A multi-dimensional AI self-checking and fault prediction method, device, equipment and medium

Abstract

The present application relates to the technical field of information processing, and particularly relates to a multi-dimensional AI self-checking and fault prediction method, device, equipment and medium, which comprises the following steps: inputting multi-source detection data into a preset lightweight analysis model for analysis and processing to obtain an environmental abnormal change sequence, a security and fire safety abnormal change sequence and a communication abnormal change sequence; determining target self-checking control information based on the abnormal amplitude, duration and change frequency in the environmental abnormal change sequence, the security and fire safety abnormal change sequence and the communication abnormal change sequence; controlling the corresponding detection equipment to perform target self-checking according to the target self-checking control information to obtain self-checking feedback data; determining fault prediction result information based on the target fault characteristic information and the same type fault evolution track in the historical fault data; and outputting alarm information when the fault prediction result information meets a preset condition.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a multi-dimensional AI self-inspection and fault prediction method, apparatus, equipment and medium. Background Technology

[0002] With the development of intelligent monitoring technology, various detection devices have been widely used in industrial operations, warehouse management, park monitoring, and comprehensive security scenarios to continuously monitor the operating status, environmental status, and connectivity status of target objects. Real-time collection and analysis of detection data can promptly identify anomalies and provide a basis for equipment maintenance, operation scheduling, and risk warning.

[0003] In existing technologies, the identification of the operational status of monitored objects mostly relies on manual inspection, threshold judgment, or single-rule analysis. These methods typically focus on independent analysis of single-type monitoring data, lacking comprehensive utilization of the correlations between multi-source monitoring data. Therefore, when faced with multi-dimensional, continuous, and dynamic anomalies, they often struggle to promptly identify anomaly trends and accurately determine the source of the anomaly. Especially when there are a large number of monitored objects, complex data types, and frequent changes in operating conditions, traditional single-point monitoring methods are prone to misjudgments, missed detections, or delayed responses.

[0004] On the other hand, existing fault handling methods are typically based on a reactive approach after an anomaly occurs, meaning that troubleshooting and maintenance are only carried out after obvious fault symptoms are detected, lacking the ability to identify and predict the fault evolution process in advance. Current technologies often fail to fully mine historical detection and fault data, making it difficult to combine the current state with the development patterns of similar historical faults, thus hindering the effective inference of fault occurrence probability and estimated fault time. Therefore, how to utilize multi-source detection data to achieve intelligent self-inspection and combine it with historical fault evolution information to achieve fault prediction has become a pressing technical problem that needs to be solved in related fields. Summary of the Invention

[0005] To address the problem of insufficient detection and analysis capabilities in existing technologies, this application provides a multi-dimensional AI self-inspection and fault prediction method, apparatus, equipment, and medium.

[0006] The above-mentioned objective of this application is achieved through the following technical solution:

[0007] A multi-dimensional AI self-inspection and fault prediction method, comprising:

[0008] Acquire multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer;

[0009] The multi-source detection data is input into a preset lightweight analysis model for analysis and processing to obtain environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences.

[0010] Based on the abnormal amplitude, duration, and frequency of change in the environmental abnormal change sequence, the security and fire protection abnormal change sequence, and the communication abnormal change sequence, the target self-inspection control information is determined.

[0011] According to the target self-inspection control information, the corresponding detection equipment is controlled to perform the target self-inspection to obtain self-inspection feedback data;

[0012] The self-test feedback data and the corresponding abnormal change sequence are jointly compared to obtain the target fault characterization information.

[0013] Based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, the fault prediction result information is determined.

[0014] When the fault prediction result information meets the preset conditions, an alarm message is output.

[0015] By adopting the above technical solution, multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer can be uniformly acquired and collaboratively analyzed. With the help of a lightweight analysis model, environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences are extracted from the multi-source detection data. Then, based on the anomaly amplitude, duration, and frequency of change, target self-test control information is determined. Then, the corresponding detection equipment is controlled to perform target self-test according to the target self-test control information. The joint comparison of self-test feedback data and anomaly change sequences forms target fault characterization information. Furthermore, based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, fault prediction result information is determined. Thus, a coherent processing link is formed between anomaly identification, self-test control, fault characterization, and fault prediction, improving the intelligence level of the detection process and the fault prediction capability.

[0016] Preferably, the step of inputting the multi-source detection data into a preset lightweight analysis model for analysis and processing to obtain anomaly change sequences in the environment, security and fire protection, and communication includes:

[0017] The multi-source detection data are classified and organized according to the collection timestamp and detection dimension to obtain environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data.

[0018] The environmental perception layer detection data, the security and fire protection perception layer detection data, and the communication connection layer detection data are segmented according to a preset analysis time window to obtain multiple continuous detection data segments.

