Method, device and storage medium for detecting abnormal operation status of mold opening equipment

Abstract

The present application relates to the field of data processing technology, and provides a method, device and storage medium for detecting abnormal operation status of mold opening equipment. By acquiring and analyzing the operation status log of the mold opening equipment in real time, abnormal detection is effectively performed, thereby achieving accurate grasp of the operation status of the equipment. This process utilizes a special operation status abnormal detection network, and generates abnormal state discrimination viewpoints by deeply analyzing the status log fields, further improving the accuracy of abnormal detection. Moreover, the solution can not only identify whether there is an abnormality in the equipment, but also determine the specific abnormality type, such as excessive temperature, excessive speed, etc. Based on these detailed abnormal labels, the solution can generate targeted equipment operation and maintenance decision recommendations, thereby significantly improving the equipment's operating efficiency and production quality. Effectively integrating data collection, abnormality detection and decision recommendations provides a comprehensive and efficient solution for the operation and maintenance of mold opening equipment.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device and storage medium for detecting abnormal operation status of mold opening equipment. Background Art

[0002] Mold opening equipment is a key equipment for the production of various industrial products, especially in the field of energy-saving equipment, such as environmentally friendly and energy-saving ventilation equipment, waste gas treatment equipment, fire-proof and highly corrosion-resistant pipelines, etc. The role of mold opening equipment is indispensable in the manufacturing process. However, since these equipment involve many complex physical and chemical reactions during operation, their operating status is often difficult to accurately grasp directly through observation or measurement.

[0003] Traditional equipment maintenance methods mainly rely on manual regular inspections or simple sensor monitoring. However, these methods are time-consuming and labor-intensive, and due to the lack of in-depth analysis, they often fail to promptly discover and resolve hidden problems during equipment operation, such as abnormal conditions such as overheating and overspeeding. This not only affects the normal operation of the equipment, but may also cause equipment damage and even lead to safety accidents.

[0004] With the development of big data and artificial intelligence technology, a more intelligent, efficient and accurate solution is needed for abnormal detection of the operating status of mold opening equipment. Summary of the invention

[0005] In order to improve the above-mentioned problems, the present application provides a method, device and storage medium for detecting abnormal operation status of mold opening equipment.

[0006] In a first aspect, an embodiment of the present application provides a method for detecting abnormal operation status of a mold opening device, which is applied to an abnormal operation status detection device, and the method includes:

[0007] Obtain the operating status log of the mold opening equipment to be tested for abnormalities;

[0008] Using the operation status anomaly detection network, anomaly detection is performed on the operation status log of the mold opening equipment to be detected for anomaly, and an abnormal state discrimination view of the operation status log of the mold opening equipment to be detected for anomaly is generated; the abnormal state discrimination view represents the abnormal label corresponding to each state log field in the operation status log of the mold opening equipment to be detected for anomaly;

[0009] Based on the abnormal labels corresponding to the various status log fields in the operation status log of the mold opening equipment to be detected for abnormality, the equipment operation and maintenance decision suggestions of the operation status log of the mold opening equipment to be detected for abnormality are determined.

[0010] In a preferred embodiment of the first aspect, the debugging step of the operation status abnormality detection network includes:

[0011] Based on the state element vector mined by the state element mining subnet in the operation state anomaly detection network for the target operation state log, the state element vector chain of the target operation state log is determined; the operation state anomaly detection network is obtained by pre-debugging through the initial operation state log and the prior training knowledge of each state log field in the initial operation state log;

[0012] Optimizing the state elements according to the state element vector chain to obtain a state element optimization vector chain, wherein the state element optimization vector of the state log field in the state element optimization vector chain is obtained by optimizing the state elements of the context state log field of the state log field;

[0013] Determining the influence weight of each state log field in the target operation state log based on the state element vector chain and the state element optimization vector chain;

[0014] Based on the influence weight of each status log field in the target operation status log and the status text vector of each status log field in the status element vector chain, a key status log field set to be annotated for status log fields is selected in the target operation status log; the influence weight is used to adjust the status text vector difference between two status log fields;

[0015] Based on the priori training knowledge of the status log fields in the key status log field set and the target operating status log, transfer learning debugging is performed on the operating status anomaly detection network.

[0016] In a preferred embodiment of the first aspect, the selecting a set of key status log fields to be annotated with status log fields in the target operation status log based on the influence weight of each status log field in the target operation status log and the status text vector of each status log field in the status element vector chain includes:

[0017] Determine a first status log field set, wherein a first status log field in the first status log field set is a status log field in the target running status log;

[0018] For each second status log field in the second status log field set, determine in the first status log field set a first target status log field having the smallest status text vector difference with the second status log field; the status text vector difference is determined based on the status text vectors of the two status log fields; the second status log field in the second status log field set is a status log field in the target running status log, and there is no overlapping field between the second status log field set and the first status log field set;

[0019] By using the influence weight of the first target state log field corresponding to the second state log field, the state text vector difference between the second state log field and the corresponding first target state log field is adjusted to obtain the target difference between the second state log field and the first state log field set, wherein the target difference is in a set quantitative relationship with the influence weight;

[0020] Determine, in the second status log field set, a second target status log field having the largest target difference with the first status log field set;

[0021] adding the second target status log field to the first status log field set;

[0022] The key status log field set is determined based on the first status log field set that has been updated.

[0023] In a preferred embodiment of the first aspect, determining the key status log field set based on the first status log field set that has completed the update includes:

[0024] Recording the number of status log fields of the first status log field set that have completed updating;

[0025] If it is determined based on the number of the status log fields that the annotation condition is met, the first status log field set that has completed the update is used as the key status log field set;

[0026] If it is determined based on the number of status log fields that the annotation condition is not met, the second target status log field is deleted from the second status log field set to obtain an adjusted second status log field set, and the process jumps to each second status log field in the second status log field set, and the first target status log field with the smallest status text vector difference between the second status log field and the first status log field is determined in the first status log field set.

[0027] In a preferred embodiment of the first aspect, the operating state abnormality detection network further includes a first abnormality discrimination operator; for each second state log field in the second state log field set, before determining in the first state log field set a first target state log field having the smallest state text vector difference with the second state log field, the method further includes:

[0028] The first abnormal state discrimination operator performs abnormality detection according to the result of state element mining on the target operation state log by the state element mining subnet, and obtains a first abnormal state discrimination viewpoint, wherein the first abnormal state discrimination viewpoint represents the possibility that each state log field in the target operation state log belongs to each abnormal label of multiple abnormal labels;

[0029] For each state log field, determine the maximum X target possibilities for the state log field among the possibilities that the state log field belongs to multiple abnormal labels, where X is a positive integer;

[0030] Based on the X target possibilities determined for each state log field, an annotation index of each state log field is calculated;

[0031] Select the status log field whose annotation index is greater than the annotation index threshold from the target running status log, and generate the second status log field set according to the status log field whose annotation index is greater than the annotation index threshold.

[0032] In a preferred embodiment of the first aspect, X is 2, and the X target possibilities determined for a status log field include a maximum possibility and a second maximum possibility among possibilities that the status log field belongs to multiple abnormal tags; the annotation index of each status log field is obtained by calculation based on the X target possibilities determined for each status log field, including:

[0033] For each state log field, calculate and obtain the sum of the maximum possibility value and the second largest possibility value corresponding to the state log field, and obtain the reference possibility of the state log field;

[0034] For each status log field, the annotation index of the status log field is determined based on the difference result of subtracting the reference possibility corresponding to the status log field from the set constant.

[0035] In a preferred embodiment of the first aspect, the state element optimization is performed according to the state element vector chain to obtain the state element optimization vector chain, including:

[0036] Performing a regional feature extraction operation on the state element vector chain to obtain a regional state feature relationship network, wherein the state text vector value of the reference state log field of the feature extraction core of each round of feature extraction in the regional state feature relationship network is 0;

[0037] The regional state feature relationship network and the state element vector chain are integrated to obtain the state element optimization vector chain.

[0038] In a preferred embodiment of the first aspect, performing a regional feature extraction operation on the state element vector chain to obtain a regional state feature relationship network includes:

[0039] Performing sliding average processing on the state element vector chain to obtain a state element vector chain to be processed;

[0040] Combining the state element vector chain to be processed with the linear feature coverage vector corresponding to the feature extraction kernel to obtain a target state element vector chain;

[0041] The state element vectors in the target state element vector chain are subjected to quantized feature mapping to obtain the regional state feature relationship network.

[0042] In a preferred embodiment of the first aspect, determining the influence weight of each status log field in the target running status log based on the status element vector chain and the status element optimization vector chain includes:

[0043] For each status log field, obtain the status text vector of the status log field from the status element vector chain and obtain the status element optimization vector of the status log field from the status element optimization vector chain;

[0044] The vector similarity between the state text vector of the state log field and the state element optimization vector of the state log field is obtained by operation;

[0045] According to the vector similarity of the status log field, the influence weight of the status log field is determined.

