performance of the network.

[0119] When debugging the device state classification network to be debugged according to the estimated cost of the first target state, the internal parameters of the multi-task state identification component and the cost integration component in the device state classification network to be debugged can be corrected according to the estimated cost of the first target state, and the device state classification network to be debugged after the parameter correction is repeatedly debugged until the set debugging cutoff requirements are met (such as the number of debugging times reaches the maximum number of debugging times, the difference in cost between the two previous and subsequent debugging times is less than a threshold, etc.) to obtain the device state classification network.

[0120] Step S150 , performing a standardization operation on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component, and obtaining a target device state classification network.

[0121] For example, the first cost influence coefficient and the second cost influence coefficient of the cost integration component are standardized, so that the transformed metric integration component includes the corresponding first metric adjustment influence coefficient and the second metric adjustment influence coefficient, wherein the first metric adjustment influence coefficient is used to adjust the confidence measure (also called confidence measure) of the predicted information during the result reasoning process of the state reasoning evaluation component that outputs the first state reasoning information, and the second metric adjustment influence coefficient is used to adjust the confidence measure of the predicted information during the reasoning process of the normalized mapping reasoning component that outputs the second state reasoning information. By adjusting the confidence measure, the confidence of the network output is more in line with the actual situation or more explainable, thereby improving the reliability and practicality of the network.

[0122] In the above process of the embodiment of the present application, the adaptive integration of the estimated costs of each state of the multi-task state reasoning component in the debugging link is completed based on the cost integration component. At the same time, the target device state classification network including the metric integration component is obtained according to the standardized operation of the influence coefficient variable in the cost integration component of the device state classification network, and the prediction and debugging costs are separated. The adaptive coordination and integration of the prediction and debugging costs are completed respectively, which is effectively applied to the imbalance effect situation during the identification of abnormal state of the device, so that the target device state classification network not only has high-quality recognition accuracy for the state classification with more debugging samples (such as high-frequency device abnormal state classification), but also has high-quality recognition accuracy for the state classification with fewer debugging samples (such as low-frequency device abnormal state classification).

[0123] In order to make the confidence measures of different state reasoning information of the multi-task state recognition component closer, facilitate the integration of prediction information, and increase the recognition accuracy of the network in each state classification, optionally, after step S120, the following steps may also be included:

[0124] Step S510, determining a first debugging cost and a second debugging cost according to an error between the first state reasoning information and the second state reasoning information of each industrial big data verification sample;

[0125] The first debugging cost is used to make the first state reasoning information in the debugging link close to the second state reasoning information, and the second debugging cost is used to make the second state reasoning information in the debugging link close to the first state reasoning information, so that the classification components in the multi-task state reasoning component constrain each other. For example, the first debugging cost and the second debugging cost are both cross entropy cost functions.

[0126] Step S520: Add the first debugging cost and the second debugging cost to obtain a second target state estimated cost.

[0127] Then, when step S140 is executed to debug the device state classification network to be debugged according to the first target state estimated cost until the device state classification network meets the set debugging deadline requirement, it includes:

[0128] Step S530, determining the total state estimated cost according to the first target state estimated cost and the second target state estimated cost.

[0129] For example, the first target state estimated cost and the second target state estimated cost are added, and the addition result is determined as the total state estimated cost.

[0130] Step S540 , debugging the device state classification network to be debugged according to the total state estimated cost until it meets the set debugging deadline requirement, thereby obtaining the device state classification network.

[0131] For example, the parameters of the multi-task state recognition component and the cost integration component in the device state classification network to be debugged are corrected according to the total state estimated cost, and the device state classification network to be debugged is repeatedly debugged according to the corrected parameters until the device state classification network meets the set debugging cutoff requirements.

[0132] Optionally, step S110, obtaining an industrial big data debugging sample set and an industrial big data verification sample set, specifically includes: obtaining an industrial big data database; the industrial big data database includes industrial big data set samples of a set number of state categories; dividing the industrial big data database into a support data set and a development data set; wherein the support data set and the development data set both have industrial big data set samples of the set number of state categories; performing multiple debugging data sampling on the support data set and the development data set, each time arbitrarily sampling a first set number of industrial big data set samples in each state category of the support data set to obtain an industrial big data debugging sample set of corresponding debugging data, and arbitrarily sampling a second set number of industrial big data set samples in each state category of the development data set to obtain an industrial big data verification sample set of the debugging data; based on any debugging data among the multiple debugging data, obtaining an industrial big data debugging sample set and an industrial big data verification sample set.

