Method and system for collecting and fusing multi-element device data based on internet of things

Abstract

The application provides a multi-device data fusion fault prediction method and system based on Internet of Things collection. First, a multi-device data set containing different device component operating state data and environmental influence data is obtained through Internet of Things collection. Each data carries a collection time and a device component identifier. Then, according to the device component identifier, the corresponding sub-library in the preset device knowledge base is associated. The sub-library stores component normal operation data characteristics and historical fault correlation data. Then, the multi-device data is scene-labeled. A fault prediction large model is called to fuse and infer the labeled data. In the inference, the sub-library is called in real time to supplement the associated data to generate fault trend characteristics. Finally, combined with historical fault development rule data, a device fault prediction result and a response suggestion are generated, including possible fault components, occurrence stages and influence ranges, greatly improving the efficiency and reliability of device operation and maintenance, and reducing the impact of device faults on production.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Internet of Things, in particular to a multi-element equipment data fusion fault prediction method and system based on Internet of Things collection. BACKGROUND

[0002] In the field of equipment operation and maintenance, with the continuous improvement of equipment automation and intelligence, the amount of data generated by various equipment is growing explosively. These equipment data are widely sourced, covering the running state data of different components of the equipment and various environmental influence data of the environment in which the equipment is located.

[0003] Traditional equipment fault prediction methods often have obvious limitations. On the one hand, they mostly focus on the data of a single component or a few components of the equipment, and fail to fully consider the complex correlation between multi-element equipment data and the influence of environmental factors on equipment operation, resulting in incomplete and inaccurate judgment of equipment faults. On the other hand, the existing fault prediction methods lack deep mining and effective use of historical fault data of the equipment, making it difficult to summarize rules from historical faults and accurately predict the development trend of equipment faults. In addition, in the process of data processing and analysis, there is a lack of close combination with professional knowledge of the equipment, making the fault prediction results lack of professionalism and pertinence, and difficult to meet the high requirements of equipment reliability and stability in actual production. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a multi-element equipment data fusion fault prediction method and system based on Internet of Things collection.

[0005] According to a first aspect of the present application, a multi-element equipment data fusion fault prediction method based on Internet of Things collection is provided, which comprises:

[0006] obtaining a multi-element equipment data set collected by Internet of Things, the multi-element equipment data set containing running state data of different equipment components and environmental influence data of the environment in which the equipment is located, each data carrying a collection time and a corresponding equipment component identifier;

[0007] According to the equipment component identifier corresponding to each data in the multi-element equipment data set, associating the corresponding equipment component knowledge base sub-library in the preset equipment knowledge base, the equipment knowledge base being divided into a plurality of equipment component knowledge base sub-libraries according to the type of equipment components, and each equipment component knowledge base sub-library storing normal running data features and historical fault correlation data of the corresponding component;

[0008] Based on the normal running data features and historical fault correlation data in the equipment component knowledge base sub-library, the running state data and environmental influence data in the multi-element equipment data set are labeled in a data scene to obtain a labeled multi-element equipment data set;

[0009] The preset fault prediction large model is called to perform fusion inference processing on the labeled multi-element equipment data set, and corresponding equipment component knowledge base sub-library supplementary associated data is called in real time in the inference process of the fault prediction large model to generate equipment fault trend characteristics;

[0010] According to the equipment fault trend characteristics, historical fault development law data in the equipment component knowledge base sub-library is combined to generate an equipment fault prediction result and a corresponding fault response suggestion, and the equipment fault prediction result includes a possible fault equipment component, a fault occurrence stage, and an associated impact range.

[0011] According to a second aspect of the present application, a multi-element equipment data fusion fault prediction system based on Internet of Things collection is provided, which includes a machine readable storage medium and a processor, the machine readable storage medium stores machine executable instructions, and the processor executes the machine executable instructions, and the multi-element equipment data fusion fault prediction system based on Internet of Things collection realizes the multi-element equipment data fusion fault prediction method based on Internet of Things collection.

[0012] According to a third aspect of the present application, a computer readable storage medium is provided, and the computer readable storage medium stores computer executable instructions, and when the computer executable instructions are executed, the multi-element equipment data fusion fault prediction method based on Internet of Things collection is realized.

[0013] According to any one of the above aspects, the technical effect of the present application is that:

[0014] By obtaining the multi-element equipment data set collected by the Internet of Things, the running state data and environmental influence data of different equipment components are comprehensively integrated, a preset equipment knowledge base is associated according to the equipment component identifier, the equipment professional knowledge is closely combined with the actual collected data, the multi-element equipment data is scene annotated, the running state of the equipment in different scenes can be accurately described, the fault prediction large model is called to perform fusion inference processing, and the equipment component knowledge base sub-library supplementary associated data is called in real time in the inference process, the powerful inference ability of the large model and the auxiliary role of the equipment professional knowledge are fully played, and the accuracy of fault trend characteristic extraction is effectively improved. Finally, the historical fault development law data is combined to generate an equipment fault prediction result and a response suggestion, not only the possible fault equipment component, the fault occurrence stage and the associated impact range are accurately predicted, but also practical response measures are provided for the operation and maintenance personnel, the efficiency and reliability of equipment operation and maintenance are greatly improved, and the impact of equipment failure on production is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0016] Figure 1 The flowchart of the multi-device data fusion fault prediction method based on Internet of Things collection provided by the embodiments of the present application is shown.

[0017] Figure 2 The component structure diagram of the multi-device data fusion fault prediction system based on Internet of Things collection provided by the embodiments of the present application is shown. DETAILED DESCRIPTION

[0018] The embodiments of the present application will be described below in conjunction with the drawings in the present application. It should be understood that the embodiments described below in conjunction with the drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions of the embodiments of the present application.

[0019] Figure 1 The flowchart of the multi-device data fusion fault prediction method and system based on Internet of Things collection provided by the embodiments of the present application is shown. It should be understood that in other embodiments, the order of some steps of the multi-device data fusion fault prediction method based on Internet of Things collection of the present embodiment can be shared according to actual needs, or some steps can be omitted or maintained. The detailed steps of the multi-device data fusion fault prediction method based on Internet of Things collection include:

[0020] Step S110: Obtain a multi-device data set collected by Internet of Things, the multi-device data set containing running state data of different device components and environmental influence data of the environment where the device is located, each data carrying a collection time and a corresponding device component identifier.

[0021] In the Internet of Things scenario, data collection covers different components of various Internet of Things devices. The running state data is derived from sensors and running logs inside the device, covering working parameters of the core components of the device, such as running intensity parameters of the power component, signal transmission parameters of the control component, action response parameters of the execution component, etc. The environmental influence data is obtained through environmental perception devices deployed around the device, including climate parameters, electromagnetic environment parameters, power supply parameters of the space where the device is located, etc.

[0022] Each collected data is accompanied by accurate collection time information, recorded in standard time format, ensuring the traceability of data in the time dimension. At the same time, each data corresponds to a unique device component identifier, which is composed of device type code, component category code and serial number. Through the identifier, the specific device and component to which the data belongs can be accurately located.

[0023] During data collection, multiple privacy protection techniques are used for privacy-sensitive data that may be involved. The collected data is anonymized to remove information that can be directly linked to the user's identity. Data transmission uses an encrypted channel, and encrypted protocols are used to package and transmit data to prevent data leakage during transmission. Data storage uses hierarchical access management, and only authorized modules can access sensitive data, and access operations are logged throughout.

[0024] Step S120: According to the device component identifier corresponding to each data in the multi-device data set, associate the corresponding device component knowledge base sub-library in the preset device knowledge base. The device knowledge base is divided into multiple device component knowledge base sub-libraries according to the type of device components. Each device component knowledge base sub-library stores the normal operation data characteristics and historical fault association data of the corresponding component.

[0025] The device knowledge base is a pre-constructed structured database, which is divided according to the functional type of the device component. Different functional types of components correspond to different sub-libraries, such as power component sub-library, control component sub-library, and sensor component sub-library. Each sub-library stores data independently to avoid cross interference of different types of component data.