[0046] In a preferred embodiment of the first aspect, the operation status anomaly detection network further includes a first anomaly discrimination operator; the state element optimization vector chain is obtained by optimizing the regional feature proposal model according to the state element vector chain; the prior training knowledge of the state log field in the key state log field set and the target operation status log are used to perform transfer learning debugging on the operation status anomaly detection network, including:

[0047] Obtaining a first abnormal state discrimination viewpoint obtained by performing anomaly detection on a state element vector mined by the state element mining subnet on a target operation state log by a first abnormal state discrimination operator;

[0048] Determining a first training cost function based on the first abnormal state discrimination viewpoint and the prior training knowledge of the state log fields in the key state log field set;

[0049] Determining a state element optimization training cost function based on the state element vector chain and the state element optimization vector chain;

[0050] Based on the first training cost function and the state element optimization training cost function, at least one variable of the operating state anomaly detection network and the regional feature proposal model is improved.

[0051] In a preferred embodiment of the first aspect, the state element vector chain is obtained by downsampling the first state element vector chain through a first residual model, and the first state element vector chain is the state element vector mined by the state element mining subnet for the target running state log;

[0052] Then the method further includes: performing anomaly detection based on the state element vector chain by a second anomaly discrimination operator to obtain a second abnormal state discrimination viewpoint; determining a second training cost function based on the second abnormal state discrimination viewpoint and prior training knowledge of the state log field in the key state log field set;

[0053] Then, based on the first training cost function and the state element optimization training cost function, improving the variables of at least one of the operating status anomaly detection network and the regional feature proposal model, includes: if the number of loops is less than a first preset number, determining the target training cost function based on the first training cost function, the second training cost function and the state element optimization training cost function; based on the target training cost function, improving the variables of the operating status anomaly detection network, the regional feature proposal model, the first residual model and the second anomaly discrimination operator.

[0054] In a preferred embodiment of the first aspect, the optimizing the training cost function based on the first training cost function and the state element to improve at least one variable of the operating state anomaly detection network and the regional feature proposal model further includes:

[0055] If the number of cycles is not less than a first preset number, improving the variables of the running state abnormality detection network based on the first training cost function;

[0056] And based on the second training cost function and the weighted cost function of the state element optimization training cost function, the variables of the regional feature proposal model, the first residual model and the second anomaly discrimination operator are adjusted.

[0057] In a preferred embodiment of the first aspect, the method further includes: performing transfer learning debugging on the operation status anomaly detection network according to the first initial operation status log and the learning annotation of the first initial operation status log.

[0058] In a preferred embodiment of the first aspect, performing transfer learning debugging on the operation status anomaly detection network according to the first initial operation status log and the learning annotation of the first initial operation status log includes:

[0059] The state element mining subnet performs state element mining on the first initial operation state log to obtain a first state element vector chain of the first initial operation state log;

[0060] The first abnormal state discrimination operator performs abnormality detection according to the first state element vector chain of the first initial operation state log to obtain a third abnormal state discrimination viewpoint;

[0061] Determine a first training cost function for the first initial operating state log based on the prior training knowledge of the first initial operating state log and the third abnormal state discrimination viewpoint;

[0062] Determining a state element vector chain of the first initial running state log according to the first state element vector chain of the first initial running state log;

[0063] Performing state element optimization on the state element vector chain of the first initial operation state log by using the regional feature proposal model to obtain a state element optimized vector chain of the first initial operation state log;

[0064] Determine a state element optimization training cost function for the first initial running state log based on the state element vector chain of the first initial running state log and the state element optimization vector chain of the first initial running state log;

[0065] Determine a third training cost function for the first initial running state log based on the possibility that each state log field belongs to a priori abnormal label and the possibility that each state log field belongs to a noise abnormal label in the third abnormal state discrimination viewpoint;

[0066] According to the first training cost function, the state element optimization training cost function and the third training cost function for the first initial running state log, the variables of at least one of the running state anomaly detection network and the regional feature proposal model are improved.

[0067] In a second aspect, an embodiment of the present application provides an abnormal operation status detection device, comprising at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the method described in the first aspect.

[0068] In a third aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method described in the first aspect when run.

[0069] In the embodiment of the present application, it is possible to perform real-time and accurate mold opening equipment anomaly detection and generate equipment operation and maintenance decision suggestions. In detail, by obtaining the operation status log of the mold opening equipment to be detected for anomaly and analyzing it using the operation status anomaly detection network, the solution can identify various possible equipment anomalies in real time. The operation status anomaly detection network performs in-depth analysis of each status log field of the operation log and generates an abnormal state discrimination viewpoint, that is, the corresponding abnormal label, which greatly improves the accuracy of anomaly detection.

[0070] Furthermore, the solution can not only identify whether there is an abnormality in the equipment, but also determine the specific type of abnormality, such as over-temperature, over-speed, etc., which is crucial for subsequent troubleshooting and repair. At the same time, based on the acquired abnormal labels, the solution can generate targeted equipment operation and maintenance decision suggestions, providing strong decision support for equipment maintenance personnel, thereby greatly improving the equipment's operating efficiency and production quality.

[0071] In general, this technical solution effectively integrates the processes of data collection, anomaly detection and decision-making recommendations, providing a comprehensive and efficient solution for the operation and maintenance of mold opening equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A flowchart of a method for detecting abnormal operating status of a mold opening device provided in an embodiment of the present application.

[0073] Figure 2 A schematic diagram of the structure of an abnormal operation status detection device 200 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0074] In order to better understand the above technical scheme, the technical scheme of the present application is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present application and the specific features in the embodiments are detailed descriptions of the technical scheme of the present application, rather than limitations on the technical scheme of the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments can be combined with each other.

[0075] Figure 1A method for detecting abnormal operation status of a mold opening device is shown, which is applied to an abnormal operation status detection device. The method includes the following steps 110 to 130.

[0076] Step 110: Obtain the operation status log of the mold opening equipment to be tested for abnormality.

[0077] Step 120: Use the operation status anomaly detection network to perform anomaly detection on the operation status log of the mold opening equipment that is to be detected for anomaly, and generate an abnormal state judgment view of the operation status log of the mold opening equipment that is to be detected for anomaly. The abnormal state judgment view represents the abnormal labels corresponding to each status log field in the operation status log of the mold opening equipment that is to be detected for anomaly.

[0078] Step 130: Based on the abnormal labels corresponding to the various status log fields in the operation status log of the mold opening equipment to be detected for abnormality, determine the equipment operation and maintenance decision suggestions for the operation status log of the mold opening equipment to be detected for abnormality.

[0079] In the above technical solution, the mold opening equipment operation status log is the record information generated by the equipment during operation, which usually includes various operation parameters of the equipment (such as temperature, pressure, speed, etc.), as well as timestamp and other information. For example, an operation status log may be as follows: "2022-03-01 10:00:00, temperature: 80 degrees, pressure: 5MPa, speed: 300rpm".

[0080] The operation status anomaly detection network is a model or algorithm used to detect whether the operation status of the device is normal, usually based on machine learning or deep learning technology. For example, a deep neural network may be used as an anomaly detection network, input the device operation status log, and output the anomaly label of each status log field.

[0081] Anomaly detection is the process of inputting the operation status log into the anomaly detection network to obtain the anomaly label of each state log field. For example, the above operation status log may be input into the anomaly detection network, and the result may be: "Temperature: normal, pressure: normal, speed: abnormal".

[0082] The abnormal state discrimination view is the judgment result of the anomaly detection network on whether each state log field is abnormal, usually expressed as a series of abnormal labels. For example, "temperature: normal, pressure: normal, speed: abnormal" is an abnormal state discrimination view.

[0083] The status log fields are the specific parameters in the operation status log, such as temperature, pressure, speed, etc. Each field corresponds to a value, indicating the specific operation status of the device at a certain moment.

[0084] The abnormal label is the judgment result of the anomaly detection network on whether each state log field is abnormal, usually expressed as "normal" or "abnormal". For example, "normal" and "abnormal" in "temperature: normal, pressure: normal, speed: abnormal" are abnormal labels.

[0085] Equipment operation and maintenance decision suggestions are suggestions for equipment operation and maintenance based on the abnormal state judgment point of view. For example, if the speed is abnormal, the equipment operation and maintenance decision suggestion may be: "Check and adjust the operating speed of the equipment."

[0086] In some more detailed examples, the power of deep learning can be used to deal with this problem. One exemplary idea is to use an autoencoder for anomaly detection.

[0087] An autoencoder is an unsupervised learning neural network whose goal is to encode the input data through a hidden layer (also called the encoding layer) and then decode it so that the output data is as close to the input data as possible. Under normal circumstances, the autoencoder can learn the main features of the data and reconstruct the input data well.

[0088] However, when the input data contains anomalies, since the autoencoder has not learned the features of these anomalies, it will produce large errors when reconstructing these anomaly data. Therefore, anomalies can be detected by comparing the difference between the input data and the reconstructed data. If the difference exceeds a certain threshold, then the input data is considered to contain anomalies.

[0089] For example, for the operating status log of the mold opening equipment, various status log fields (such as temperature, pressure, speed, etc.) can be extracted as input data to train an autoencoder. During the training process, the autoencoder will learn the characteristics of the normal operating status.

[0090] Then, the new running status log can be input into the trained autoencoder to obtain the reconstructed running status log. The difference between the original log and the reconstructed log is compared. If the difference exceeds the threshold, then the running status log is considered to contain anomalies.