[0133] The above process, based on the organization of the industrial big data database, performs multiple debugging data sampling on the support data set and the development data set to obtain multiple debugging data, each debugging data includes an industrial big data debugging sample set sampled according to the support data set and an industrial big data verification sample set sampled according to the development data set. Subsequently, steps S110 to S140 of the embodiment of the present application can be implemented on any of the multiple debugging data to obtain a device status classification network debugged according to each debugging data.

[0134] Optionally, step S150 may specifically include: obtaining a device state classification network obtained by debugging respectively according to multiple debugging data; determining a state reasoning effect parameter of each device state classification network according to a development data set; determining a candidate device state classification network from multiple device state classification networks according to the state reasoning effect parameter of each device state classification network; the state reasoning effect parameter of the candidate device state classification network is better than the state reasoning effect parameter of any remaining device state classification network; and performing a standardization operation on an influence coefficient variable of a cost integration component in the candidate device state classification network to transform the cost integration component in the candidate device state classification network into a metric integration component to obtain a target device state classification network.

[0135] In the above steps, based on determining the candidate device state classification network with the best state reasoning effect parameters, the influence coefficient variables of its cost integration components are standardized to obtain the target device state classification network, which can increase the state recognition effect of the target device state classification network.

[0136] For the device state classification network to be debugged, its network architecture may include a multi-task state recognition component and a cost integration component, wherein the multi-task state recognition component includes a representation vector extraction component (core component, BackboneNetwork), a centroid adjustment component and a multi-task state reasoning component, the multi-task state reasoning component includes a state reasoning evaluation component and a normalized mapping reasoning component, the cost integration component includes an adjustable cost influence coefficient, and the cost influence coefficient includes a cost influence coefficient for integrating the state estimated cost of the state reasoning evaluation component and the state estimated cost of the normalized mapping reasoning component;

[0137] The representation vector extraction component is used to implement step S210 to obtain the training representation vector of each industrial big data debugging sample and the verification representation vector of each industrial big data verification sample; the centroid adjustment component is used to implement step S220 to obtain the state classification representation vector of each state classification; the state reasoning evaluation component is used to implement step S230 to obtain the first state reasoning information of each industrial big data verification sample; the normalized mapping reasoning component is used to implement step S240 to obtain the second state reasoning information of each industrial big data verification sample. According to the first state reasoning information and the second state reasoning information, the state estimated cost of the state reasoning evaluation component, i.e., the first state estimated cost, and the state estimated cost of the normalized mapping reasoning component, i.e., the second state estimated cost, can be obtained; the first state estimated cost and the second state estimated cost are loaded into the cost integration component to obtain the first target state estimated cost. Then, based on the first target state estimated cost and the second target state estimated cost, the total state estimated cost is obtained, and feedback is performed based on the total state estimated cost to correct the parameters in the multi-task state recognition component and the cost integration component, and debugging is performed to obtain the device state classification network. Then, the cost influence coefficient of the cost integration component in the device state classification network is standardized (the cost integration component after the standardized operation is the metric integration component) to obtain the target device state classification network.

[0138] At this point, the introduction of the debugging process of the device status classification network is completed. The following describes the process of implementing the industrial big data simulation method based on digital twins provided in the embodiment of the present application based on the target device status classification network obtained by the debugging, including the following steps:

[0139] Step S100, obtaining an industrial big data set to be analyzed.

[0140] In the embodiment of the present application, the industrial big data set to be analyzed is industrial data including the target industrial production area (such as production line, warehouse, logistics, energy facility room). Specifically, the industrial big data set to be analyzed includes the operation data of industrial equipment in the target industrial production area, such as sensor data, equipment operation data, etc. Among them, the operation data corresponding to different industrial equipment may be different. For example, sensor data can represent the environmental data of industrial equipment, such as temperature data, humidity data, pressure data, etc., and equipment operation data can include fault data, operation data (such as speed, parameters, working hours, switch status, energy consumption, etc.), position data, etc. It can be understood that the above operation data can come from different industrial equipment in the target industrial production area, and each industrial equipment contains multiple address bits. For discrete data (such as fault status, switch status), it can be represented by valves (such as 0 and 1), 0 is fault, 1 is normal, for example, if an industrial equipment contains 5 address bits, according to the order of the address bits, the following binary data string "00110" is obtained, indicating that the equipment points corresponding to the 1st, 2nd and 5th address bits have faults. In this way, the binary record of discrete data is completed, or data encoding can be used to complete the conversion from discrete to continuous values ​​for discrete data, such as one-hot encoding, etc. By obtaining the operating data of each industrial equipment, the corresponding industrial big data set can be constructed. The industrial big data set can be an M×N data matrix, each row represents an industrial equipment, and each column represents the operating data of the corresponding industrial equipment, for example, refer to the matrix shown in the following table.