[0026] Normal operation data characteristics include parameter fluctuation range, parameter change trend characteristics, and parameter correlation characteristics of components under different working conditions. Historical fault association data records the operating parameter characteristics, environmental parameter characteristics, parameter evolution process before failure, failure type and failure impact range, and other information when the component fails in the past.

[0027] Step S121: Extract the device component identifier corresponding to each data from the multi-device data set to form a device component identifier list, which contains the corresponding relationship between the collection time of each data and the device component identifier.

[0028] Through the data analysis program, the multi-device data set is traversed, and the device component identifier field in each data is extracted one by one. During the extraction process, the identifier format is verified to ensure the integrity and standardization of the identifier. For abnormal format identifiers, mark them as pending and record abnormal logs.

[0029] The extracted equipment component identifier is arranged in sequence with the collection time of the corresponding data to form an equipment component identifier list. The list adopts a two-dimensional structure, each row containing two fields of collection time and equipment component identifier, and the fields are distinguished by a separator for subsequent program parsing and processing. After the list is generated, integrity check is performed to ensure that the identifiers of all data have been correctly extracted and there is no repetition or omission.

[0030] Step S122: calling an index management module of the equipment knowledge base, the index management module of the equipment knowledge base storing a mapping relationship between the equipment component identifier and the equipment component knowledge base sub-library.

[0031] The index management module is a core navigation component of the equipment knowledge base, responsible for maintaining the association between the equipment component identifier and the sub-library. The module internally stores multiple index structures, including identifier prefix index, function type index, sub-library path index, etc., to improve query efficiency through multi-dimensional indexing.

[0032] The calling process is implemented through a standardized interface, sending an index query request to the index management module. The request contains a query type identifier and a reference to the equipment component identifier list to be queried. After receiving the request, the index management module starts a query processing thread, parses the request and schedules the corresponding index for query operation.

[0033] Step S1221: sending an identifier query request to the index management module, the identifier query request containing all equipment component identifiers in the equipment component identifier list.

[0034] The identifier query request adopts a structured data format, including a request header, a request body and a verification tail. The request header contains a request number, a request time, and a request source identifier; the request body contains the number of equipment component identifiers and the specific identifier list; and the verification tail is used for integrity verification of the request to prevent tampering during transmission.

[0035] Before sending the request, the request data is compressed to reduce the amount of transmission data. The request is sent to the receiving port of the index management module through the internal communication bus, and a timeout monitoring mechanism is started at the same time. If no response is received within the preset time, the request is resent and a retry log is recorded.

[0036] Step S1222: after receiving the identifier query request, the index management module matches the corresponding equipment function type of each equipment component identifier in the equipment component identifier classification table, and filters out the equipment component knowledge base sub-library category associated with the equipment function type.

[0037] After receiving the request, the index management module first decompresses and integrity checks the request. After passing the check, the device component identifier list in the request body is parsed, and each identifier is processed one by one. The identifier is extracted by the identifier parsing algorithm, and the feature code is matched in the device component identifier classification table according to the feature code.

[0038] The device component identifier classification table is stored in a tree structure, with the root node being a device category, the child node being a function type, and the leaf node being a specific component subdivision type. The matching process starts from the root node and traverses layer by layer until the corresponding function type node is found, and then the device function type corresponding to the identifier is determined. According to the function type, the corresponding sub-library category is filtered out from the pre-set association table.

[0039] Step S1223: In the device component knowledge base sub-library mapping table, according to the device component knowledge base sub-library category and device component identifier, the device component knowledge base sub-library name corresponding to each device component identifier is accurately matched.

[0040] The device component knowledge base sub-library mapping table stores the correspondence between the sub-library category, component identifier feature, and sub-library name. The mapping table uses a hash structure to optimize query speed, and each sub-library category corresponds to multiple sub-library name entries, each entry containing a specific component identifier feature code.

[0041] According to the sub-library category and device component identifier determined in step S1222, the subdivision feature code in the identifier is extracted, and the feature code comparison is performed under the corresponding sub-library category entry in the mapping table. The comparison uses a combination of fuzzy matching and accurate matching, first narrows the range through fuzzy matching, and then determines the unique corresponding sub-library name through accurate matching.

[0042] Step S1224: According to the device component knowledge base sub-library name, the corresponding storage path and update time are extracted from the device component knowledge base sub-library mapping table, and the update time is used to confirm whether the device component knowledge base sub-library data is the latest state.

[0043] Each sub-library name entry in the device component knowledge base sub-library mapping table is associated with a storage path and an update time field. The storage path uses a multi-level directory structure to represent, including main storage node identifier, sub-library classification directory, data file name, etc. The update time is accurate to seconds, recording the time point of the last update of the sub-library data.

[0044] After extracting the storage path, the path format is standardized to ensure that it meets the requirements of the data access interface. The extracted update time is compared with the current system time, and the time difference is calculated. If the time difference exceeds the pre-set threshold, it is marked that the sub-library data may have update lag, and a prompt is given in the subsequent data extraction process.

[0045] Step S1225: The index query result of each device component identifier, storage path and update time is sorted, and the index query result is used for subsequent access to the corresponding device component knowledge base sub-library.

[0046] The index query result is organized in a list structure, each device component identifier corresponds to a row record, and each row record contains the identifier itself, sub-library name, storage path, update time and data state identifier. The data state identifier is generated according to the update time judgment result, and is divided into normal, to be updated and expired.

[0047] After the result is generated, format conversion is performed to convert the list into a structured data format that can be directly recognized by the data access module. At the same time, the result is signed to ensure that the result is not tampered with in the subsequent use process. The processed index query result is stored in the temporary cache area, and a result ready notification is generated.

[0048] Step S123: Each device component identifier in the device component identifier list is input into the index management module to query the corresponding device component knowledge base sub-library name and storage path of each device component identifier.

[0049] Through a loop processing mechanism, the identifiers in the device component identifier list are input one by one into the query interface of the index management module. The query of each identifier is independent, and the query of the next identifier is started after the query of the previous identifier is completed, avoiding resource conflicts caused by concurrent queries.

[0050] During the query process, the query state is monitored in real time, and for identifiers that query timeout or query failure, the failure reason is recorded and retried. If the retry number reaches the preset upper limit and still fails, the identifier is marked as abnormal and handled separately later. For identifiers that query successfully, the sub-library name and storage path in the returned result are extracted and stored in association with the identifier itself.

[0051] Step S124: According to the storage path, access the corresponding device component knowledge base sub-library, extract the normal operation data features and historical fault associated data basic description information in each device component knowledge base sub-library, and form a device component knowledge base sub-library data description table.

[0052] According to the storage path, a connection with the device component knowledge base sub-library is established through a data access interface. Identity verification is performed during the connection process, and access rights of the sub-library are obtained after the verification is passed. According to the preset data extraction rules, the metadata of the normal operation data features is read from the sub-library, including the basic description information such as feature name, feature dimension and feature value range description.

[0053] Meanwhile, the basic description information of the historical fault association data is extracted, including fault type classification, typical feature description of each fault, time span of fault data, number of data samples, and the like. The extracted information is classified and arranged according to the sub-library names to form a device component knowledge library sub-library data description table, and each record in the table corresponds to the basic data description of a sub-library.

[0054] Step S125: An association relationship table of each data in the multi-device data set and the corresponding device component knowledge library sub-library is established, and the association relationship table includes data identification, device component identification, corresponding device component knowledge library sub-library name, and description information of the associated data in the device component knowledge library sub-library.

[0055] A unique data identification is generated for each data in the multi-device data set, and the data identification is generated in a combination of a time stamp and a random sequence to ensure uniqueness. The original data is associated with the device component identification through the data identification, and then the data is associated with the sub-library according to the correspondence between the device component identification and the sub-library.

[0056] The association relationship table adopts a relational database table structure and includes fields such as data identification, device component identification, sub-library name, associated data type, associated data description, and association strength. The associated data description field records the specific data entry name and feature description associated with the data in the sub-library, and the association strength field indicates the close degree of the association between the data and the sub-library data.

[0057] Step S126: Each data in the multi-device data set is bound to the corresponding device component knowledge library sub-library through the association relationship table to obtain a data sub-library binding result.