[0091] In this way, the operating status of the mold opening equipment can be monitored in real time, and an early warning can be issued when an abnormality is detected, so as to discover and solve the problem in advance.

[0092] In other more detailed examples, anomaly detection can also be achieved through a neural network model, namely a Long Short-Term Memory (LSTM) network.

[0093] LSTM is a special recurrent neural network (RNN) that is suitable for processing and predicting important events with very long intervals and delays in time series. In the abnormal operation status detection of mold opening equipment, the operation log of the equipment can be regarded as a time series data, and then LSTM can be used to learn and predict the normal operation mode of the equipment.

[0094] Specifically, various status log fields (such as temperature, pressure, speed, etc.) can be extracted from the historical operation status log and organized into a sequence in chronological order. Then, an LSTM model can be trained to predict future status based on past status.

[0095] When using LSTM for anomaly detection, the new running status log is input into the trained LSTM model to obtain the predicted next state. Then, the actual next state is compared with the predicted next state. If the difference between the two exceeds a certain threshold, then the running status log is considered to contain anomalies.

[0096] For example, if the LSTM model predicts that the temperature at the next moment should be 80 degrees, but the actual temperature is 90 degrees, and this difference exceeds the set threshold, then it can be considered that the device may have an abnormality.

[0097] In this way, not only can the operating status of the mold opening equipment be monitored in real time, but also possible abnormalities of the equipment in the future can be predicted, so that measures can be taken in advance to avoid problems.

[0098] In some possible application scenarios, the abnormal label is not necessarily a binary classification of normal / abnormal. In more complex scenarios, more detailed classification of abnormalities may be required to more accurately identify and handle problems.

[0099] For example, for mold opening equipment, the following types of exception labels can be defined.

[0100] “Normal”: The device is operating normally and all parameters are within the normal range.

[0101] "Temperature Abnormal": The operating temperature of the device exceeds the normal range, which may be caused by a cooling system failure or other reasons.

[0102] "Abnormal pressure": The operating pressure of the equipment exceeds the normal range, which may be caused by hydraulic system failure or other reasons.

[0103] "Abnormal speed": The operating speed of the equipment exceeds the normal range, which may be caused by a drive system failure or other reasons.

[0104] "Multiple parameter abnormalities": Multiple parameters of the device exceed the normal range at the same time, which may mean that there are multiple problems with the device or a key component has failed.

[0105] In order to obtain these anomaly labels, a multi-classification model can be trained, such as a multi-class support vector machine (SVM), a multi-class logistic regression, a CNN or RNN in deep learning, etc. During the training process, each status log field in the historical log data needs to be input into the model together with its corresponding anomaly label, so that the model can learn various types of anomalies.

[0106] In actual operation, the new log data collected is input into the model. If the model determines that the equipment may be abnormal, it will output the corresponding abnormal label. Then, fault diagnosis and repair can be performed based on these abnormal labels. By defining different abnormal labels, equipment problems can be more accurately located and solved, thereby improving the equipment's operating efficiency and production quality.

[0107] In addition, for each anomaly type, more specific subcategories can be defined to describe and handle the problem more finely. Taking temperature anomaly as an example, the following subcategories can be defined.

[0108] "Overtemperature": The device's operating temperature exceeds the upper limit of the normal range, which may be caused by a cooling system failure or other reasons.

[0109] "Temperature is too low": The device's operating temperature is below the lower limit of the normal range, which may be caused by insufficient heat source or other reasons.

[0110] "Severe temperature fluctuations": The device's operating temperature fluctuates greatly over a short period of time, which may mean there is a problem with the device's temperature control system.

[0111] Temperature jump: The device's operating temperature suddenly increases or decreases and then returns to the normal range, which may be caused by a sensor failure or other temporary problem with the device.

[0112] In order to obtain these more specific anomaly labels, a multi-level classification model can be used, or cluster analysis can be performed on the anomalies. For example, one model can be used to determine whether there is an anomaly, and then another model can be used to determine the specific anomaly type.

[0113] The specific implementation method can be selected according to the actual situation. For example, a decision tree or random forest can be used for multi-level classification, or a clustering algorithm such as K-means or DBSCAN can be used to classify anomalies.

[0114] In actual operation, the new log data collected is input into the model. If the model determines that the equipment may be abnormal, it will output the corresponding abnormal label. Then, fault diagnosis and repair can be performed based on these abnormal labels. By defining more specific abnormal labels, equipment problems can be more accurately located and solved, thereby improving the equipment's operating efficiency and production quality.

[0115] Likewise, speed anomalies can also have multiple subcategories.

[0116] "Excessive speed": The equipment's operating speed exceeds the upper limit of the normal range, which may be caused by a drive system failure or a problem with the control system.

[0117] "Speed ​​too slow": The equipment's operating speed is below the lower limit of the normal range, which may be caused by insufficient driving force or excessive load.

[0118] "Severe speed fluctuations": The operating speed of the equipment fluctuates greatly up and down in a short period of time, which may mean that there is a problem with the speed control system of the equipment.

[0119] "Speed ​​jump": The device's operating speed suddenly increases or decreases and then returns to the normal range. This may be caused by a sensor failure or other temporary problem with the device.

[0120] For these specific anomaly labels, you can use a multi-level classification model or perform cluster analysis on the anomalies. For example, you can first use a model to determine whether there is an anomaly, and then use another model to determine the specific anomaly type. The specific implementation method can be selected according to the actual situation. For example, you can use decision trees or random forests for multi-level classification, or you can use clustering algorithms such as K-means or DBSCAN to classify anomalies.

[0121] In the field of energy-saving equipment, the manufacturing of products such as environmentally friendly and energy-saving ventilation equipment, exhaust gas treatment equipment, fire-proof and highly corrosion-resistant pipelines, etc. is inseparable from the relevant mold opening equipment. Therefore, in some examples, the mold opening equipment can be used as an example of a centrifugal fan mold opening equipment, and the operation status abnormality detection device is responsible for abnormality detection and equipment operation and maintenance decision recommendations.

[0122] At the beginning of step 110, the operation status abnormality detection device first obtains the operation status log of the centrifugal fan mold opening device to be detected for abnormality. These logs may contain various operation parameters such as the operation speed, temperature, vibration frequency and the corresponding timestamp of the device.

[0123] In step 120, the operation state abnormality detection device inputs these operation state logs into the trained operation state abnormality detection network for abnormality detection. This network can identify various abnormality types and generate a corresponding abnormality label for each abnormality. These abnormality labels constitute the abnormal state discrimination viewpoint, which represents the abnormality type corresponding to each state log field in the operation state log.

[0124] In step 130, the operation status abnormality detection device determines the equipment operation and maintenance decision suggestion according to the abnormality label corresponding to each status log field in the operation status log. For example, if the abnormality label indicates "temperature is too high", the equipment operation and maintenance decision suggestion may be "check and adjust the cooling system"; if the abnormality label indicates "speed is too fast", the equipment operation and maintenance decision suggestion may be "check and adjust the motor drive". In this way, the operation status abnormality detection device can provide specific and targeted operation and maintenance decision suggestions for the mold opening equipment of the centrifugal fan, so as to facilitate more accurate and effective fault diagnosis and repair.

[0125] It can be seen that by applying the above steps 110 to 130, it is possible to perform real-time and accurate mold opening equipment anomaly detection and generate equipment operation and maintenance decision suggestions. In detail, by obtaining the operation status log of the mold opening equipment to be detected for anomaly and analyzing it using the operation status anomaly detection network, the solution can identify various possible equipment anomalies in real time. The operation status anomaly detection network performs in-depth analysis on each status log field of the operation log and generates an abnormal state judgment point of view, that is, the corresponding abnormal label, which greatly improves the accuracy of anomaly detection.

[0126] Furthermore, the solution can not only identify whether there is an abnormality in the equipment, but also determine the specific type of abnormality, such as over-temperature, over-speed, etc., which is crucial for subsequent troubleshooting and repair. At the same time, based on the acquired abnormal labels, the solution can generate targeted equipment operation and maintenance decision suggestions, providing strong decision support for equipment maintenance personnel, thereby greatly improving the equipment's operating efficiency and production quality.

[0127] In general, this technical solution effectively integrates the processes of data collection, anomaly detection and decision-making recommendations, providing a comprehensive and efficient solution for the operation and maintenance of mold opening equipment.

[0128] In some possible embodiments, the debugging steps of the operation status abnormality detection network include steps 210 to 250.

[0129] Step 210: Determine the state element vector chain of the target running status log based on the state element vector mined by the state element mining subnet in the running status anomaly detection network; the running status anomaly detection network is pre-debugged through the initial running status log and the prior training knowledge of each state log field in the initial running status log.

[0130] Step 220: Optimize the state elements according to the state element vector chain to obtain a state element optimization vector chain, wherein the state element optimization vector of the state log field in the state element optimization vector chain is obtained by optimizing the state elements of the context state log field of the state log field.

[0131] Step 230: Determine the influence weight of each status log field in the target running status log based on the status element vector chain and the status element optimization vector chain.

[0132] Step 240: Based on the impact weight of each status log field in the target operation status log and the status text vector of each status log field in the status element vector chain, select a key status log field set to be annotated in the target operation status log; the impact weight is used to adjust the status text vector difference between two status log fields.