[0019] Each of the continuous detection data segments is input into the lightweight analysis model to obtain the corresponding offset feature value and difference feature value;

[0020] The offset feature value and the difference feature value are combined to represent the abnormal fluctuation identification value;

[0021] The abnormal fluctuation identifier values ​​of the environmental perception layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain an environmental abnormal change sequence. The abnormal fluctuation identifier values ​​of the security and fire protection perception layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain a security and fire protection abnormal change sequence. The abnormal fluctuation identifier values ​​of the communication connection layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain a communication abnormal change sequence.

[0022] By adopting the above technical solution, multi-source detection data can be first classified and organized according to the collection timestamp and detection dimension. Then, multiple continuous detection data segments are formed according to the preset analysis time window. The lightweight analysis model is used to extract offset feature values ​​and difference feature values ​​for each continuous detection data segment. Subsequently, the offset feature values ​​and difference feature values ​​are combined to form abnormal fluctuation identification values. Furthermore, environmental abnormal change sequences, security and fire protection abnormal change sequences, and communication abnormal change sequences are constructed in chronological order. This makes the analysis process of multi-source detection data more hierarchical and temporal, improves the accuracy of abnormal change characterization and the effectiveness of subsequent abnormal identification.

[0023] Preferably, determining the target self-inspection control information based on the abnormal amplitude, duration, and frequency of change in the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence includes:

[0024] Within a preset judgment period, the abnormal amplitude, duration, and frequency of change of the environmental abnormal change sequence, the security and fire protection abnormal change sequence, and the communication abnormal change sequence are extracted respectively.

[0025] By combining and characterizing the abnormal amplitude, the duration, and the frequency of change, we can obtain the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value, and the communication self-inspection intensity value.

[0026] The environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence are time-series aligned, and synchronization fluctuation analysis is performed on each aligned anomaly change sequence to obtain dimension correlation correction values.

[0027] Based on the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value, the communication self-inspection intensity value, and the dimension correlation correction value, the self-inspection triggering order and self-inspection execution method corresponding to each detection dimension are determined;

[0028] The self-test triggering sequence and self-test execution method corresponding to each detection dimension are combined to obtain the target self-test control information.

[0029] By adopting the above technical solution, the abnormal amplitude, duration, and frequency of change corresponding to the abnormal environmental change sequence, the abnormal security and fire protection change sequence, and the abnormal communication change sequence can be extracted within a preset judgment period. The abnormal amplitude, duration, and frequency of change are then combined to form environmental self-inspection intensity values, security and fire protection self-inspection intensity values, and communication self-inspection intensity values. Simultaneously, by performing time-series alignment processing and synchronous fluctuation analysis on the environmental, security and fire protection, and communication abnormal change sequences, dimension correlation correction values ​​are obtained. Furthermore, by combining the environmental self-inspection intensity values, security and fire protection self-inspection intensity values, communication self-inspection intensity values, and dimension correlation correction values, the self-inspection triggering sequence and self-inspection execution method corresponding to each detection dimension are determined, and these are combined to form target self-inspection control information. This allows the generation of target self-inspection control information to not only reflect the degree of abnormal change in a single detection dimension but also the synchronous change relationship between different detection dimensions, improving the pertinence and coordination of self-inspection arrangements.

[0030] Preferably, the step of controlling the corresponding detection device to perform target self-testing according to the target self-testing control information to obtain self-testing feedback data includes:

[0031] The target self-inspection control information is parsed to obtain the environmental self-inspection control content corresponding to the environmental perception layer, the security and fire protection self-inspection control content corresponding to the security and fire protection perception layer, and the communication self-inspection control content corresponding to the communication connection layer.

[0032] According to the self-test triggering sequence corresponding to the target self-test control information, the detection device corresponding to the control environment perception layer performs environmental self-test according to the environmental self-test control content and obtains environmental self-test feedback data; the detection device corresponding to the control security and fire protection perception layer performs security and fire protection self-test according to the security and fire protection self-test control content and obtains security and fire protection self-test feedback data; the detection device corresponding to the control communication connection layer performs communication self-test according to the communication self-test control content and obtains communication self-test feedback data.

[0033] The environmental self-inspection feedback data is compared with the environmental abnormal change sequence, the security and fire protection self-inspection feedback data is compared with the security and fire protection abnormal change sequence, and the communication self-inspection feedback data is compared with the communication abnormal change sequence to obtain the abnormal continuation result;

[0034] If any of the above-mentioned abnormal continuation results indicate that there is abnormal continuation in the corresponding detection dimension, adjust the self-inspection control content corresponding to the other detection dimensions to obtain the adjusted self-inspection control content, and control the corresponding detection device to perform target self-inspection according to the adjusted self-inspection control content;

[0035] The self-inspection feedback data of the environment, the self-inspection feedback data of the security and fire protection, and the self-inspection feedback data of the communication are integrated to obtain the self-inspection feedback data.