[0058] Based on the association relationship table, a data binding algorithm is used to establish a mapping relationship between each data in the multi-device data set and the corresponding sub-library. During the binding process, a sub-library reference pointer is added to each data, pointing to the corresponding device component knowledge library sub-library. Meanwhile, a reverse reference index is added to the sub-library to record the associated data identification and the record position in the association relationship table.

[0059] The data sub-library binding result is stored in the form of an index file, including a data identification list, a corresponding sub-library identification list, and a binding state identification. The binding state identification is used to indicate whether the binding is successful, and for the data that fails to be bound, the failure reason is recorded and separately marked in the result file. After the binding is completed, a consistency check is performed on the result to ensure that the data and the sub-library are accurately referenced in both directions.

[0060] Step S130: Based on the normal operation data features and historical fault correlation data in the device component knowledge base sub-library, the running state data and environmental influence data in the multi-element device data set are marked with data scenes to obtain a marked multi-element device data set.

[0061] The data scene marking is a process of classifying and marking the collected data according to the reference data in the sub-library. By comparing the collected data with the normal operation data features, the running state of the device reflected by the data is determined; in combination with the historical fault correlation data, the fault hidden danger features possibly existing in the data are identified. The marking process comprehensively considers the time characteristics, parameter characteristics and correlation characteristics of the data, so as to ensure that the marking result can accurately reflect the actual scene where the data is located.

[0062] The marked multi-element device data set not only contains original data, but also has scene marking information. The marking information adopts a standardized label format, which is convenient for model analysis and processing.

[0063] Step S131: Extracting the device component knowledge base sub-library corresponding to each data from the data sub-library binding result, obtaining the normal operation data features in the device component knowledge base sub-library, the normal operation data features including the running parameter range of the corresponding component under different environmental conditions.

[0064] By traversing the data sub-library binding result, the device component knowledge base sub-library identifier corresponding to each data is extracted. According to the sub-library identifier, the normal operation data features are read from the sub-library through a data access interface. The normal operation data features are stored according to environmental conditions, and each environmental condition category includes the feature description of multiple running parameters.

[0065] The running parameter range feature is represented in the form of an interval, including the lower limit value, upper limit value, typical value, fluctuation allowed range and other information of the parameter. For parameters with time series characteristics, the normal change rate range of the parameter over time, change period characteristics and other information are also included. The obtained normal operation data features are stored according to data identifiers and associated with the corresponding collected data.

[0066] Step S132: Comparing the running state data in the multi-element device data set with the running parameter range in the normal operation data features of the corresponding device component knowledge base sub-library to determine the running scene category to which the running state data belongs, the running scene category including normal scene and scene to be concerned.

[0067] The comparison process adopts a multi-dimensional parameter comparison method, in which each parameter in the running state data is compared with the range of the corresponding parameter in the normal operation data features one by one. The comparison contents include whether the current value of the parameter is within the normal range, whether the change rate of the parameter is within the allowed range, whether the ratio of the parameter to other associated parameters is normal, etc.

[0068] According to the comparison result, scene category determination is performed. If all parameters are within the normal range and the relationship between parameters conforms to the normal characteristics, it is determined as a normal scene; if there are parameters close to the boundary of the normal range, abnormal parameter change trend but not exceeding the range, and partial associated parameter ratio deviating from the normal, etc., it is determined as a scene to be concerned. The determination result is stored in association with the running state data, and the determination basis is recorded.

[0069] Step S133: Extract historical fault correlation data in the corresponding equipment component knowledge base sub-library, which contains the running state data features and environmental influence data features at the time of past fault occurrence.

[0070] According to the data sub-library binding result, the corresponding equipment component knowledge base sub-library is located, and historical fault correlation data is extracted through a query interface. Historical fault correlation data is organized by fault type, and each fault type entry contains multiple fault case records.

[0071] Each fault case record contains fault occurrence time, fault duration, running state data sequence before and after fault occurrence, contemporaneous environmental influence data sequence, fault phenomenon description, fault cause analysis, etc. Running state data features include abnormal fluctuation mode of parameters before fault occurrence, parameter mutation point characteristics, parameter correlation change characteristics, etc.; environmental influence data features include environmental parameter abnormal value at the time of fault occurrence, environmental parameter change trend, etc.

[0072] Step S134: Associate and match the environmental influence data in the multi-element equipment data set with the environmental influence data features in the historical fault correlation data to determine whether the environmental influence data has a fault correlation tendency.

[0073] The association and matching use a feature similarity analysis method. First, feature extraction is performed on the environmental influence data to obtain a feature vector containing data distribution characteristics, change trend characteristics, and abnormal point characteristics. At the same time, environmental influence data feature vectors are extracted from historical fault correlation data to form a historical feature vector library.

[0074] The similarity between the current environmental influence data feature vector and each vector in the historical feature vector library is calculated, and the similarity calculation considers the matching degree of each dimension in the feature vector. According to the similarity calculation result, the historical feature vectors with high similarity are selected, the corresponding fault occurrence conditions are analyzed, and then it is judged whether the current environmental influence data has a fault correlation tendency.

[0075] Step S1341: Extract all environmental influence data features at the time of historical fault occurrence from the historical fault correlation data of the equipment component knowledge base sub-library to form a historical environmental feature set.

[0076] Traverse all historical fault association data in the device component knowledge base sub-library, and extract the features of environmental impact data for each fault case. Feature extraction covers statistical features of data, such as mean value features, variance features, extreme value features, etc.; time domain features, such as change rate features, periodicity features, etc.; frequency domain features, such as main frequency component features, spectral distribution features, etc.

[0077] Organize the extracted features according to fault type and time sequence, and form a historical environmental feature set. The set is stored in a multi-dimensional array structure, each dimension corresponds to an environmental feature type, and the array elements are specific feature values or feature descriptions. The historical environmental feature set is processed to remove duplicate or highly similar feature items, reducing redundant data.

[0078] Step S1342: The environmental impact data in the multi-element device data set is decomposed into multiple environmental factor data, each corresponding to an environmental impact dimension.

[0079] The environmental impact data decomposition uses a dimension division algorithm to decompose the original data into multiple independent environmental factor data according to the type of environmental impact. Environmental impact dimensions include climate dimensions, electromagnetic dimensions, power dimensions, spatial dimensions, etc., each of which corresponds to a type of environmental factor.

[0080] For example, the climate dimension includes temperature factors, humidity factors, and air pressure factors; the electromagnetic dimension includes electromagnetic intensity factors and frequency interference factors. During the decomposition process, data cleaning is performed on each environmental factor data to remove noise data and outliers, ensuring the accuracy and integrity of each environmental factor data. The decomposed environmental factor data is stored according to the dimension identifier.

[0081] Step S1343: Compare each environmental factor data with the environmental data of the corresponding dimension in the historical environmental feature set, and extract the similar feature data in the historical environmental feature set.

[0082] For each environmental factor data, determine its environmental impact dimension, and locate the corresponding environmental data in the historical environmental feature set. The comparison uses an intra-dimension feature comparison method, which compares the features of the current environmental factor data with the features of the historical data.

[0083] The comparison includes the closeness of the feature values, the similarity of the feature change trend, and the matching degree of the feature distribution pattern. By setting a similarity threshold, the feature data in the historical environmental feature set that has a similarity to the current environmental factor data exceeding the threshold is extracted as similar feature data.

[0084] Step S1344: Count the number of failures corresponding to the similar feature data in the historical failure correlation data, and the correlation description of the feature data and the failure occurrence.

[0085] Trace the extracted similar feature data to find its corresponding failure case record in the historical failure correlation data. Count the total number of failures in these failure cases, and classify and count them by failure type to obtain the number of occurrences of each failure type.

[0086] At the same time, extract the description information of the relationship between the feature data and the failure occurrence in the historical failure correlation data, including the time interval between the occurrence of the feature data and the failure, the relationship between the change amplitude of the feature data and the severity of the failure, the synergistic effect description of the feature data and other failure causes. Organize the statistical results and correlation descriptions into structured data as the basis for subsequent analysis.

[0087] Step S1345: Determine the influence tendency of the current environmental factor data on the equipment component failure according to the number of failures and the correlation description.