[0133] Step 250: Based on the prior training knowledge of the status log fields in the key status log field set and the target operating status log, transfer learning debugging is performed on the operating status anomaly detection network.

[0134] In this technical solution, some new technical terms are first introduced.

[0135] The state element mining subnet is a component of the operation state anomaly detection network. It is responsible for mining key state elements from the operation state log, that is, important parameters that affect the operation state of the device. The state element vector chain is a vector chain composed of state elements mined by the state element mining subnet. Each state element vector represents the characteristics of a state log field. Initial operation state log and target operation state log: The initial operation state log is the original data used to train the network, while the target operation state log is the new data for anomaly detection. Prior training knowledge is the network parameter and structure information obtained by training the initial operation state log, which can be used for pre-debugging and transfer learning debugging. State element optimization is an optimization method that optimizes the state element vector chain by considering the context state log field to generate the state element optimization vector chain. The influence weight is determined based on the state element vector chain and the state element optimization vector chain, which is used to measure the influence of each state log field on the operation state of the device. The key state log field set is a set of important state log fields to be annotated for the state log field, which is selected based on the influence weight and the difference of the state text vector. Transfer learning debugging is a debugging method that uses prior training knowledge and target running status logs to adjust the network so that it can better adapt to new data.

[0136] For example, the open-molding equipment of the exhaust gas treatment equipment is used as the target equipment. The operation status anomaly detection network has been pre-debugged with the initial operation status log and the prior training knowledge of each status log field.

[0137] At the beginning of step 210, the state element mining subnet in the operation state anomaly detection network processes the target operation state log. For example, a dimensionality reduction technique such as principal component analysis (PCA) or autoencoder may be used to convert multiple state parameters in each log into a state element vector, thereby forming a state element vector chain.

[0138] In step 220, the state element optimization is performed according to the state element vector chain to obtain the state element optimized vector chain. This process may involve some deep learning methods, such as long short-term memory (LSTM) network, to consider the context information of the state log field.

[0139] Step 230 determines the influence weight of each state log field in the target operation state log based on the state element vector chain and the state element optimization vector chain. For example, algorithms such as support vector machine (SVM) or decision tree can be used for feature selection to determine the influence weight of each state log field.

[0140] In step 240, based on the impact weight and the state element vector chain, a set of key state log fields to be annotated are selected from the target operation state log. This process can be achieved by comparing the state text vector differences between different state log fields.

[0141] Finally, in step 250, based on the prior training knowledge of the state log fields in the key state log field set and the target operation state log, the operation state anomaly detection network is transferred learning and debugged. For example, a pre-trained neural network model (such as a convolutional neural network CNN or a recurrent neural network RNN) can be used to fine-tune its parameters to adapt it to the new task, thereby completing the transfer learning and debugging.

[0142] In this technical solution, by using technologies such as state element mining subnets, state element optimization, and influence weights, the key factors affecting the operating status of the equipment can be more accurately identified, and targeted equipment operation and maintenance decision recommendations can be generated. At the same time, through transfer learning debugging, the network can be adjusted to an optimal state, thereby improving the efficiency and accuracy of anomaly detection. In addition, this technical solution also emphasizes the importance of contextual information, that is, the state of a status log field may be affected by other status log fields around it. By considering contextual information, the operating status of the equipment can be more comprehensively understood, thereby making more accurate anomaly judgments and equipment operation and maintenance decisions. In general, this technical solution achieves deeper and more comprehensive anomaly detection and equipment operation and maintenance decisions by introducing new technical terms and methods, and has higher technical effects.

[0143] In some other possible embodiments, step 240 described above selects a set of key status log fields to be annotated with status log fields in the target running status log based on the impact weight of each status log field in the target running status log and the status text vector of each status log field in the status element vector chain, including steps 241-244.

[0144] Step 241: Determine a first status log field set, where the first status log field in the first status log field set is a status log field in the target running status log.

[0145] Step 242: for each second status log field in the second status log field set, determine in the first status log field set a first target status log field having the smallest status text vector difference with the second status log field; the status text vector difference is determined based on the status text vectors of the two status log fields; the second status log field in the second status log field set is a status log field in the target running status log, and there is no overlapping field between the second status log field set and the first status log field set.

[0146] Step 243: Adjust the status text vector difference between the second status log field and the corresponding first target status log field through the influence weight of the first target status log field corresponding to the second status log field, and obtain the target difference between the second status log field and the first status log field set, wherein the target difference is in a set quantitative relationship with the influence weight.

[0147] Step 244, determining in the second status log field set a second target status log field having the largest target difference with the first status log field set; adding the second target status log field to the first status log field set; and determining the key status log field set based on the updated first status log field set.

[0148] In this example, the specific implementation of steps 241 to 244 is introduced by still taking the mold opening equipment of the exhaust gas treatment equipment as an example.

[0149] At the beginning of step 241, a first state log field set is determined, which is a state log field in the target operation state log. For example, the first state log field set may include parameters such as temperature, pressure, and operation speed of the device.

[0150] In step 242, for each second status log field in the second status log field set, a first target status log field having the smallest status text vector difference with the second status log field is determined in the first status log field set. The second status log field set is also selected from the target operation status log and has no overlapping fields with the first status log field set. The status text vector difference can be determined by calculating the Euclidean distance or cosine similarity of the status text vectors of the two status log fields.

[0151] In step 243, the state text vector difference between the second state log field and the corresponding first target state log field is adjusted by the influence weight of the first target state log field corresponding to the second state log field, so as to obtain the target difference between the second state log field and the first state log field set. The influence weight may be calculated based on information such as the state element vector chain and the state element optimization vector chain.

[0152] Finally, in step 244, the second target state log field with the largest target difference from the first state log field set is determined from the second state log field set, and added to the first state log field set, and the key state log field set is determined based on the updated first state log field set. This process can be achieved by methods such as sorting or threshold selection.

[0153] The technical effect of this technical solution is mainly reflected in the more refined status log field analysis and selection. It can not only find out the key status log fields that affect the operating status of the equipment, but also optimize them according to the impact weights of these fields and the differences in status text vectors, further improving the accuracy of anomaly detection and the effectiveness of equipment operation and maintenance decisions. In addition, by dynamically updating the status log field set, the solution can also better adapt to changes in the operating status of the equipment, and has strong practicality and adaptability.

[0154] In some preferred embodiments, the key status log field set is determined based on the first status log field set that has been updated in step 244, including: recording the number of status log fields of the first status log field set that has been updated; if it is determined based on the number of status log fields that the annotation condition is met, the first status log field set that has been updated is used as the key status log field set; if it is determined based on the number of status log fields that the annotation condition is not met, the second target status log field is deleted from the second status log field set to obtain an adjusted second status log field set, and jumps to each second status log field in the second status log field set, and determines in the first status log field set the first target status log field with the smallest status text vector difference with the second status log field.

[0155] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the process of determining the key status log field set in step 244 is continued to be introduced.

[0156] When this process starts, the number of status log fields in the first status log field set that has completed the update is first recorded. For example, if there are 10 status log fields in the first status log field set, the number of status log fields recorded is 10.

[0157] Next, it is determined whether the annotation condition is met according to the number of status log fields. The annotation condition may be a preset threshold or ratio. For example, if the number of status log fields reaches 20, or reaches 50% of all status log fields, the annotation condition is considered to be met.

[0158] If the annotation condition is met, the first set of status log fields that have been updated is regarded as the key set of status log fields. This means that these status log fields are considered to have the greatest impact on the device operation status and should be given priority for anomaly detection and device operation and maintenance decisions.

[0159] If the annotation condition is not met, the second target status log field is deleted from the second status log field set to obtain an adjusted second status log field set, and then jump to each second status log field in the second status log field set, determine the first target status log field with the smallest status text vector difference with the second status log field in the first status log field set, and then repeat the previous steps.

[0160] The technical effect of this technical solution is mainly reflected in more flexible status log field selection and more refined anomaly detection. It can not only dynamically update the status log field set to adapt to changes in the operating status of the device, but also flexibly determine the key status log field set according to the preset annotation conditions, thereby achieving more accurate and comprehensive anomaly detection. In addition, by deleting the second target status log field and jumping to the next round of judgment, the solution can also avoid excessive analysis of irrelevant or minor status log fields, thereby improving the efficiency of anomaly detection.

[0161] In some examples, the operation state abnormality detection network further includes a first abnormality discrimination operator. Then, for each second state log field in the second state log field set, before determining the first target state log field in the first state log field set with the smallest state text vector difference with the second state log field, the method further includes steps 310 to 340.

[0162] Step 310: The first abnormal state discrimination operator performs anomaly detection based on the result of state element mining on the target operation state log by the state element mining subnet to obtain a first abnormal state discrimination viewpoint, wherein the first abnormal state discrimination viewpoint represents the possibility that each state log field in the target operation state log belongs to each abnormal label in multiple abnormal labels.

[0163] Step 320: For each status log field, determine the maximum X target possibilities for the status log field among the possibilities that the status log field belongs to multiple abnormal labels, where X is a positive integer.

[0164] Step 330: Obtain an annotation index for each status log field based on the X target possibilities determined for each status log field.