[0141]

[0142] Step S200, according to the target state classification representation vector of each state classification in a set number of state classifications, a multi-task state recognition component based on the target device state classification network performs device state reasoning on the industrial big data set to be analyzed to obtain first state reasoning information and second state reasoning information.

[0143] Among them, the first state reasoning information indicates the commonality coefficient between the industrial big data set to be analyzed and each state classification, and the second state reasoning information indicates the confidence of the industrial big data set to be analyzed corresponding to each state classification. In other words, the reasoning result indicated by the first state reasoning information is the similarity between the industrial big data set to be analyzed and each state classification (that is, the equipment operation state category), which can be compared from the feature level of the industrial big data set to be analyzed and the feature level of the state classification (that is, the state classification representation vector) to obtain the corresponding commonality coefficient. The reasoning result of the second state reasoning information is the possibility that the industrial big data set to be analyzed belongs to each state classification.

[0144] The target state classification representation vector of each state classification is determined according to the target training representation vector (i.e., sample features) corresponding to the target industrial big data debugging sample of each state classification in the target industrial big data debugging sample set, and the target training representation vector is obtained by extracting the representation vector according to the target industrial big data debugging sample corresponding to the target device state classification network. Among them, in order to facilitate the distinction between the industrial big data debugging sample set used in the network application link and the industrial big data debugging sample set used in the network debugging link during the introduction, the industrial big data debugging sample set in the network application link is regarded as the target industrial big data debugging sample set. In other words, the acquisition method and state classification of the target industrial big data debugging sample set and the industrial big data debugging sample set in the network debugging link are the same. The target industrial big data debugging sample set can be P×H industrial big data set samples sampled from the supporting data set of the industrial big data debugging sample set extracted from the subsequent introduction. In a feasible design, the target state classification representation vector of each state classification in the set number of state classifications can be obtained by processing each target training representation vector of the target industrial big data debugging sample set through the steps S410 to S450 in the introduction of the embodiment of this application.

[0145] In a feasible design, the multi-task state recognition component of the target device state classification network includes a representation vector extraction component and a multi-task state reasoning component, and the multi-task state reasoning component includes a state reasoning evaluation component and a normalized mapping reasoning component. Step S120 may specifically include: extracting a representation vector of the industrial big data set to be analyzed based on the representation vector extraction component of the multi-task state recognition component in the target device state classification network to obtain a representation vector of the industrial big data set to be analyzed; determining the commonality coefficient between the representation vector of the industrial big data set to be analyzed and the representation vectors of each target state classification based on the state reasoning evaluation component of the multi-task state recognition component in the target device state classification network to obtain first state reasoning information; determining the confidence of the representation vector of the industrial big data set to be analyzed corresponding to each state classification based on the normalized mapping reasoning component of the multi-task state recognition component in the target device state classification network to obtain second state reasoning information.

[0146] Step S300 , performing a metric integration operation on the first state reasoning information and the second state reasoning information according to the metric integration component of the target device state classification network to obtain integrated state reasoning information.

[0147] Because the transformed metric integration component includes a first metric adjustment influence coefficient and a second metric adjustment influence coefficient, wherein the first metric adjustment influence coefficient can adjust the confidence metric of the first state reasoning information, and the second metric adjustment influence coefficient can adjust the confidence metric of the second state reasoning information, after loading the first state reasoning information and the second state reasoning information into the metric integration component, the metric integration component can perform weighted summation of the first state reasoning information and the second state reasoning information according to the first metric adjustment influence coefficient and the second metric adjustment influence coefficient to complete the integration of the first state reasoning information and the second state reasoning information.

[0148] Based on the metric adjustment influence coefficients of different classification components (state reasoning evaluation component, normalized mapping reasoning component) learned during the debugging process, the confidence metric of the prediction information of each classification component can be globally adjusted to achieve adaptive coordination and integration.