[0088] The influence tendency determination adopts a weighted scoring method. The basic weight is set according to the number of failures. The more the number of failures, the higher the weight. Combined with the influence degree information in the correlation description, adjust the coefficient for different correlation description items. For example, the shorter the time interval between the feature data and the failure occurrence, the larger the adjustment coefficient; the stronger the positive correlation between the change amplitude of the feature data and the severity of the failure, the larger the adjustment coefficient.

[0089] Multiply the basic weight and the adjustment coefficient to obtain the influence score corresponding to each similar feature data. Summarize the influence scores of all similar feature data to obtain the total influence score of the current environmental factor data. According to the preset score interval where the total influence score is located, determine the influence tendency level, such as low tendency, medium tendency, and high tendency. At the same time, combined with the specific content in the correlation description, generate a detailed explanation of the influence tendency, including the main influencing factors, the influence mechanism, etc.

[0090] Step S1346: Integrate the influence tendencies of all environmental factor data to form the failure correlation tendency determination result corresponding to the environmental influence data, which includes the tendency level and the explanation of the associated environmental factors.

[0091] Traverse the influence tendency results of all environmental factor data, extract the tendency level and influence explanation of each environmental factor. Use a weighted comprehensive method to integrate the tendency levels of various environmental factors. According to the importance of different environmental factors to the equipment component failure, set the weight, the higher the importance, the greater the weight.

[0092] The tendency level of each environmental factor is converted into a corresponding numerical score, multiplied by the weight and summed up to obtain the comprehensive tendency score of the environmental impact data. The overall tendency level is determined according to the comprehensive tendency score. At the same time, the influence of each environmental factor is summarized, and the environmental factors with higher scores and their influence mechanisms are selected to form the related environmental factor description, which clearly indicates which environmental factors may have a related impact on the equipment component failure and the specific influence mode.

[0093] Step S135: According to the running scene category and the fault correlation tendency, a scene annotation label is added to each data in the multi-element equipment data set, which includes a normal scene label, a scene to be concerned label and a fault correlation tendency scene label.

[0094] A scene annotation rule library is established to clearly indicate the corresponding annotation label of the combination of different running scene categories and fault correlation tendencies. If the running scene category is a normal scene and the fault correlation tendency is low, a normal scene label is added. If the running scene category is a scene to be concerned, a scene to be concerned label is added regardless of the fault correlation tendency. If the fault correlation tendency is medium or high, a fault correlation tendency scene label is added regardless of the running scene category. When the labels corresponding to the running scene category and the fault correlation tendency conflict, the label is determined according to the fault correlation tendency priority principle.

[0095] When adding a label to each data, a label field is added to the data header to store the corresponding scene annotation label. At the same time, the basis for adding the label is recorded in the data metadata, including the running scene category determination result reference and the fault correlation tendency determination result reference, which facilitates the subsequent tracing of the generation logic of the label.

[0096] Step S136: All data with added scene annotation labels are integrated to form an annotated multi-element equipment data set, each data in the annotated multi-element equipment data set carrying a scene annotation label and a corresponding equipment component knowledge base sub-library association identifier.

[0097] All data with added scene annotation labels are grouped according to the equipment component identifier, and the data of the same equipment component are arranged in chronological order. An association identifier of the corresponding equipment component knowledge base sub-library is added to each data, and the association identifier is composed of the abbreviation code of the sub-library name and the sub-library version number.

[0098] The integrated data set is subjected to a completeness check to ensure that all data have added labels and association identifiers without omission or error. After the check passes, the data set is stored in a dedicated data storage area, and a data index file is generated to record the storage location, label information and association identifier of the data, improving the subsequent data retrieval efficiency.

[0099] Step S140: A preset fault prediction large model is called to perform fusion reasoning processing on the labeled multi-element equipment data set, and real-time corresponding equipment component knowledge base sub-library supplementary associated data is called in the inference process of the fault prediction large model to generate equipment fault trend characteristics.

[0100] The fault prediction large model adopts a deep learning architecture, including a feature association layer, a fusion reasoning layer, a feature output layer, and multiple functional layers. The model is initialized by loading pre-training parameters to ensure that the model has basic fault prediction capabilities. The labeled multi-element equipment data set is used as input data of the model, which is processed by each layer of the model to finally generate equipment fault trend characteristics.

[0101] During the inference process, the model establishes real-time communication with the equipment component knowledge base sub-library through an interface. When additional data is needed, a data request is sent and associated data is received to achieve dynamic data supplementation and improve inference accuracy.

[0102] Step S141: The labeled multi-element equipment data set is grouped according to equipment component identifiers to obtain multiple equipment component data groups, each containing labeled data and environmental impact data for the corresponding component.

[0103] The labeled multi-element equipment data set is traversed by a data grouping program, and the data is grouped according to the equipment component identifier. During the grouping process, a temporary data structure is created to store the data of each equipment component, which includes a labeled data list and an environmental impact data list.

[0104] The labeled data list stores the running state data of the equipment component and the corresponding scene annotation label in chronological order; the environmental impact data list stores the environmental impact data related to the equipment component and the fault association tendency determination result. After grouping, the data volume of each equipment component data group is counted, and the start time, end time, and data number are recorded.

[0105] Step S142: Each equipment component data group is input into the feature association layer of the fault prediction large model, which first calls the corresponding equipment component knowledge base sub-library of the equipment component to extract historical fault feature parameters from the equipment component knowledge base sub-library.

[0106] After receiving the equipment component data group, the feature association layer parses the equipment component identifier in the data group to determine the corresponding equipment component knowledge base sub-library. A historical fault feature parameter extraction request is sent to the sub-library through a data interface, which includes the equipment component identifier, data time range, and other information.

[0107] After receiving the request, the device component knowledge base sub-library queries the historical fault characteristic parameters of the device component in the same time span, including parameter change characteristics before the fault occurs, typical fault characteristic patterns, parameter correlation rules, and the like. The extracted historical fault characteristic parameters are arranged in the form of a feature vector and fed back to the feature correlation layer.

[0108] Step S143: The feature correlation layer is used to associate and map the labeled data in the device component data set with the extracted historical fault characteristic parameters to generate a data fault characteristic correlation matrix.

[0109] The feature correlation layer pre-processes the labeled data and the historical fault characteristic parameters, converts non-numeric data into numeric codes, and standardizes numeric data to ensure that the data is on the same order of magnitude. After pre-processing, the feature correlation algorithm is used to calculate the correlation between the labeled data and the historical fault characteristic parameters.

[0110] The correlation calculation considers the time correlation, parameter correlation, and feature pattern correlation of the data, and realizes nonlinear correlation mapping through a multi-layer neural network structure. The correlation results are organized in matrix form, with the rows representing the feature dimensions of the labeled data and the columns representing the dimensions of the historical fault characteristic parameters. The matrix element values represent the correlation degree between the corresponding dimensions, forming a data fault characteristic correlation matrix.

[0111] Step S144: The data fault characteristic correlation matrix is input into the fusion reasoning layer of the fault prediction large model, and the fusion reasoning layer cooperatively analyzes the running state data correlation features and the environmental influence data correlation features in the data fault characteristic correlation matrix to generate a preliminary reasoning result.

[0112] The fusion reasoning layer adopts an architecture combining attention mechanisms and recurrent neural networks. First, the data fault characteristic correlation matrix is feature-extracted to identify running state data correlation features and environmental influence data correlation features. The attention mechanism is used to assign weights to different features, focusing on features with high correlation to fault prediction.

[0113] The recurrent neural network models the feature sequence in time series to capture the variation of the features over time and analyze the dynamic correlation between the running state features and the environmental influence features. Combined with pre-set reasoning rules and knowledge obtained through model training, the features are comprehensively judged to generate a preliminary reasoning result containing possible fault types, fault occurrence probability trends, feature support evidence, and the like.

[0114] Step S1441: The fusion reasoning layer is used to perform time series continuity analysis on the running state data correlation features in the data fault characteristic correlation matrix to extract the feature change trend between different collection times.