[0165] Step 340: Select the status log fields whose annotation index is greater than the annotation index threshold from the target running status log, and generate the second status log field set according to the status log fields whose annotation index is greater than the annotation index threshold.

[0166] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of step 310 to step 340 is further introduced.

[0167] First, we need to explain the newly emerged technical terms: The first anomaly discrimination operator is a component of the running state anomaly detection network. It is responsible for anomaly detection based on the results of the state element mining subnet and generates the first abnormal state discrimination view. The anomaly discrimination operator can be constructed using a classifier algorithm or a decision tree algorithm. The annotation index is an index calculated based on the target probability of each state log field, which is used to evaluate the importance of the state log field. The annotation index threshold is a preset threshold. Only when the annotation index of the state log field is greater than this threshold will it be selected into the second state log field set.

[0168] In step 310, the first abnormality discrimination operator performs abnormality detection based on the result of state element mining performed by the state element mining subnet on the target operation state log, and obtains a first abnormal state discrimination view. For example, if a deep neural network is used as the first abnormality discrimination operator, it may output a probability distribution indicating the possibility that each state log field belongs to multiple abnormal labels.

[0169] In step 320, for each state log field, the maximum X target possibilities are determined for the state log field among the possibilities that the state log field belongs to multiple abnormal labels. This process can be implemented by methods such as sorting or threshold selection.

[0170] In step 330, the annotation index of each state log field is calculated based on the X target possibilities determined for each state log field. For example, the X target possibilities can be added or averaged to obtain the annotation index.

[0171] Finally, in step 340, status log fields whose annotation index is greater than the annotation index threshold are selected from the target running status log, and a second status log field set is generated based on these fields.

[0172] The technical effect of this technical solution is mainly reflected in more accurate and comprehensive anomaly detection. By using the first anomaly discrimination operator, the solution can perform more in-depth anomaly detection based on the results of state element mining and generate a first abnormal state discrimination view with detailed anomaly information. In addition, by calculating the annotation index and setting the annotation index threshold, the solution can effectively filter out important state log fields, further improving the accuracy and efficiency of anomaly detection.

[0173] In some examples, X is 2, and the X target possibilities determined for a state log field include the maximum possibility and the second largest possibility among the possibilities that the state log field belongs to multiple abnormal tags. Then, the annotation index of each state log field is calculated based on the X target possibilities determined for each state log field, as described in step 330, including steps 331-332.

[0174] Step 331: For each status log field, calculate and obtain the sum of the maximum possibility value and the second largest possibility value corresponding to the status log field to obtain the reference possibility of the status log field.

[0175] Step 332: For each status log field, determine the annotation index of the status log field based on the difference result of subtracting the reference possibility corresponding to the status log field from the set constant.

[0176] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of step 331 to step 332 is continued to be introduced.

[0177] First, in this example, X is set to 2, which means that the maximum and second largest values ​​of the possibility that each status log field belongs to multiple abnormal labels are focused on. These two values ​​will serve as the basis for the subsequent calculation of the annotation index.

[0178] In step 331, for each status log field, the sum of the corresponding maximum and second-largest likelihood values ​​is calculated to obtain the reference likelihood of the status log field. For example, if the likelihood of a status log field belonging to the "overheating" abnormality tag is 0.7 (maximum likelihood), and the likelihood of belonging to the "excessive current" abnormality tag is 0.2 (second-largest likelihood), then its reference likelihood is 0.7+0.2=0.9.

[0179] Then, in step 332, the annotation index of the status log field is determined based on the difference between the reference probability corresponding to each status log field and the set constant. Assuming the set constant is 0.5, the annotation index of the status log field in the previous example is 0.9-0.5=0.4.

[0180] The technical effect of this technical solution is mainly reflected in the more accurate evaluation of the importance of the status log field. By only focusing on the maximum and second largest values ​​of the possibility that each status log field belongs to the abnormal label, it is possible to more attentively analyze the most likely abnormalities, thereby improving the accuracy of anomaly detection. In addition, by introducing set constants and calculating the difference results, the calculation method of the annotation index can be flexibly adjusted to make it more adaptable to different anomaly detection needs. This method also helps to reduce excessive attention to unimportant status log fields, further improving the efficiency of anomaly detection.

[0181] Under some design ideas, the state element optimization is performed according to the state element vector chain in step 220 to obtain the state element optimization vector chain, including steps 221 and 222.

[0182] Step 221: Perform a regional feature extraction operation on the state element vector chain to obtain a regional state feature relationship network, wherein the state text vector value of the baseline state log field of the feature extraction core of each round of feature extraction in the regional state feature relationship network is 0.

[0183] Step 222: Integrate the regional state feature relationship network and the state element vector chain to obtain the state element optimization vector chain.

[0184] In some preferred embodiments, the step 221 performs a regional feature extraction operation on the state element vector chain to obtain a regional state feature relationship network, including steps 2211 to 2213.

[0185] Step 2211: Perform sliding average processing on the state element vector chain to obtain the state element vector chain to be processed.

[0186] Step 2212: Combine the state element vector chain to be processed with the linear feature coverage vector corresponding to the feature extraction kernel to obtain the target state element vector chain.

[0187] Step 2213: Perform quantized feature mapping on the state element vectors in the target state element vector chain to obtain the regional state feature relationship network.

[0188] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of step 221 to step 222 and its sub-steps will be continued to be introduced.

[0189] First, we need to explain the emerging technical terms: Regional feature extraction is a machine learning method that extracts relevant or important features from the state element vector chain and generates a regional state feature relationship network containing these features. Sliding average processing is a data preprocessing technique that smoothes the data by calculating the average value of each state element vector over a period of time, which helps to reduce data noise and fluctuations. Linear feature coverage vector is a tool for feature extraction that can map the state element vector chain to be processed to a new feature space to help identify or extract more valuable features. Quantitative feature mapping is a feature processing method that obtains a set of discrete and comparable features by quantizing the state element vector.

[0190] In step 221, a regional feature extraction operation is performed on the state element vector chain to obtain a regional state feature relationship network. In the network, the state text vector value of the baseline state log field of the feature extraction core of each round of feature extraction is 0, which means that in each round of feature extraction, a certain state log field is used as a benchmark for comparison.

[0191] Then, in step 222, the regional state feature relationship network and the state element vector chain are integrated to obtain the state element optimization vector chain. This step fuses the original state element vector chain with the regional state feature relationship network obtained by feature extraction to generate a more representative state element optimization vector chain containing richer information.

[0192] In some preferred embodiments, step 221 can be further refined into steps 2211 to 2213.

[0193] In step 2211, the state element vector chain is subjected to sliding average processing to obtain the state element vector chain to be processed. Then, in step 2212, the state element vector chain to be processed is combined with the linear feature coverage vector corresponding to the feature extraction kernel to obtain the target state element vector chain. Finally, in step 2213, the state element vectors in the target state element vector chain are subjected to quantitative feature mapping to obtain the regional state feature relationship network.

[0194] Exemplarily, the main calculation steps involved in the above technical solution can be expressed by the following formula.

[0195] In step 2211, a sliding average process is performed on the state element vector chain. If the state element vector chain is represented as V = {v1, v2, ..., vn}, then the sliding average process can be represented as: V' = {SMA (v1), SMA (v2), ..., SMA (vn)};

[0196] Among them, SMA(vi) represents the sliding average of vi within a time window.

[0197] In step 2212, the state element vector chain to be processed is combined with the linear feature coverage vector corresponding to the feature extraction kernel. Assuming that the linear feature coverage vector is represented by L = {l1, l2, ..., ln}, then this step can be expressed as: V" = {combine (V', L)};

[0198] Wherein, combine(V', L) represents the operation of combining the state element vector chain to be processed and the linear feature coverage vector.

[0199] In step 2213, the state element vectors in the target state element vector chain are subjected to quantization feature mapping to obtain a regional state feature relationship network. This step can be expressed as: Q = {quantize(v”)|v”∈V”};

[0200] Among them, quantize(v”) represents the operation of performing quantized feature mapping on v”.

[0201] Finally, in step 222, the regional state feature relationship network and the state element vector chain are integrated to obtain the state element optimized vector chain. This step can be expressed as: V_optimized = integrate (Q, V);

[0202] Among them, integrate(Q, V) represents the operation of integrating the regional state feature relationship network and the state element vector chain.

[0203] The technical effect of this technical solution is mainly reflected in deeper feature extraction and more effective state factor optimization. Through methods such as regional feature extraction, sliding average processing, linear feature coverage vector combination and quantitative feature mapping, this solution can extract more valuable features from the original state factor vector chain and integrate these features into the state factor optimization vector chain, thereby achieving more accurate and comprehensive anomaly detection.

[0204] In some alternative embodiments, the step 230 described in determining the influence weight of each status log field in the target running status log based on the status element vector chain and the status element optimization vector chain includes steps 231 to 233.

[0205] Step 231: For each status log field, obtain the status text vector of the status log field from the status element vector chain and obtain the status element optimization vector of the status log field from the status element optimization vector chain.

[0206] Step 232: Calculate and obtain the vector similarity between the status text vector of the status log field and the status element optimization vector of the status log field.