[0149] Since the state reasoning evaluation component that outputs the first state reasoning information ignores the connection between state classifications during the classification process, it has better results when there are fewer samples. The normalized mapping reasoning component in the multi-task state reasoning component takes into account the connection between state classifications based on the feature information learned from all state classifications, which has better results when there are more samples. Then, in order to further increase the recognition accuracy for all state classifications in the case of imbalance effect, in a feasible design, step S300 specifically includes: determining the first state reasoning information influence coefficient and the second state reasoning information influence coefficient based on the number of industrial big data set samples corresponding to each state classification in the supporting data set; wherein the first state reasoning information influence coefficient is negatively correlated with the number of industrial big data set samples corresponding to each state classification, and the second state reasoning information influence coefficient is positively correlated with the number of industrial big data set samples corresponding to each state classification, and the sum of the first state reasoning information influence coefficient and the second state reasoning information influence coefficient is the set value; multiplying the first state reasoning information influence coefficient by the first state reasoning information to obtain the target first state reasoning information; and multiplying the second state reasoning information influence coefficient by the second state reasoning information to obtain the target second state reasoning information; loading the target first state reasoning information and the target second state reasoning information into the metric integration component of the target device state classification network to perform metric integration operations to obtain integrated state reasoning information.

[0150] The supporting data set is used to sample and obtain the industrial big data debugging sample set (including the target industrial big data debugging sample set) in the embodiment of the present application. In other words, the target industrial big data debugging sample set is obtained based on the supporting data set including industrial big data set samples of a set number of state classifications.

[0151] The set value can be 1. For example, the integrated state inference information can be obtained according to the following formula (10):

[0152] D=e·F1·D1+(1-e)·F2·D2

[0153] Where, e=1 / x, x=(x1,x2……x p ), is the number of industrial big data set samples supporting each state classification in the data set, P is the set number of state classifications; F1 is the first metric adjustment influence coefficient, F2 is the second metric adjustment influence coefficient, D1 is the first state reasoning information, D2 is the second state reasoning information; e is the first state reasoning information influence coefficient; 1-e is the second state reasoning information influence coefficient; e·F1·D1 is the target first state reasoning information; (1-e)·F2·D2 is the target second state reasoning information.

[0154] In the above steps, for state classifications with more samples, the influence coefficient of the second state reasoning information is large, and for state classifications with fewer samples, the influence coefficient of the first state reasoning information is large. If the first state reasoning information is obtained through the output of the state reasoning evaluation component and the second state reasoning information is obtained through the output of the normalized mapping reasoning component, the state classification with more samples is completed, the influence coefficient of the normalized mapping reasoning component is large, and the state classification with fewer samples is completed, and the influence coefficient of the state reasoning evaluation component is large. In this way, the value of the two classification components can be utilized as much as possible, and the influence coefficient is adjusted based on different metrics at the same time, and the confidence metrics of the prediction information of the two classification components are globally adjusted to achieve adaptive coordinated integration, so that the integrated state reasoning information after integration is more accurate, the recognition ability is strong in the imbalance effect situation, and the recognition accuracy of all state classifications is high.

[0155] Step S400: determining target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information.

[0156] For example, the integrated state reasoning information is used as the target state reasoning information corresponding to the industrial big data set to be analyzed, and the state classification corresponding to the maximum confidence in the integrated state reasoning information is determined as the target state classification of the industrial big data set to be analyzed for output.

[0157] Optionally, step S100, obtaining the industrial big data set to be analyzed may include: obtaining an initial industrial big data set, performing region of interest identification processing on the initial industrial big data set, obtaining a region of interest annotation window and region of interest identification status classification information corresponding to the region of interest annotation window; obtaining an industrial data cluster corresponding to the region of interest annotation window from the initial industrial big data set, and obtaining the industrial big data set to be analyzed in step S100.

[0158] The region of interest recognition can be implemented based on a previously debugged region of interest recognition network, which can be, for example, a YOLO (You Only Look Once) network, a region proposal network (RPN), an R-CNN (Region-based Convolutional Neural Network) or other neural network. For details, please refer to the prior art.

[0159] The initial industrial big data set is loaded into the region of interest recognition network for region of interest recognition, and then the region of interest recognition result is output, which includes the region of interest annotation window and the region of interest recognition state classification information corresponding to the region of interest annotation window. The region of interest recognition state classification information represents the confidence level of the target industrial production area in the corresponding region of interest annotation window for each state classification.

[0160] Step S400 may specifically include: determining target state reasoning information according to the region of interest recognition state classification information and integration state reasoning information corresponding to the region of interest annotation window.

[0161] The identification state classification information of the region of interest corresponding to the region of interest annotation window is summed with the integrated state reasoning information to obtain the target state reasoning information, and the state classification corresponding to the maximum confidence in the target state reasoning information is used as the target state classification of the industrial big data set to be analyzed.

[0162] Step S500: Perform corresponding state display in the industrial digital twin model based on the target state inference information.