[0115] The time sequence analysis module in the fusion reasoning layer sorts the operating state data correlation features according to the collection time to form a time sequence feature sequence. The sliding window technology is used for segmented processing of the time sequence feature sequence, and the window size is determined according to the data collection frequency and the fault evolution period of the device component.

[0116] The trend of each feature in each window is analyzed, and the change rate, fluctuation amplitude, extreme point distribution and other indexes of the feature are calculated. By comparing the feature indexes of adjacent windows, the continuous change mode of the feature is identified, such as upward trend, downward trend, periodic fluctuation trend, etc. The extracted change trend is sorted in time sequence to form a sequence of operating state feature change trends.

[0117] Step S1442: Perform influence degree analysis on the environmental influence data correlation features in the data fault feature correlation matrix to determine the action direction of the environmental factors on the operating state data correlation features.

[0118] The influence degree analysis module calculates the correlation coefficient between the environmental influence data correlation features and the operating state data correlation features, and judges the action direction through the positive and negative of the correlation coefficient. Positive correlation indicates that the environmental factor enhances the change of the operating state feature, and negative correlation indicates that the environmental factor inhibits the change of the operating state feature.

[0119] At the same time, the causal inference algorithm is used to analyze the causal relationship between the environmental factors and the operating state feature change, and to exclude false correlation. According to the absolute value of the correlation coefficient and the causal inference result, the influence strength of the environmental factors on the operating state feature is determined, which is divided into strong, medium and weak three levels. The action direction and influence strength are sorted as environmental influence analysis results.

[0120] Step S1443: Correlate the feature change trend and the environmental factor action direction to form a multi-factor collaborative analysis model.

[0121] The correlation modeling module takes the feature change trend sequence and the environmental influence analysis result as input to build a multi-factor collaborative analysis model. The model uses a graph neural network structure, taking the feature change trend and the environmental factor as nodes, and the edges between the nodes representing the correlation between the two, and the weight of the edge representing the correlation strength.

[0122] The model learns the collaborative mode between the feature change trend and the environmental factor action direction through model training, and captures how the environmental factors affect the evolution process of the feature change trend. The model output includes the prediction results of the feature change trend under different combinations of environmental factors.

[0123] Step S1444: Through the multi-factor collaborative analysis model, compare the feature change rule before the historical fault occurs in the device component knowledge base sub-library and the environmental action direction to determine the correlation degree of the current feature change and the historical fault feature.

[0124] The multi-factor synergistic analysis model loads the feature change rule data and the environmental action direction data before the historical fault occurs in the equipment component knowledge base sub-library as a reference template. The current feature change trend and the environmental factor action direction are input into the model and compared and analyzed with the reference template.

[0125] The comparison adopts a feature matching algorithm to calculate the similarity of the current feature and the historical fault feature in terms of change mode, environmental action direction, and change rate. According to the similarity calculation result, the correlation degree of the current feature change and the historical fault feature is determined, and the correlation degree is expressed by a percentage. The higher the percentage, the closer the correlation.

[0126] Step S1445: According to the correlation degree, a preliminary reasoning result containing possible fault types and fault occurrence probability tendencies is generated, and each possible fault type in the preliminary reasoning result corresponds to a set of associated feature change descriptions and environmental action descriptions.

[0127] According to the correlation degree ranking, the historical fault types with higher correlation degrees are selected as possible fault types. For each possible fault type, the fault occurrence probability tendency is calculated according to the correlation degree. The higher the correlation degree, the greater the probability tendency.

[0128] At the same time, the current feature change description associated with the possible fault type is extracted, including the feature change trend, the change of the key feature point, etc.; and the corresponding environmental action description, including the main influencing environmental factors, the action direction and intensity. The possible fault type, the fault occurrence probability tendency, the feature change description and the environmental action description are arranged into a structured preliminary reasoning result.

[0129] Step S145: In the process of generating the preliminary reasoning result, if it is detected that the data correlation feature has information gaps, the corresponding device component knowledge base sub-library is called in real time to supplement the associated data of the historical similar scene, and the preliminary reasoning result is perfected.

[0130] The fusion reasoning layer has an information gap detection mechanism built in, which monitors the integrity of the data correlation feature in real time during reasoning. When an information gap is detected, a data supplement process is triggered, relevant data is obtained from the device component knowledge base sub-library through an interface, and reasoning is performed again after the gap is supplemented to ensure the accuracy of the preliminary reasoning result.

[0131] Step S1451: The fusion reasoning layer monitors the integrity of the feature parameters in real time when analyzing the data correlation feature. If any feature parameter is missing or has insufficient data in the current data fault feature correlation matrix, it is determined that there is an information gap.

[0132] The feature parameter integrity monitoring module checks each feature parameter in the data fault feature correlation matrix, records the existence state and data amount of the parameter. If there is no corresponding value of a feature parameter in the matrix, it is determined as missing; if the data amount of the feature parameter is lower than the preset minimum data amount threshold, it is determined as insufficient data amount.

[0133] For the feature parameters determined as missing or insufficient data amount, mark them as information gaps, and record the feature parameter type, data location and other information of the gap.

[0134] Step S1452: According to the feature parameter type and equipment component identifier corresponding to the information gap, a data supplement request is generated, which contains the feature parameter type, equipment component identifier and part of the existing feature data.

[0135] The data supplement request generation module constructs a standardized data supplement request according to the feature parameter type and equipment component identifier recorded in the information gap. The request contains the specific name of the gap feature parameter, data format requirement, time range and other information; the equipment component identifier is used to locate the corresponding equipment component knowledge base sub-library; part of the existing feature data is used as a reference basis for similarity matching.

[0136] After the request is generated, the request content is verified to ensure the accuracy and integrity of the information, and to avoid sending invalid requests.

[0137] Step S1453: The data supplement request is sent to the corresponding equipment component knowledge base sub-library. After receiving the request, the equipment component knowledge base sub-library queries the associated feature data in the historical similar scene according to the feature parameter type and equipment component identifier; the historical similar scene is determined by comparing the similarity of the part of the existing feature data with the historical data in the equipment component knowledge base sub-library, and the feature data corresponding to the historical scene with a similarity meeting the requirements is selected.

[0138] After receiving the data supplement request, the equipment component knowledge base sub-library parses the feature parameter type, equipment component identifier and part of the feature data in the request. According to the feature parameter type and equipment component identifier, the corresponding historical data storage area in the sub-library is located.

[0139] Using a similarity matching algorithm, the part of the existing feature data is compared with the historical data to calculate the similarity. A similarity threshold is set to select the historical scene data with a similarity higher than the threshold, extract the associated feature data in these scenes, and ensure that the supplemented data has high similarity with the current scene.

[0140] Step S1454: The selected associated feature data is fed back to the fusion reasoning layer, and the supplemented associated feature data is filled into the information gap through the fusion reasoning layer to update the data fault feature correlation matrix.

[0141] The device component knowledge base sub-library organizes the screened associated feature data into the format required by the request, adds data source identification and a timestamp, and feeds back to the fusion reasoning layer. After receiving the data, the fusion reasoning layer verifies the validity and integrity of the data, and ensures that the data format matches the gap feature parameters.

[0142] The supplemented associated feature data is filled into the corresponding position of the data fault feature association matrix, covering the original gap area. After updating, the consistency of the data fault feature association matrix is checked to ensure that the dimensions and data distribution of the data fault feature association matrix are reasonable.

[0143] Step S1455: Based on the updated data fault feature association matrix, re-perform collaborative analysis to correct the original preliminary reasoning result to obtain a perfected preliminary reasoning result.

[0144] The fusion reasoning layer calls the collaborative analysis module to re-analyze using the updated data fault feature association matrix. Repeat the collaborative analysis process of steps S1441 to S1445 to generate a new reasoning result.

[0145] Compare the new reasoning result with the original preliminary reasoning result, analyze the difference points and the causes. According to the difference analysis result, correct the original reasoning result, retain the reasonable part, and correct the error part, and finally obtain a perfected preliminary reasoning result, to ensure that the reasoning result is more accurate and reliable.

[0146] Step S146: Convert the perfected preliminary reasoning result into a time-sequenced feature sequence, which is arranged in order of collection time, and each collection time corresponds to a group of feature parameters. The feature sequence is a device fault trend feature.