[0207] Step 233: Determine the influence weight of the status log field according to the vector similarity of the status log field.

[0208] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of steps 231 to 233 is continued to be introduced.

[0209] First, we need to explain some new technical terms: vector similarity is a measure of the similarity between two vectors. There are many ways to calculate vector similarity, such as cosine similarity, Euclidean distance, etc.

[0210] In step 231, for each state log field, the state text vector of the state log field is obtained from the state element vector chain, and the state element optimization vector of the state log field is obtained from the state element optimization vector chain. For example, if the state log field of interest is "mold opening current", the state text vector and state element optimization vector corresponding to this field can be obtained from the two vector chains.

[0211] Then, in step 232, the vector similarity between the status text vector of the status log field and the status element optimization vector of the status log field is calculated. For example, the cosine similarity can be used to calculate the similarity of the two vectors: similarity = cos(theta) = (A·B) / (||A||||B||);

[0212] Among them, A and B represent the state text vector and the state element optimization vector respectively, and theta represents the angle between the two vectors.

[0213] Finally, in step 233, the influence weight of the status log field is determined according to the vector similarity of the status log field. For example, the similarity can be directly used as the weight, or the similarity can be converted into the weight through a certain mapping function.

[0214] The technical effect of this technical solution is mainly reflected in more accurately evaluating the impact weight of the status log field. By comparing the similarity between the status text vector and the status element optimization vector, the solution can take into account the information difference between the status log field in the original state and the optimized state, thereby more accurately determining its impact weight. This helps to improve the accuracy and robustness of anomaly detection.

[0215] In some examples, the operation status anomaly detection network further includes a first anomaly discrimination operator; the state element optimization vector chain is obtained by optimizing the regional feature proposal model according to the state element vector chain. Then, the operation status anomaly detection network is transferred and learned based on the prior training knowledge of the state log field in the key state log field set and the target operation status log as described in step 250, including steps 251-254.

[0216] Step 251: Obtain a first abnormal state discrimination viewpoint obtained by performing anomaly detection on a state element vector mined by the state element mining subnet on a target operation state log using a first abnormal state discrimination operator.

[0217] Step 252: Determine a first training cost function based on the first abnormal state discrimination viewpoint and the prior training knowledge of the state log fields in the key state log field set.

[0218] Step 253: Determine a state element optimization training cost function based on the state element vector chain and the state element optimization vector chain.

[0219] Step 254: Based on the first training cost function and the state factor optimization training cost function, improve at least one variable of the operating state anomaly detection network and the regional feature proposal model.

[0220] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of steps 251 to 254 is continued to be introduced.

[0221] First, we need to explain the new technical terms: Training cost function: also known as loss function or objective function, is the quantity that needs to be minimized (or maximized) in the optimization process of machine learning models. It reflects the gap between the model prediction results and the true value.

[0222] In step 251, a first abnormal state discrimination viewpoint is obtained by performing anomaly detection on the state element vector mined by the state element mining subnet on the target operation state log by the first abnormal state discrimination operator. For example, the first abnormal state discrimination operator may find an abnormality in the "opening current" field.

[0223] Then, in step 252, a first training cost function is determined based on the first abnormal state discrimination viewpoint and the prior training knowledge of the state log field in the key state log field set. For example, a binary cross entropy loss function can be constructed based on the first abnormal state discrimination viewpoint and the abnormal labels of the key state log fields.

[0224] Next, in step 253, a state element optimization training cost function is determined based on the state element vector chain and the state element optimization vector chain. This cost function can be used to measure the gap between the state element optimization vector chain and the target state, for example, a cross entropy loss function can also be used.

[0225] Finally, in step 254, the training cost function is optimized based on the first training cost function and the state elements to improve at least one variable in the running state anomaly detection network and the regional feature proposal model. For example, the model parameters can be adjusted by an optimization algorithm such as gradient descent to minimize the training cost function.

[0226] The technical effect of this technical solution is mainly reflected in more accurate transfer learning debugging. By combining multiple viewpoints and training cost functions, the solution can more comprehensively consider various factors, thereby more accurately adjusting the operating status anomaly detection network and regional feature proposal model. This helps to improve the accuracy and generalization ability of anomaly detection, while also making the model more adaptable and flexible.

[0227] In some examples, the state element vector chain is obtained by downsampling the first state element vector chain through the first residual model, and the first state element vector chain is the state element vector mined by the state element mining subnet for the target running state log. Then the method also includes steps 410-420.

[0228] Step 410: A second abnormal state discrimination operator performs abnormality detection based on the state element vector chain to obtain a second abnormal state discrimination viewpoint.

[0229] Step 420: Determine a second training cost function based on the second abnormal state discrimination viewpoint and the prior training knowledge of the state log fields in the key state log field set.

[0230] Then the step 254 described in which the training cost function is optimized based on the first training cost function and the state element to improve at least one variable in the operating state anomaly detection network and the regional feature proposal model includes steps 2541 and 2542.

[0231] Step 2541: If the number of cycles is less than a first preset number, the training cost function is optimized based on the first training cost function, the second training cost function and the state factor to determine a target training cost function.

[0232] Step 2542: Based on the target training cost function, improve the variables of the operating status anomaly detection network, the regional feature proposal model, the first residual model and the second anomaly discrimination operator.

[0233] In this example, still taking the mold opening equipment of the exhaust gas treatment equipment as an example, the specific implementation of steps 410 to 420 and steps 2541 to 2542 is continued to be introduced.

[0234] First, we need to explain the new technical terms: The second anomaly discrimination operator is similar to the first anomaly discrimination operator. It is also a tool or model for anomaly detection based on the state element vector chain. The first residual model is a model for downsampling the input state element vector chain to obtain a new state element vector chain. Downsampling is usually used to reduce the complexity of the data and the computational requirements.

[0235] In step 410, the second abnormality discrimination operator performs abnormality detection based on the state element vector chain to obtain a second abnormal state discrimination viewpoint. For example, the second abnormality discrimination operator may find an abnormality in the "cooling water flow" field.

[0236] Then, in step 420, a second training cost function is determined based on the second abnormal state discrimination viewpoint and the prior training knowledge of the state log field in the key state log field set. For example, a new binary cross entropy loss function can be constructed based on the second abnormal state discrimination viewpoint and the abnormal labels of the key state log fields.

[0237] Next, in step 2541, if the number of cycles is less than the first preset number, the target training cost function is determined based on the first training cost function, the second training cost function and the state factor optimization training cost function. For example, the average of the three training cost functions can be used as the target training cost function.

[0238] Finally, in step 2542, based on the target training cost function, the variables of the operating state anomaly detection network, the regional feature proposal model, the first residual model, and the second anomaly discrimination operator are improved. For example, the parameters of these models can be adjusted by an optimization algorithm such as gradient descent to minimize the target training cost function.

[0239] The technical effect of this technical solution is mainly reflected in more comprehensive and accurate transfer learning debugging. By introducing the second anomaly discrimination operator and the first residual model, the solution can more comprehensively consider various factors, thereby more accurately adjusting the operating state anomaly detection network, regional feature proposal model, etc. In addition, by setting the number of cycles, the solution can also flexibly adjust the training process according to actual needs, thereby improving the efficiency and stability of training.

[0240] In some preferred embodiments, the step 254 of optimizing the training cost function based on the first training cost function and the state element to improve the variables of at least one of the operating state anomaly detection network and the regional feature proposal model also includes: if the number of loops is not less than the first preset number of times, improving the variables of the operating state anomaly detection network based on the first training cost function; and adjusting the variables of the regional feature proposal model, the first residual model and the second abnormality discrimination operator based on the weighted cost function of the second training cost function and the state element optimization training cost function.

[0241] In this example, the specific implementation of step 254 is still introduced by taking the mold opening equipment of the exhaust gas treatment equipment as an example.

[0242] In a preferred embodiment, step 254 includes the following two parts: if the number of loops is not less than the first preset number of times, based on the first training cost function, the variables of the running state anomaly detection network are improved. For example, if the number of loops reaches the first preset number of times, only the first training cost function is used to adjust the parameters of the running state anomaly detection network. Based on the weighted cost function of the second training cost function and the state element optimization training cost function, the variables of the regional feature proposal model, the first residual model and the second anomaly discrimination operator are adjusted. For example, the weighted average of the second training cost function and the state element optimization training cost function can be used as a new target cost function, and then the parameters of these models are adjusted by optimization algorithms such as gradient descent to minimize the new target cost function.

[0243] The technical effect of this technical solution is mainly reflected in more flexible and effective transfer learning debugging. By selecting different training cost functions according to the number of cycles, the solution can flexibly adjust the training process, thereby improving the efficiency and stability of training. At the same time, by using a weighted cost function, the solution can more effectively integrate various training information, thereby more accurately adjusting model parameters. This helps to improve the accuracy and generalization ability of anomaly detection, while also making the model more adaptable and flexible.

[0244] In some other possible embodiments, the method further includes: performing transfer learning debugging on the operation status anomaly detection network according to the first initial operation status log and the learning annotation of the first initial operation status log.