[0163] It can be understood that in the industrial data digital twin, the real industrial data is classified by equipment status in order to better understand and analyze the data. By classifying industrial data, it can be organized into meaningful groups, making it easier to identify patterns, trends, and anomalies in the data. In the preset digital twin model, the corresponding status is displayed based on the equipment status classification results. For example, the equipment status of industrial equipment can be divided into equipment status classifications such as normal, abnormal, and faulty, and different equipment states can be simulated in the digital twin model with different colors or icons. In this way, users can intuitively understand the current status of industrial equipment through the digital twin model and take corresponding measures in a timely manner. The embodiment of the present application does not involve the construction process of the industrial digital twin model, and please refer to the prior art for this process.

[0164] Based on the foregoing embodiments, the embodiments of the present application provide an industrial big data simulation device, and the various units included in the device, and the various modules included in each unit, can be implemented by a processor in a computer device; of course, they can also be implemented by a specific logic circuit; in the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP) or a field programmable gate array (FPGA), etc.

[0165] Figure 2 A schematic diagram of the composition structure of an industrial big data simulation device provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, the industrial big data simulation device 200 includes:

[0166] A data acquisition module 210 is used to acquire a large industrial data set to be analyzed;

[0167] A state reasoning module 220 is used to perform device state reasoning on the industrial big data set to be analyzed based on the target state classification representation vector of each state classification in the set number of state classifications and the multi-task state recognition component based on the target device state classification network to obtain first state reasoning information and second state reasoning information;

[0168] The first state reasoning information indicates a commonality coefficient between the industrial big data set to be analyzed and each of the state classifications, and the second state reasoning information indicates a confidence level of the industrial big data set to be analyzed corresponding to each of the state classifications;

[0169] A state integration module 230, configured to perform a metric integration operation on the first state reasoning information and the second state reasoning information according to a metric integration component of the target device state classification network to obtain integrated state reasoning information;

[0170] A state determination module 240, configured to determine target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information;

[0171] A twin display module 250, used to perform corresponding state display in the industrial digital twin model according to the target state reasoning information;

[0172] The target state classification representation vector of each state classification is determined according to the target training representation vector corresponding to the target industrial big data debugging sample of each state classification in the target industrial big data debugging sample set, and the target training representation vector is obtained by extracting the representation vector of the corresponding target industrial big data debugging sample according to the target device state classification network; wherein, the target device state classification network is debugged according to the following steps:

[0173] Obtain an industrial big data debugging sample set and an industrial big data verification sample set; the industrial big data debugging sample set includes industrial big data debugging samples of a set number of status classifications, and the status classification of the industrial big data verification samples in the industrial big data verification sample set is the same as that of the industrial big data debugging sample set;

[0174] The industrial big data debugging sample set and the industrial big data verification sample set are loaded into the multi-task state recognition component of the state classification network of the device to be debugged to perform device state reasoning, and the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample are obtained; the first state reasoning information indicates the commonality coefficient between the corresponding industrial big data verification sample and each of the state classifications, and the second state reasoning information indicates the confidence of the corresponding industrial big data verification sample corresponding to each of the state classifications;

[0175] Determine a first state estimated cost and a second state estimated cost according to errors between the first state reasoning information and the second state reasoning information corresponding to each of the industrial big data verification samples and the state classification of the industrial big data verification samples;

[0176] Loading the first state estimated cost and the second state estimated cost into the cost integration component of the state classification network of the device to be debugged to perform a cost integration operation to obtain a first target state estimated cost;

[0177] Debugging the device state classification network to be debugged according to the first target state estimated cost until it meets the set debugging cutoff requirement, thereby obtaining a device state classification network;

[0178] A standardization operation is performed on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component to obtain a target device state classification network.

[0179] The description of the above device embodiment is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. In some embodiments, the functions or modules included in the device provided in the embodiment of the present application can be used to execute the method described in the above method embodiment. For technical details not disclosed in the device embodiment of the present application, please refer to the description of the method embodiment of the present application for understanding.

[0180] It should be noted that in the embodiment of the present application, if the above-mentioned industrial big data simulation method based on digital twins is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can be essentially or partly reflected in the form of a software product that contributes to the relevant technology. The software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a disk or an optical disk. In this way, the embodiment of the present application is not limited to any specific hardware, software or firmware, or any combination of hardware, software, and firmware.

[0181] An embodiment of the present application provides an industrial big data simulation system, including a memory and a processor, wherein the memory stores a computer program that can be executed on the processor, and when the processor executes the program, some or all of the steps in the above method are implemented.