[0147] The result conversion module extracts features from the perfected preliminary reasoning result, converts information such as possible fault types and fault occurrence probability trends in the reasoning result into numerical feature parameters. The feature parameters are sorted in order of collection time to ensure that each collection time corresponds to a complete set of feature parameters.

[0148] The feature parameters include fault probability features, feature association strength features, environmental influence features, and multiple dimensions. Each dimension of feature parameters reflects the fault-related state of the device at that time. These feature parameters are combined to form a time-sequenced feature sequence, and each element in the sequence represents a device fault trend feature at a collection time. The overall sequence reflects the change of the device fault trend over time.

[0149] Step S150: According to the device failure trend feature, combined with the historical failure development rule data in the device component knowledge base sub-library, the device failure prediction result and the corresponding failure response suggestion are generated, and the device failure prediction result includes the device component that may fail, the failure development stage and the associated impact range.

[0150] The device failure trend feature reflects the dynamic change of the device failure state, and combined with the historical failure development rule data, the occurrence of the failure can be more accurately predicted. By comparative analysis and trend extrapolation, the device component that may fail, the failure development stage and the possible impact range are determined, and the targeted failure response suggestion is generated according to the historical response experience.

[0151] Step S151: Extract the feature parameters of each collection time from the device failure trend feature, and construct a feature change sequence in the order of collection time.

[0152] The feature extraction module traverses the device failure trend feature, extracts the feature parameters of each collection time, including the failure probability parameter, the feature correlation strength parameter, the environmental influence parameter, etc. The extracted feature parameters are arranged in the order of collection time to form a feature change sequence.

[0153] Each element in the feature change sequence contains all the feature parameters and the corresponding time stamp of this time. Through the feature change sequence, the change of each feature parameter over time can be clearly observed.

[0154] Step S152: According to the device component identifier corresponding to the feature change sequence, the corresponding device component knowledge base sub-library is called to extract the historical failure development rule data in the device component knowledge base sub-library, and the historical failure development rule data includes the feature evolution process and the corresponding time length from the initial feature to the failure.

[0155] According to the device component identifier in the feature change sequence, the corresponding device component knowledge base sub-library is determined. The historical failure development rule data extraction request is sent to the sub-library through the data interface, and the request includes the device component identifier and the time range of the feature change sequence.

[0156] After receiving the request, the device component knowledge base sub-library queries the historical failure record of the device component, and extracts the feature evolution process data and the corresponding time length data. The feature evolution process data includes the change sequence of each feature parameter in the whole process from the initial abnormal feature to the final failure of the historical failure; the corresponding time length data includes the time interval from the initial feature to the failure, the duration of each failure development stage, etc.

[0157] Step S153: Compare the feature change sequence with the feature evolution process in the historical fault development rule data to determine the fault development stage in which the current feature change sequence is located.

[0158] The comparison module compares the feature change sequence with the feature evolution process in the historical fault development rule data in multiple dimensions. The comparison content includes the similarity of the change trend of the feature parameters, the order similarity of the feature abnormal points, and the change similarity of the correlation between the feature parameters.

[0159] According to the comparison result, the most similar historical feature evolution process to the current feature change sequence is found. Referring to the fault development stage division standard corresponding to the historical evolution process, and combining the progress of the current feature change sequence, the current fault development stage is determined, such as the initial abnormal stage, the feature obvious stage, and the fault approaching stage.

[0160] Step S154: Based on the fault development stage and the corresponding time length in the historical fault development rule data, the time range in which the current device component may fail is calculated.

[0161] From the historical fault development rule data, the historical time length data corresponding to the current fault development stage is extracted, including the average time length, the shortest time length, and the longest time length from the start of the stage to the next stage. Combined with the feature change rate of the current fault development stage, the feature change rate of the historical same stage is compared. If the current feature change rate is faster than the historical average rate, the historical shortest time length is referred to for calculation; if the current feature change rate is slower than the historical average rate, the historical longest time length is referred to for calculation; if the rate is close to the historical average rate, the historical average time length is mainly referred to. By comprehensively analyzing the historical time length data and the current rate comparison result, the time range in which the current device component may fail is determined, which is presented in the form of a starting time point and an ending time point. The starting time point is the earliest possible failure time calculated based on the current stage progress, and the ending time point is the latest possible failure time.

[0162] Step S155: Combined with the fault development stage and the time range in which the failure may occur, the device component that may fail, the failure occurrence stage, and the associated impact range are determined to form the device failure prediction result.

[0163] According to the determination result of the fault development stage, the current device component that may fail is determined, which is the device component corresponding to the feature change sequence. The determined fault development stage, such as the initial abnormal stage, the feature obvious stage, the fault approaching stage, etc., is included in the prediction result. The associated influence analysis data in the device component knowledge base sub-library is called, which records the other device components that may be affected by this type of device component at different fault stages and the influence degree. Combined with the current fault development stage and the time range in which the fault may occur, the diffusion path and range of the fault influence are analyzed, the associated influence range is determined, and the associated influence range is presented in the form of a device component list and an influence level. The influence level is divided according to the influence degree of this type of fault on other components in the historical fault data. The device component that may fail, the fault occurrence stage, the time range in which the fault may occur, and the associated influence range are integrated to form a structured device fault prediction result, and each part is accompanied by corresponding analysis basis to ensure the traceability of the prediction result.

[0164] Step S156: According to the fault type and fault development stage in the device fault prediction result, the corresponding fault response measure data in the device component knowledge base sub-library is called to generate a fault response suggestion matching the fault prediction result, and the fault response suggestion includes intervention operation steps and operation timing suggestions.

[0165] The fault type and fault development stage information are extracted from the device fault prediction result, and the corresponding fault response measure data is queried in the device component knowledge base sub-library by taking this as the retrieval condition. The fault response measure data is stored according to fault type and development stage, including standard intervention procedures for different stages, recommended tools or spare parts, operation precautions, etc. According to the current fault development stage, the intervention operation steps suitable for this stage are selected, and the operation steps are arranged in execution order, each step including operation target, operation content description, operation specification requirement, etc. Combined with the time range in which the fault may occur, appropriate operation timing is recommended for each intervention operation step. If it is in the initial abnormal stage and the time range is long, it is recommended to perform preventive intervention as soon as possible; if it is in the fault approaching stage and the time range is short, it is recommended to perform emergency response operation preferentially. The intervention operation steps and operation timing suggestions are integrated to form a complete fault response suggestion, and the suggestion also includes operation effect evaluation indicators for subsequent evaluation of the effectiveness of the intervention operation.

[0166] For example, step S1551: The fault development stage is divided into multiple stages according to the feature change degree, and the multiple stages include an initial warning stage, a feature aggravation stage, and a fault approaching stage, each stage corresponding to a group of feature change thresholds.

[0167] By analyzing the degree difference of feature change in historical failure development rule data, multiple feature change threshold intervals are set, and each interval corresponds to a failure development stage. The feature change threshold interval corresponding to the initial warning stage is that the feature parameter starts to deviate from the normal range, but the deviation is small, and the feature change rate is slow; the threshold interval corresponding to the feature aggravation stage is that the deviation of the feature parameter increases and exceeds the slight deviation range, and the feature change rate significantly accelerates; the threshold interval corresponding to the failure approaching stage is that the feature parameter approaches or reaches the failure trigger threshold, and the feature change presents a trend of violent fluctuation or mutation. The feature change threshold of each stage is obtained based on a large amount of historical data statistics, and contains the threshold range of multiple key feature parameters, and the threshold will be dynamically adjusted according to the model, service life and other factors of the device component, to ensure the accuracy of stage division.

[0168] Step S1552: In combination with the time range in which the failure may occur, the remaining duration corresponding to the failure development stage in which the current device component is located is determined, and the remaining duration is the time interval from the current time to the time when the failure may occur.