[0245] In some other preferred embodiments, the operation status anomaly detection network is subjected to transfer learning debugging based on the first initial operation status log and the learning annotation of the first initial operation status log, including: the state element mining subnet performs state element mining on the first initial operation status log to obtain a first state element vector chain of the first initial operation status log; the first abnormality discrimination operator performs anomaly detection based on the first state element vector chain of the first initial operation status log to obtain a third abnormal state discrimination point of view; based on the prior training knowledge of the first initial operation status log and the third abnormal state discrimination point of view, determines a first training cost function for the first initial operation status log; determines the state element vector chain of the first initial operation status log based on the first state element vector chain of the first initial operation status log; and The feature proposal model optimizes the state elements of the state element vector chain of the first initial running state log to obtain the state element optimization vector chain of the first initial running state log; based on the state element vector chain of the first initial running state log and the state element optimization vector chain of the first initial running state log, determine the state element optimization training cost function for the first initial running state log; based on the possibility that each state log field belongs to the prior abnormal label and the possibility that it belongs to the noise abnormal label in the third abnormal state discrimination viewpoint, determine the third training cost function for the first initial running state log; according to the first training cost function, the state element optimization training cost function and the third training cost function for the first initial running state log, improve at least one variable in the running state anomaly detection network and the regional feature proposal model.

[0246] In this example, we still take the mold opening equipment of the exhaust gas treatment equipment as an example to continue to introduce the specific implementation of the above steps.

[0247] First, we need to explain the emerging technical terms: The first initial operating status log is the original data set for transfer learning, which contains a series of log information about the operating status of the mold opening equipment. Learning annotation is the process of annotating the original data set, which usually includes specifying one or more labels for each data sample. Here, learning annotation may include specifying whether each status log field has an anomaly. The third abnormal state discrimination viewpoint is the result of anomaly detection based on the first state element vector chain of the first initial operating status log.

[0248] Next, the following are specific examples of the above technical solution.

[0249] In this embodiment, the state element mining subnet is first used to mine the state elements of the first initial operation state log to obtain a first state element vector chain. Then, the first abnormality discrimination operator is used to perform abnormality detection based on the vector chain to obtain a third abnormal state discrimination viewpoint.

[0250] Based on the prior training knowledge of the first initial operation state log and the third abnormal state discrimination viewpoint, a first training cost function for the first initial operation state log is determined. At the same time, a state element vector chain needs to be determined according to the first state element vector chain of the first initial operation state log.

[0251] Next, the state element vector chain is optimized using the regional feature proposal model to obtain the state element optimized vector chain. Based on the optimized vector chain and the state element vector chain, a state element optimization training cost function for the first initial running state log is determined.

[0252] In addition, it is also necessary to determine a third training cost function for the first initial running state log based on the possibility that each state log field belongs to a priori abnormal label and the possibility that it belongs to a noise abnormal label in the third abnormal state discrimination viewpoint.

[0253] Finally, according to the first training cost function, the state factor optimization training cost function and the third training cost function, at least one variable in the operating state anomaly detection network and the regional feature proposal model is improved.

[0254] The technical effect of this technical solution is mainly reflected in that by introducing more training cost functions and more comprehensive training information, the running state anomaly detection network and the regional feature proposal model can more accurately adjust the model parameters, thereby improving the accuracy and generalization ability of anomaly detection. At the same time, through detailed analysis and processing of the initial running state log, the solution can also have a deeper understanding of data characteristics, thereby further optimizing model performance.

[0255] In addition, the process of the above technical solution can be explained by the following series of steps and related formulas.

[0256] State element mining: First, there is a state element mining subnet, which converts the first initial running state log into the first state element vector chain. If the function representing the state element mining subnet is f, and the first initial running state log is Log, then this process can be expressed as: V = f(Log);

[0257] Among them, V is the first state element vector chain obtained.

[0258] Anomaly detection: Then, the first anomaly discrimination operator takes V as input to perform anomaly detection and obtains the third abnormal state discrimination point of view. If the function representing the first anomaly discrimination operator is g, then this process can be expressed as: P = g(V);

[0259] Among them, P is the third abnormal state judgment viewpoint.

[0260] Determine the first training cost function: Based on the prior training knowledge K and P of the first initial running state log, the first training cost function J1 for the first initial running state log can be determined. If the function representing this process is h, then this process can be represented as: J1 = h(K, P);

[0261] State element optimization: Then, the regional feature proposal model optimizes the state elements of V and obtains the state element optimization vector chain V'. If the function representing the regional feature proposal model is m, then this process can be expressed as: V' = m(V);

[0262] Determine the state element optimization training cost function: Based on V and V', the state element optimization training cost function J2 for the first initial running state log can be determined. If the function representing this process is n, then this process can be represented as: J2 = n(V, V');

[0263] Determine the third training cost function: Based on the probability P_e of each state log field in P belonging to the prior abnormal label and the probability P_n of each state log field belonging to the noise abnormal label, the third training cost function J3 for the first initial operation state log can be determined. If the function representing this process is q, then this process can be represented as: J3 = q (P_e, P_n);

[0264] Improve model variables: Finally, according to J1, J2 and J3, at least one variable in the operation status anomaly detection network and regional feature proposal model can be improved. If the function representing this process is r, then this process can be represented as: r(J1, J2, J3);

[0265] In this way, by mining the state elements of the original log data, and then performing anomaly detection through the anomaly discrimination operator, and combining the prior training knowledge to determine the training cost function, and then improving the variables of at least one item in the running state anomaly detection network and the regional feature proposal model by optimizing the training cost function, efficient anomaly detection of the running state log can be achieved.

[0266] Figure 2 Schematic diagram of a running state abnormality detection device 200 provided in an embodiment of the present application. Figure 2 The operating status abnormality detection device 200 shown includes a processor 210, and the processor 210 can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0267] Alternatively, if Figure 2 As shown, the operation status abnormality detection device 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0268] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0269] Alternatively, if Figure 2 As shown, the abnormal operation status detection device 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices, specifically, may send information or data to other devices, or receive information or data sent by other devices.

[0270] Optionally, the operation status abnormality detection device 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they will not be repeated here.

[0271] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method embodiment can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and performed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0272] It can be understood that the memory in the embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile 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), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct RAM bus random access memory (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0273] It should be understood that the above-mentioned memory is exemplary but not restrictive. For example, the memory in the embodiments of the present application may also be static random access memory (static RAM, SRAM), dynamic random access memory (dynamic RAM, DRAM), synchronous dynamic random access memory (synchronous DRAM, SDRAM), double data rate synchronous dynamic random access memory (double data rate SDRAM, DDR SDRAM), enhanced synchronous dynamic random access memory (enhanced SDRAM, ESDRAM), synchronous link dynamic random access memory (synch link DRAM, SLDRAM) and direct memory bus random access memory (Direct Rambus RAM, DR RAM), etc. That is to say, the memory in the embodiments of the present application is intended to include but not limited to these and any other suitable types of memory.

[0274] Based on the above, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program implements the above method when running.

[0275] In summary, by acquiring and analyzing the operating status log of the mold opening equipment in real time, anomaly detection is effectively performed, thereby achieving accurate grasp of the equipment's operating status. This process utilizes a dedicated operating status anomaly detection network, and through in-depth analysis of the status log fields, it generates abnormal state discrimination viewpoints, further improving the accuracy of anomaly detection. More importantly, this solution can not only identify whether there is an abnormality in the equipment, but also determine the specific type of abnormality, such as excessive temperature, excessive speed, etc. Based on these detailed abnormality labels, the solution can generate targeted equipment operation and maintenance decision recommendations for equipment maintenance personnel, thereby significantly improving the equipment's operating efficiency and production quality. This technical solution effectively integrates data collection, anomaly detection, and decision recommendations, providing a comprehensive and efficient solution for the operation and maintenance of mold opening equipment.

[0276] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations.

Claims

1. A method for detecting abnormal operation status of mold opening equipment, characterized in that: Applied to a running state abnormality detection device, the method comprises: Obtain the operating status log of the mold opening equipment to be tested for abnormality; Using the operation status anomaly detection network, anomaly detection is performed on the operation status log of the mold opening equipment to be detected for anomaly, and an abnormal state discrimination view of the operation status log of the mold opening equipment to be detected for anomaly is generated; the abnormal state discrimination view represents the abnormal label corresponding to each state log field in the operation status log of the mold opening equipment to be detected for anomaly; Determine equipment operation and maintenance decision suggestions for the operation status log of the mold opening equipment to be detected for abnormality based on the abnormality labels corresponding to the status log fields in the operation status log of the mold opening equipment to be detected for abnormality; The debugging steps of the operation status abnormality detection network include: Based on the state element vector mined by the state element mining subnet in the operation state anomaly detection network for the target operation state log, the state element vector chain of the target operation state log is determined; the operation state anomaly detection network is obtained by pre-debugging through the initial operation state log and the prior training knowledge of each state log field in the initial operation state log; Optimizing state elements according to the state element vector chain to obtain a state element optimization vector chain, wherein the state element optimization vector of the state log field in the state element optimization vector chain is obtained according to the state element optimization of the context state log field of the state log field; wherein the state element optimization is implemented by a long short-term memory network; Based on the state element vector chain and the state element optimization vector chain, determining the influence weight of each state log field in the target operation state log; wherein the step of determining the influence weight comprises: performing feature selection using a support vector machine or a decision tree algorithm to determine the influence weight of each state log field; Based on the influence weight of each status log field in the target operation status log and the status text vector of each status log field in the status element vector chain, a key status log field set to be annotated for status log fields is selected in the target operation status log; the influence weight is used to adjust the status text vector difference between two status log fields; Based on the prior training knowledge of the status log fields in the key status log field set and the target operating status log, performing transfer learning debugging on the operating status anomaly detection network; Wherein, the state element mining subnet is used to mine key state elements from the operation state log, and the state element mining subnet is a principal component analysis algorithm or an automatic encoder; The state element vector chain is a vector chain composed of state elements mined by the state element mining subnet. Each state element vector represents the characteristics of a state log field. The initial run status log is the raw data used to train the network; The target operation status log is the new data that we want to detect anomalies. Prior training knowledge is the network parameters and structure information obtained by training the initial operation status log, which is used for pre-debugging and transfer learning debugging; The impact weight is determined based on the state element vector chain and the state element optimization vector chain, and is used to measure the impact of each state log field on the device operating status; The key status log field set is a set of important status log fields to be annotated that are selected based on the impact weight and status text vector difference.