[0182] The embodiment of the present application provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, some or all of the steps in the above method are implemented. The computer-readable storage medium can be transient or non-transient.

[0183] An embodiment of the present application provides a computer program, including a computer-readable code. When the computer-readable code is run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0184] The embodiment of the present application provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium, and in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK) and the like.

[0185] It should be noted here that the description of the various embodiments above tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. The description of the above device, storage medium, computer program and computer program product embodiments is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiments of the device, storage medium, computer program and computer program product of this application, please refer to the description of the method embodiment of this application for understanding.

[0186] Figure 3 A hardware entity diagram of an industrial big data simulation system provided in an embodiment of the present application is shown in FIG. Figure 3 As shown, the hardware entity of the industrial big data simulation system 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program.

[0187] The memory 1002 stores a computer program that can be run on the processor. The memory 1002 is configured to store instructions and applications executable by the processor 1001. It can also cache data to be processed or processed by the processor 1001 and each module in the industrial big data simulation system 1000 (for example, image data, audio data, voice communication data, and video communication data). This can be achieved through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0188] When the processor 1001 executes the program, the steps of any of the above-mentioned industrial big data simulation methods based on digital twins are implemented. The processor 1001 generally controls the overall operation of the industrial big data simulation system 1000.

[0189] An embodiment of the present application provides a computer storage medium, which stores one or more programs. The one or more programs can be executed by one or more processors to implement the steps of the industrial big data simulation method based on digital twins in any of the above embodiments.

[0190] It should be noted here that the description of the above storage medium and device embodiments is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments. For the technical details not disclosed in the storage medium and device embodiments of the present application, please refer to the description of the method embodiments of the present application for understanding. The above processor can be at least one of a target application integrated circuit (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD), a programmable logic device (Programmable Logic Device, PLD), a field programmable gate array (Field Programmable Gate Array, FPGA), a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, and a microprocessor. It can be understood that the electronic device that realizes the above processor function can also be other, and the embodiments of the present application are not specifically limited.

[0191] The above-mentioned computer storage medium / memory can 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), a magnetic random access memory (FRAM), a flash memory (Flash Memory), a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM) and the like; it can also be various terminals including one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0192] It should be understood that the "one embodiment" or "one embodiment" mentioned throughout the specification means that the specific features, structures or characteristics related to the embodiment are included in at least one embodiment of the present application. Therefore, "in one embodiment" or "in one embodiment" appearing in various places throughout the specification may not necessarily refer to the same embodiment. In addition, these specific features, structures or characteristics can be combined in one or more embodiments in any suitable manner. It should be understood that in various embodiments of the present application, the size of the sequence number of each step / process mentioned above does not mean the order of execution, and the execution order of each step / process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application. The above-mentioned sequence number of the embodiment of the present application is only for description and does not represent the advantages and disadvantages of the embodiment. It should be noted that, in this article, the term "include", "comprise" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.

[0193] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. 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 can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0194] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; 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 present embodiment.