[0169] According to the time difference calculation between the end time point in the time range in which the failure may occur and the current time, the remaining duration is obtained. If the time range is a time period, the end time point of the time period is taken as the calculation reference, and the remaining duration is ensured to be the longest possible time interval, leaving sufficient preparation time for failure response. The remaining duration is compared with the typical duration of the current failure development stage, if the remaining duration is less than the typical duration of the stage, it means that the failure development may accelerate, and the response strategy needs to be adjusted; if the remaining duration is greater than the typical duration, it means that the failure development is relatively flat, and the response operation can be executed according to the normal process. The remaining duration is recorded in a standard time format and is synchronized to the failure response suggestion module as an important basis for operation timing suggestion.

[0170] Step S1553: According to the device component identifier, the associated component mapping table of the component in the device knowledge base is called, and the associated component mapping table records other device components that may be affected by the failure of the component and the influence degree description.

[0171] The corresponding associated component mapping table is located in the device knowledge base through the device component identification, and the associated component mapping table is stored in a matrix form, the rows represent the current device components, the columns represent other device components that can be affected, and the matrix elements are impact degree descriptions. The impact degree description is determined based on the actual impact of other components after the failure of the component in the historical failure case, including impact probability, impact range size, impact duration, etc. For example, if a power component failure has a high probability of causing the transmission component connected thereto to be overloaded, the impact degree description in the corresponding position of the mapping table is "high probability of causing overload, and the impact lasts until the power component is repaired". During the calling process, the version of the mapping table is checked to ensure that the latest updated associated relationship data is used.

[0172] Step S1554: Based on the associated component mapping table and the current failure development stage, determine the associated impact range that can be affected by the current component failure.

[0173] According to the severity of the current failure development stage, combined with the impact degree description in the associated component mapping table, the associated components that can be affected are screened out. In the initial warning stage, only the associated components with "extremely high" impact degree are considered; in the feature aggravation stage, the associated components with "high" and "extremely high" impact degree are included; in the failure approaching stage, the associated components with "medium" and above impact degree are included. The screened associated components are hierarchically divided, the first-level associated components are the directly affected components, the second-level associated components are the components affected by the first-level associated components, and so on, to form a multi-level associated impact range. At the same time, according to the remaining time length of the failure development stage, the diffusion speed of the associated impact range is predicted, if the remaining time length is short, the diffusion range is relatively small; if the remaining time length is long, more extensive potential impact components need to be considered.

[0174] Step S1555: Integrate the device component identification that can fail, the current failure development stage, the time range in which the failure can occur, and the associated impact range to form a structured device failure prediction result; each part in the device failure prediction result corresponds to an associated basis description, and the basis description is derived from the historical failure development rule data and the associated component impact data in the device component knowledge base sub-library.

[0175] The device component that may fail is identified and filled into the designated field of the prediction result in a standardized coding form, and the basic information description of the component is attached. The current failure development stage is explicitly marked as the initial warning stage, the feature aggravation stage or the failure approaching stage, and the key feature basis for determining the stage is briefly explained, such as the feature parameter deviation degree, the change rate, etc. The time range in which the failure may occur is presented in the format of "start time-end time", and the historical time data and the current feature change rate situation referred to for calculating the time range are noted. The associated influence range is listed in a hierarchical structure, and each level contains the associated component identification, the influence degree and the possible influence performance description. At the end of each part of the prediction result, the basis description is added, which refers to the corresponding historical failure case number, the data statistical conclusion or the associated analysis report number in the device component knowledge base sub-library, so as to facilitate subsequent tracing and verification.

[0176] Figure 2 A multi-device data fusion fault prediction system based on Internet of Things collection provided in the embodiments of the present application is shown, which includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, such as through a bus 1002. Optionally, the multi-device data fusion fault prediction system based on Internet of Things collection can also include a transceiver 1004, which can be used for data interaction between the multi-device data fusion fault prediction system based on Internet of Things collection and other multi-device data fusion fault prediction systems based on Internet of Things collection, such as data transmission and / or data reception. It should be noted that the transceiver 1004 is not limited to one in actual scheduling, and the structure of the multi-device data fusion fault prediction system based on Internet of Things collection does not constitute a limitation on the embodiments of the present application.

[0177] The memory 1003 is used to store the program code for executing the embodiments of the present application, and is controlled by the processor 1001 to execute. The processor 1001 is used to execute the program code stored in the memory 1003 to realize the steps shown in the foregoing method embodiments.

[0178] The embodiments of the present application provide a computer readable storage medium, which stores program code, and the program code is executed by a processor to realize the steps and corresponding contents of the foregoing method embodiments.

[0179] The above is only an optional implementation of some implementation scenarios of the present application, and it should be noted that for those skilled in the art, other similar implementation manners according to the technical idea of the present application without departing from the technical concept of the present application also belong to the protection scope of the embodiments of the present application.