2. The method according to claim 1, characterized in that The step of selecting a key status log field set to be annotated with status log fields in the target operation status log based on the influence weight of each status log field in the target operation status log and the status text vector of each status log field in the status element vector chain includes: Determine a first status log field set, wherein a first status log field in the first status log field set is a status log field in the target running status log; For each second status log field in the second status log field set, determine in the first status log field set a first target status log field having the smallest status text vector difference with the second status log field; the status text vector difference is determined based on the status text vectors of the two status log fields; the second status log field in the second status log field set is a status log field in the target running status log, and there is no overlapping field between the second status log field set and the first status log field set; By using the influence weight of the first target state log field corresponding to the second state log field, the state text vector difference between the second state log field and the corresponding first target state log field is adjusted to obtain the target difference between the second state log field and the first state log field set, wherein the target difference is in a set quantitative relationship with the influence weight; Determine, in the second status log field set, a second target status log field having the largest target difference with the first status log field set; adding the second target status log field to the first status log field set; The key status log field set is determined based on the first status log field set that has been updated.

3. The method according to claim 2, characterized in that The determining the key status log field set based on the first status log field set that has been updated includes: Recording the number of status log fields of the first status log field set that have completed updating; If it is determined based on the number of the status log fields that the annotation condition is met, the first status log field set that has completed the update is used as the key status log field set; If it is determined based on the number of status log fields that the annotation condition is not met, the second target status log field is deleted from the second status log field set to obtain an adjusted second status log field set, and the process jumps to each second status log field in the second status log field set, and the first target status log field with the smallest status text vector difference between the second status log field and the first status log field is determined in the first status log field set.

4. The method according to claim 2, characterized in that: The operation state abnormality detection network further includes a first abnormality discrimination operator; for each second state log field in the second state log field set, before determining in the first state log field set a first target state log field having the smallest state text vector difference with the second state log field, the method further includes: The first abnormal state discrimination operator performs abnormality detection according to the result of state element mining on the target operation state log by the state element mining subnet, and obtains a first abnormal state discrimination viewpoint, wherein the first abnormal state discrimination viewpoint represents the possibility that each state log field in the target operation state log belongs to each abnormal label of multiple abnormal labels; For each state log field, determine the maximum X target possibilities for the state log field among the possibilities that the state log field belongs to multiple abnormal labels, where X is a positive integer; Based on the X target possibilities determined for each state log field, an annotation index of each state log field is calculated; Selecting a status log field whose annotation index is greater than a threshold of the annotation index from the target running status log, and generating the second status log field set according to the status log field whose annotation index is greater than the threshold of the annotation index; Wherein, X is 2, and the X target possibilities determined for a status log field include the maximum possibility and the second largest possibility among the possibilities that the status log field belongs to multiple abnormal labels; the step of calculating the annotation index of each status log field according to the X target possibilities determined for each status log field includes: for each status log field, calculating the sum of the maximum possibility and the second largest possibility corresponding to the status log field to obtain the reference possibility of the status log field; for each status log field, determining the annotation index of the status log field based on the difference result of subtracting the reference possibility corresponding to the status log field from a set constant.

5. The method according to claim 1, characterized in that The step of optimizing the state elements according to the state element vector chain to obtain the state element optimization vector chain includes: Performing a regional feature extraction operation on the state element vector chain to obtain a regional state feature relationship network, wherein the state text vector value of the reference state log field of the feature extraction core of each round of feature extraction in the regional state feature relationship network is 0; Integrate the regional state feature relationship network and the state element vector chain to obtain the state element optimization vector chain; Among them, the regional feature extraction operation is performed on the state element vector chain to obtain the regional state feature relationship network, including: performing sliding average processing on the state element vector chain to obtain a state element vector chain to be processed; combining the state element vector chain to be processed with the linear feature coverage vector corresponding to the feature extraction kernel to obtain a target state element vector chain; performing quantized feature mapping on the state element vectors in the target state element vector chain to obtain the regional state feature relationship network.

6. The method according to claim 1, characterized in that The determining, based on the state element vector chain and the state element optimization vector chain, the influence weight of each state log field in the target operation state log includes: For each status log field, obtain the status text vector of the status log field from the status element vector chain and obtain the status element optimization vector of the status log field from the status element optimization vector chain; The vector similarity between the state text vector of the state log field and the state element optimization vector of the state log field is obtained by operation; According to the vector similarity of the status log field, the influence weight of the status log field is determined.

7. The method according to claim 1, characterized in that The operation status anomaly detection network also includes a first anomaly discrimination operator; the state element optimization vector chain is obtained by optimizing the regional feature proposal model according to the state element vector chain; the prior training knowledge of the state log field in the key state log field set and the target operation status log are used to perform transfer learning debugging on the operation status anomaly detection network, including: Obtaining a first abnormal state discrimination viewpoint obtained by performing anomaly detection on a state element vector mined by the state element mining subnet on a target operation state log by a first abnormal state discrimination operator; Determining a first training cost function based on the first abnormal state discrimination viewpoint and the prior training knowledge of the state log fields in the key state log field set; Determining a state element optimization training cost function based on the state element vector chain and the state element optimization vector chain; Based on the first training cost function and the state element optimization training cost function, improving at least one variable of the operating state anomaly detection network and the regional feature proposal model; Among them, the state element vector chain is obtained by downsampling the first state element vector chain through the first residual model, and the first state element vector chain is the state element vector mined by the state element mining subnet for the target running state log; then the method also includes: performing anomaly detection based on the state element vector chain by the second abnormality discrimination operator to obtain a second abnormal state discrimination viewpoint; determining a second training cost function based on the second abnormal state discrimination viewpoint and the prior training knowledge of the state log field in the key state log field set; then based on the first training cost function and the state element optimization training cost function, improving at least one variable in the running state anomaly detection network and the regional feature proposal model, including: if the number of cycles is less than the first preset number, determining the target training cost function based on the first training cost function, the second training cost function and the state element optimization training cost function; based on the target training cost function, improving the variables of the running state anomaly detection network, the regional feature proposal model, the first residual model and the second abnormality discrimination operator; Among them, the step of optimizing the training cost function based on the first training cost function and the state element to improve the variables of at least one of the operating state anomaly detection network and the regional feature proposal model further includes: if the number of cycles is not less than the first preset number, improving the variables of the operating state anomaly detection network based on the first training cost function; and adjusting the variables of the regional feature proposal model, the first residual model and the second anomaly discrimination operator based on the weighted cost function of the second training cost function and the state element optimization training cost function; The method further comprises: performing transfer learning debugging on the operation state anomaly detection network according to the first initial operation state log and the learning annotation of the first initial operation state log; The step of performing transfer learning debugging on the operation status anomaly detection network according to the first initial operation status log and the learning annotation of the first initial operation status log includes: The state element mining subnet performs state element mining on the first initial operation state log to obtain a first state element vector chain of the first initial operation state log; The first abnormal state discrimination operator performs abnormality detection according to the first state element vector chain of the first initial operation state log to obtain a third abnormal state discrimination viewpoint; Determine a first training cost function for the first initial operating state log based on the prior training knowledge of the first initial operating state log and the third abnormal state discrimination viewpoint; Determining a state element vector chain of the first initial running state log according to the first state element vector chain of the first initial running state log; Performing state element optimization on the state element vector chain of the first initial operation state log by using the regional feature proposal model to obtain a state element optimized vector chain of the first initial operation state log; Determine a state element optimization training cost function for the first initial running state log based on the state element vector chain of the first initial running state log and the state element optimization vector chain of the first initial running state log; Determine a third training cost function for the first initial running state log based on the possibility that each state log field belongs to a priori abnormal label and the possibility that each state log field belongs to a noise abnormal label in the third abnormal state discrimination viewpoint; According to the first training cost function, the state element optimization training cost function and the third training cost function for the first initial operation state log, the variable of at least one of the operation state anomaly detection network and the regional feature proposal model is improved.

8. A device for detecting abnormal operation status, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program executes the method according to any one of claims 1 to 7 when running.

Images