[0195] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0196] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0197] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the relevant technology can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0198] The above is only an implementation method of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. An industrial big data simulation method based on digital twins, characterized in that: The method comprises: Obtaining industrial large data sets to be analyzed; According to the target state classification representation vector of each state classification in the set number of state classifications, a multi-task state recognition component based on the target device state classification network performs device state reasoning on the industrial big data set to be analyzed to obtain first state reasoning information and second state reasoning information; The first state reasoning information indicates a commonality coefficient between the industrial big data set to be analyzed and each of the state classifications, and the second state reasoning information indicates a confidence level of the industrial big data set to be analyzed corresponding to each of the state classifications; Performing a metric integration operation on the first state reasoning information and the second state reasoning information according to the metric integration component of the target device state classification network to obtain integrated state reasoning information; Determining target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information; Performing corresponding state display in the industrial digital twin model based on the target state reasoning information; The target state classification representation vector of each state classification is determined according to the target training representation vector corresponding to the target industrial big data debugging sample of each state classification in the target industrial big data debugging sample set, and the target training representation vector is obtained by extracting the representation vector of the corresponding target industrial big data debugging sample according to the target device state classification network; wherein, the target device state classification network is debugged and obtained according to the following steps: Obtain an industrial big data debugging sample set and an industrial big data verification sample set; the industrial big data debugging sample set includes industrial big data debugging samples of a set number of status classifications, and the status classification of the industrial big data verification samples in the industrial big data verification sample set is the same as that of the industrial big data debugging sample set; The industrial big data debugging sample set and the industrial big data verification sample set are loaded into the multi-task state recognition component of the state classification network of the device to be debugged to perform device state reasoning, and the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample are obtained; the first state reasoning information indicates the commonality coefficient between the corresponding industrial big data verification sample and each of the state classifications, and the second state reasoning information indicates the confidence of the corresponding industrial big data verification sample corresponding to each of the state classifications; Determine a first state estimated cost and a second state estimated cost according to errors between the first state reasoning information and the second state reasoning information corresponding to each of the industrial big data verification samples and the state classification of the industrial big data verification samples; Loading the first state estimated cost and the second state estimated cost into the cost integration component of the state classification network of the device to be debugged to perform a cost integration operation to obtain a first target state estimated cost; Debugging the device state classification network to be debugged according to the first target state estimated cost until it meets the set debugging cutoff requirement, thereby obtaining a device state classification network; Performing a standardization operation on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component to obtain a target device state classification network; The influence coefficient variable of the cost integration component includes a first cost influence coefficient and a second cost influence coefficient; the step of loading the first state estimated cost and the first state estimated cost into the cost integration component of the device state classification network to be debugged to perform a cost integration operation to obtain the first target state estimated cost includes: Loading the first state estimated cost into the cost integration component of the state classification network of the device to be debugged, outputting and calculating the first state estimated cost according to the first cost influence coefficient and the corresponding first cost offset, to obtain the first target sub-state estimated cost; the first cost influence coefficient is negatively correlated with the first cost offset; The second state estimated cost is loaded into the cost integration component of the state classification network of the device to be debugged, and the second state estimated cost is output and calculated according to the second cost influence coefficient and the corresponding second cost offset to obtain the second target sub-state estimated cost; the second cost influence coefficient is negatively correlated with the second cost offset; Adding the first target sub-state estimated cost and the second target sub-state estimated cost to obtain a first target state estimated cost; The obtaining of the industrial big data debugging sample set and the industrial big data verification sample set includes: Acquire an industrial big data database; the industrial big data database includes the industrial big data set samples of the set number of state classifications; The industrial big data set is divided into a support data set and a development data set; the support data set and the development data set both have the industrial big data set samples of the set number of state classifications; Performing multiple debugging data sampling on the support data set and the development data set, each time arbitrarily sampling a first set number of industrial big data set samples in each state classification of the support data set to obtain an industrial big data debugging sample set of corresponding debugging data, and arbitrarily sampling a second set number of industrial big data set samples in each state classification of the development data set to obtain an industrial big data verification sample set of the debugging data; According to any one of the plurality of debugging data, obtaining the industrial big data debugging sample set and the industrial big data verification sample set; The step of performing a standardization operation on the influence coefficient variable of the cost integration component in the device state classification network to transform the cost integration component in the device state classification network into a metric integration component to obtain a target device state classification network comprises: Acquire a device status classification network obtained by debugging respectively according to a plurality of the debugging data; Determining a state reasoning effect parameter of each of the device state classification networks based on the development data set; According to the state reasoning effect parameter of each of the device state classification networks, a candidate device state classification network is determined from the plurality of the device state classification networks; the state reasoning effect parameter of the candidate device state classification network is better than the state reasoning effect parameter of any of the remaining device state classification networks; A standardization operation is performed on the influence coefficient variable of the cost integration component in the candidate device state classification network to transform the cost integration component in the candidate device state classification network into a metric integration component to obtain a target device state classification network.

2. The method according to claim 1, characterized in that The method further comprises: Determine a first debugging cost and a second debugging cost according to an error between the first state reasoning information and the second state reasoning information of each industrial big data verification sample; the first debugging cost is used to make the first state reasoning information close to the second state reasoning information, and the second debugging cost is used to make the second state reasoning information close to the first state reasoning information; Adding the first debugging cost and the second debugging cost to obtain a second target state estimated cost; The step of debugging the device state classification network to be debugged according to the first target state estimated cost until the debugging cutoff requirement is met to obtain the device state classification network includes: Determining a total state estimated cost according to the first target state estimated cost and the second target state estimated cost; The device state classification network to be debugged is debugged according to the total state estimated cost until it meets the set debugging cutoff requirement, thereby obtaining the device state classification network.