Claims

1. A method for collecting and fusing multi-element equipment data based on Internet of Things, characterized in that, The method comprises: acquiring a plurality of device data sets collected by Internet of Things, the plurality of device data sets comprising operating state data of different device components and environmental influence data of the environment in which the device is located, each data carrying a collection time and a corresponding device component identifier; associating a corresponding device component knowledge base sub-library in a preset device knowledge base according to the device component identifier corresponding to each data in the plurality of device data sets, the device knowledge base being divided into a plurality of device component knowledge base sub-libraries according to device component types, and each device component knowledge base sub-library storing normal operating data characteristics and historical fault correlation data of a corresponding component; based on the normal operating data characteristics and historical fault correlation data in the device component knowledge base sub-library, performing data scene labeling on the operating state data and the environmental influence data in the plurality of device data sets to obtain a plurality of labeled device data sets; calling a preset fault prediction large model to perform fusion reasoning processing on the plurality of labeled device data sets, the fault prediction large model calling a corresponding device component knowledge base sub-library in real time during reasoning to supplement associated data and generate device fault trend characteristics; based on the device fault trend characteristics, combining historical fault development rule data in the device component knowledge base sub-library to generate a device fault prediction result and a corresponding fault response suggestion, the device fault prediction result comprising a device component that may fail, a fault occurrence stage, and an associated influence range; the calling of the preset fault prediction large model to perform fusion reasoning processing on the plurality of labeled device data sets, the fault prediction large model calling a corresponding device component knowledge base sub-library in real time during reasoning to supplement associated data and generate device fault trend characteristics, comprising: grouping the plurality of labeled device data sets according to device component identifiers to obtain a plurality of device component data groups, each device component data group comprising labeled data and environmental influence data of a corresponding component; inputting each device component data group into a feature association layer of the fault prediction large model, the feature association layer first calling a device component knowledge base sub-library corresponding to the device component to extract historical fault characteristic parameters in the device component knowledge base sub-library; using the feature association layer to associate and map the labeled data in the device component data group with the extracted historical fault characteristic parameters to generate a data fault feature association matrix; inputting the data fault feature association matrix into a fusion reasoning layer of the fault prediction large model, the fusion reasoning layer performing collaborative analysis on operating state data association characteristics and environmental influence data association characteristics in the data fault feature association matrix to generate a preliminary reasoning result; during generation of the preliminary reasoning result, if a data association characteristic is detected to have a missing information gap, the fusion reasoning layer calls a corresponding device component knowledge base sub-library in real time to supplement associated data of a historical similar scene and perfect the preliminary reasoning result; converting the perfected preliminary reasoning result into a time-sequenced feature sequence, the feature sequence being arranged in order of collection time, each collection time corresponding to a group of feature parameters, and the feature sequence being a device fault trend characteristic. 2.The method according to claim 1, wherein, The device component identifier corresponding to each data in the multi-device data set is extracted from the multi-device data set to form a device component identifier list, and the device component identifier list contains the corresponding relationship between the collection time of each data and the device component identifier. An index management module of the device knowledge base is called, and the index management module of the device knowledge base stores the mapping relationship between the device component identifier and the device component knowledge base sub-library. Each device component identifier in the device component identifier list is input into the index management module to query the device component knowledge base sub-library name and storage path corresponding to each device component identifier. The device component knowledge base sub-library corresponding to each device component identifier is accessed according to the storage path to extract the normal operation data features and the basic description information of the historical fault associated data in each device component knowledge base sub-library, and a device component knowledge base sub-library data description table is formed. An association relationship table of each data in the multi-device data set and the corresponding device component knowledge base sub-library is established, and the association relationship table contains the data identifier, the device component identifier, the corresponding device component knowledge base sub-library name, and the description information of the associated data in the device component knowledge base sub-library. Each data in the multi-device data set is bound to the corresponding device component knowledge base sub-library through the association relationship table to obtain a data sub-library binding result. The index management module of the device knowledge base contains a device component identifier classification table and a device component knowledge base sub-library mapping table, the device component identifier classification table classifies the device component identifiers according to the device function types, and the device component knowledge base sub-library mapping table records the name, storage path and update time of each device component identifier and the corresponding device component knowledge base sub-library. 3.The method according to claim 2, wherein, The index management module of the device knowledge base stores the mapping relationship between the device component identifier and the device component knowledge base sub-library, including: An identifier query request is sent to the index management module, and the identifier query request contains all device component identifiers in the device component identifier list. After the index management module receives the identifier query request, the corresponding device function type of each device component identifier is matched in the device component identifier classification table, and the device component knowledge base sub-library category associated with the device function type is filtered out. The device component knowledge base sub-library name corresponding to each device component identifier is accurately matched in the device component knowledge base sub-library mapping table according to the device component knowledge base sub-library category and the device component identifier. The corresponding storage path and update time are extracted from the device component knowledge base sub-library mapping table according to the device component knowledge base sub-library name, and the update time is used to confirm whether the device component knowledge base sub-library data is in the latest state. The device component knowledge base sub-library name, storage path and update time corresponding to each device component identifier are arranged as an index query result, and the index query result is used for subsequent access to the corresponding device component knowledge base sub-library. ​ 4.The method according to claim 2, wherein, The normal operation data features and historical fault correlation data in the device component knowledge base sub-library are used to perform data scene labeling on the operating state data and environmental influence data in the multi-element device data set, to obtain a labeled multi-element device data set, comprising: Extract the device component knowledge base sub-library corresponding to each data from the data sub-library binding result, and obtain the normal operation data features in the device component knowledge base sub-library, wherein the normal operation data features include the operating parameter range of the corresponding component under different environmental conditions; Compare the operating state data in the multi-element device data set with the operating parameter range in the normal operation data features of the corresponding device component knowledge base sub-library, to determine the operating scene category to which the operating state data belongs, wherein the operating scene category includes normal scenes and scenes to be concerned; Extract the historical fault correlation data in the corresponding device component knowledge base sub-library, wherein the historical fault correlation data includes the operating state data features and environmental influence data features when a fault occurred in the past; Correlate and match the environmental influence data in the multi-element device data set with the environmental influence data features in the historical fault correlation data, to determine whether the environmental influence data has a fault correlation tendency; According to the operating scene category and the fault correlation tendency, add scene labeling tags to each data in the multi-element device data set, wherein the scene labeling tags include normal scene tags, scenes to be concerned tags, and fault correlation tendency scene tags; Integrate all the data with added scene labeling tags to form a labeled multi-element device data set, wherein each data in the labeled multi-element device data set carries a scene labeling tag and a corresponding device component knowledge base sub-library association identifier. 5.The method of claim 4, wherein, The correlating and matching of the environmental influence data in the multi-element device data set with the environmental influence data features in the historical fault correlation data to determine whether the environmental influence data has a fault correlation tendency comprises: Extract the environmental influence data features when all historical faults occurred from the historical fault correlation data of the device component knowledge base sub-library to form a historical environmental feature set; Disassemble the environmental influence data in the multi-element device data set into multiple environmental factor data, each of which corresponds to an environmental influence dimension; Compare each environmental factor data with the environmental data of the corresponding dimension in the historical environmental feature set, and extract the feature data in the historical environmental feature set that is similar to the current environmental factor data; Statistically count the number of faults corresponding to the similar feature data in the historical fault correlation data, and the correlation description of the feature data and the faults; According to the number of faults and the correlation description, determine the influence tendency of the current environmental factor data on the device component fault; Integrate the influence tendencies of all environmental factor data to form a fault correlation tendency judgment result corresponding to the environmental influence data, wherein the fault correlation tendency judgment result includes a tendency level and an associated environmental factor description. 6.The method according to claim 1, wherein, The fusion reasoning layer inputs the data fault feature correlation matrix into the fault prediction large model, and performs collaborative analysis on the running state data correlation features and the environmental influence data correlation features in the data fault feature correlation matrix to generate a preliminary reasoning result, including: The fusion reasoning layer performs time sequence continuity analysis on the running state data correlation features in the data fault feature correlation matrix, and extracts feature change trends between different collection time points; The influence degree analysis is performed on the environmental influence data correlation features in the data fault feature correlation matrix to determine the action direction of environmental factors on the running state data correlation features; The feature change trends and the action direction of environmental factors are associated and modeled to form a multi-factor collaborative analysis model; Through the multi-factor collaborative analysis model, the feature change and the historical fault feature are compared to determine the correlation degree of the feature change and the historical fault feature. According to the correlation degree, a preliminary reasoning result including possible fault types and fault occurrence probability tendencies is generated, and each possible fault type in the preliminary reasoning result corresponds to a set of associated feature change descriptions and environmental action descriptions. 7.The method according to claim 1, wherein, During the generation of the preliminary reasoning result, if the fusion reasoning layer detects that the data correlation features have information gaps, the corresponding device component knowledge base sub-library is called in real time to supplement the associated data of the historical similar scene, and the preliminary reasoning result is improved, including: When analyzing the data correlation features, the fusion reasoning layer monitors the integrity of the feature parameters in real time, and if any feature parameter is missing or has insufficient data in the current data fault feature correlation matrix, it is determined that there is an information gap; According to the feature parameter type and the device component identifier corresponding to the information gap, a data supplement request is generated, which includes the feature parameter type, the device component identifier and the existing part of the feature data; The data supplement request is sent to the corresponding device component knowledge base sub-library, and the device component knowledge base sub-library queries the associated feature data in the historical similar scene according to the feature parameter type and the device component identifier after receiving the request; the historical similar scene is determined by comparing the similarity between the existing part of the feature data and the historical data in the device component knowledge base sub-library, and the feature data corresponding to the historical scene that meets the similarity requirement is selected; The filtered associated feature data is fed back to the fusion reasoning layer, and the fusion reasoning layer fills the supplemented associated feature data into the information gap and updates the data fault feature correlation matrix; Based on the updated data fault feature correlation matrix, collaborative analysis is performed again to correct the original preliminary reasoning result and obtain an improved preliminary reasoning result. 8.The method according to claim 1, wherein, According to the device fault trend feature, the historical fault development rule data in the device component knowledge base sub-library is combined to generate a device fault prediction result and a corresponding fault response suggestion, including: Feature parameters at each collection time point are extracted from the device fault trend feature, and a feature change sequence is constructed according to the collection time sequence; According to the equipment component identifier corresponding to the feature change sequence, a corresponding equipment component knowledge base sub-library is called, and historical fault development rule data in the equipment component knowledge base sub-library is extracted, the historical fault development rule data including a feature evolution process and a corresponding time length of a historical fault from an initial feature to fault occurrence; The feature change sequence is compared with the feature evolution process in the historical fault development rule data to determine a fault development stage in which the current feature change sequence is located; Based on the fault development stage and the corresponding time length in the historical fault development rule data, a time range in which the current equipment component is likely to fail is calculated; In combination with the fault development stage and the time range in which the failure is likely to occur, a device component likely to fail, a fault occurrence stage, and an associated impact range are determined to form a device fault prediction result; According to the fault type and the fault development stage in the device fault prediction result, corresponding fault response measure data in the equipment component knowledge base sub-library is called to generate a fault response suggestion matched with the fault prediction result, the fault response suggestion including an intervention operation step and an operation timing suggestion.

9. A multi-device data fusion fault prediction system based on Internet of Things collection, characterized in that, The method comprises a processor and a computer readable storage medium, the computer readable storage medium stores machine executable instructions, and the machine executable instructions are executed by the processor to implement the multi-equipment data fusion fault prediction method based on Internet of Things collection in any one of claims 1-8.

Images