3. The method according to claim 1, characterized in that The multi-task state recognition component includes a representation vector extraction component and a multi-task state reasoning component, and the multi-task state reasoning component includes a state reasoning evaluation component and a normalized mapping reasoning component; the multi-task state recognition component of the industrial big data debugging sample set and the industrial big data verification sample set are loaded into the state classification network of the device to be debugged to perform device state reasoning, and obtain the first state reasoning information and the second state reasoning information corresponding to each industrial big data verification sample, including: Loading the industrial big data debugging sample set and the industrial big data verification sample set into the representation vector extraction component to extract the representation vector, and obtaining the training representation vector of each industrial big data debugging sample and the verification representation vector of each industrial big data verification sample; Determine a state classification representation vector for each state classification according to the training representation vectors of each industrial big data debugging sample corresponding to each state classification; Determine the commonality coefficient between each verification characterization vector and each of the state classification characterization vectors according to the state reasoning evaluation component, and obtain the first state reasoning information of each of the industrial big data verification samples; Determine the confidence of each verification representation vector corresponding to each state classification according to the normalized mapping reasoning component, and obtain the second state reasoning information of each industrial big data verification sample; The step of determining the state classification representation vector of each state classification according to the training representation vector of each industrial big data debugging sample corresponding to each state classification includes: Averaging the training representation vectors of the industrial big data debugging samples corresponding to each state classification in the industrial big data debugging sample set to obtain an initial state centroid representation vector for each state classification; Determine the similarity measure between each training representation vector corresponding to each state classification and the initial state centroid representation vector of the state classification, and obtain the influence coefficient of each training representation vector corresponding to each state classification; Performing influence adjustment calculation on each training representation vector of the state classification according to the influence coefficient of each training representation vector corresponding to each state classification and completing the average to obtain the optimized state centroid representation vector of each state classification; Performing a same-state grouping operation on each training representation vector corresponding to each state classification to obtain a same-state centroid representation vector of each state classification; An integration operation is performed on the optimized state centroid representation vector of each state classification and the same-state centroid representation vector of the state classification to obtain a state classification representation vector of each state classification.

4. The method according to claim 1, characterized in that: The method of performing device state inference on the industrial big data set to be analyzed based on the target state classification characterization vector of each state classification in the set number of state classifications and the multi-task state recognition component based on the target device state classification network to obtain first state inference information and second state inference information includes: Extracting a representation vector of the industrial big data set to be analyzed according to the representation vector extraction component of the multi-task state recognition component in the target device state classification network to obtain a representation vector of the industrial big data set to be analyzed; Determine the commonality coefficient between the characterization vector of the industrial big data set to be analyzed and each of the target state classification characterization vectors according to the state reasoning evaluation component of the multi-task state recognition component in the target device state classification network, and obtain first state reasoning information; According to the normalized mapping reasoning component of the multi-task state recognition component in the target device state classification network, the confidence of the characterization vector of the industrial big data set to be analyzed corresponding to each of the state classifications is determined to obtain second state reasoning information.

5. The method according to claim 4, characterized in that The target industrial big data debugging sample set is obtained based on a supporting data set including the industrial big data set samples of the set number of state classifications; the metric integration component based on the target device state classification network performs a metric integration operation on the first state reasoning information and the second state reasoning information to obtain the integrated state reasoning information, including: According to the number of industrial big data set samples corresponding to each state classification in the support data set, determine the first state reasoning information influence coefficient and the second state reasoning information influence coefficient; the first state reasoning information influence coefficient is negatively correlated with the number of industrial big data set samples corresponding to each state classification, and the second state reasoning information influence coefficient is positively correlated with the number of industrial big data set samples corresponding to each state classification, and the sum of the first state reasoning information influence coefficient and the second state reasoning information influence coefficient is the set value; multiplying the first state reasoning information influence coefficient by the first state reasoning information to obtain target first state reasoning information; and multiplying the second state reasoning information influence coefficient by the second state reasoning information to obtain target second state reasoning information; The target first state reasoning information and the target second state reasoning information are loaded into the metric integration component of the target device state classification network to perform a metric integration operation to obtain integrated state reasoning information.

6. The method according to claim 4 or 5, characterized in that: The obtaining of the industrial big data set to be analyzed comprises: Acquire an initial industrial big data set, perform region of interest recognition processing on the initial industrial big data set, and obtain a region of interest marking window and region of interest recognition state classification information corresponding to the region of interest marking window; Acquire the industrial data cluster corresponding to the region of interest annotation window from the initial industrial big data set to obtain the industrial big data set to be analyzed; Determining target state reasoning information corresponding to the industrial big data set to be analyzed based on the integrated state reasoning information includes: The target state reasoning information is determined according to the ROI recognition state classification information corresponding to the ROI annotation window and the integrated state reasoning information.

7. An industrial big data simulation system, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 6 are implemented.

Images