[0036] By adopting the above technical solution, the target self-inspection control information can be analyzed first to obtain the environmental self-inspection control content corresponding to the environmental perception layer, the security and fire protection self-inspection control content corresponding to the security and fire protection perception layer, and the communication self-inspection control content corresponding to the communication connection layer. Then, according to the self-inspection trigger sequence corresponding to the target self-inspection control information, the corresponding detection equipment is controlled to perform environmental self-inspection, security and fire protection self-inspection, and communication self-inspection, and environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are generated. Subsequently, by comparing the environmental self-inspection feedback data with the environmental anomaly change sequence, the security and fire protection self-inspection data is analyzed. The self-inspection feedback data is compared with the security and fire protection anomaly change sequence, and the communication self-inspection feedback data is compared with the communication anomaly change sequence to obtain the anomaly continuation result. If any anomaly continuation result indicates that there is an anomaly continuation in the corresponding detection dimension, the self-inspection control content corresponding to the other detection dimensions is adjusted and the target self-inspection is executed again. Finally, the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are integrated to form self-inspection feedback data. This allows the self-inspection execution process to be dynamically adjusted according to the anomaly continuation situation, improving the adaptability of the self-inspection process and the completeness of the feedback results.

[0037] Preferably, the step of jointly comparing the self-test feedback data and the corresponding abnormal change sequence to obtain target fault characterization information includes:

[0038] The self-test feedback data is classified and analyzed to obtain the environmental self-test feedback data, the security and fire protection self-test feedback data, and the communication self-test feedback data.

[0039] The environmental self-inspection feedback data is compared with the environmental anomaly change sequence to determine the environmental anomaly reduction result; the security and fire protection self-inspection feedback data is compared with the security and fire protection anomaly change sequence to determine the security and fire protection anomaly reduction result; the communication self-inspection feedback data is compared with the communication anomaly change sequence to determine the communication anomaly reduction result.

[0040] Based on the environmental anomaly reduction results, the security and fire protection anomaly reduction results, and the communication anomaly reduction results, residual anomaly values ​​and anomaly reduction directions are extracted respectively.

[0041] By performing cross-combination analysis on each of the residual outliers and each of the outlier reduction directions, the correlation comparison results between each detection dimension are obtained.

[0042] The target fault characterization information is obtained by jointly mapping the correlation comparison results.

[0043] By adopting the above technical solution, the self-inspection feedback data can be classified and analyzed to obtain environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data. Then, the environmental self-inspection feedback data is compared with the environmental anomaly change sequence to determine the environmental anomaly reduction result, the security and fire protection self-inspection feedback data is compared with the security and fire protection anomaly change sequence to determine the security and fire protection anomaly reduction result, and the communication self-inspection feedback data is compared with the communication anomaly change sequence to determine the communication anomaly reduction result. Subsequently, based on the environmental anomaly reduction result, the security and fire protection anomaly reduction result, and the communication anomaly reduction result, residual anomalies and anomaly reduction directions are extracted respectively. By cross-combining and analyzing each residual anomaly value and each anomaly reduction direction, the correlation comparison results between each detection dimension are obtained. Furthermore, the correlation comparison results are jointly mapped to form target fault characterization information, so that the correspondence between self-inspection feedback data and anomaly change sequence can be more detailed, improving the completeness of the target fault characterization information in representing multi-dimensional abnormal states.

[0044] Preferably, determining the fault prediction result information based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data includes:

[0045] The target fault characterization information is expanded to obtain a fault state characterization sequence;

[0046] The fault state representation sequence is compared with the evolution trajectories of the same type of fault in the historical fault data to obtain a cluster of candidate evolution trajectories;

[0047] Stage matching is performed on each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster to determine the target fault evolution stage and the stage evolution rate;

[0048] Based on the target failure evolution stage and the stage evolution rate, the remaining evolution time to reach the preset failure state is estimated.

[0049] Based on the candidate evolution trajectory cluster, the target fault evolution stage, and the remaining evolution duration, the probability of fault occurrence and the expected fault time are determined, and the fault prediction result information is obtained.

[0050] By adopting the above technical solution, the target fault characterization information can be expanded and processed to form a fault state characterization sequence. Then, the fault state characterization sequence is compared with the evolution trajectories of similar faults in historical fault data to obtain candidate evolution trajectory clusters. Subsequently, the evolution trajectories of similar faults in the candidate evolution trajectory clusters are matched in stages to determine the target fault evolution stage and the stage evolution rate. Based on the target fault evolution stage and the stage evolution rate, the remaining evolution time to reach the preset failure state is extrapolated. Furthermore, the probability of fault occurrence and the expected fault time are determined according to the candidate evolution trajectory clusters, the target fault evolution stage, and the remaining evolution time, forming fault prediction result information. This allows the fault prediction process to continuously extrapolate by combining the correspondence between the current target fault characterization information and the evolution trajectories of similar faults in history, improving the accuracy of the fault prediction result information in characterizing the fault development state.

[0051] Preferably, the step of performing stage matching on each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster to determine the target fault evolution stage and the stage evolution rate includes:

[0052] The fault state representation sequence is subjected to transition identification to obtain stage transition features;

[0053] Each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster is segmented to obtain trajectory segmentation features;

[0054] The stage transition features are matched with the trajectory segmentation features, and the corresponding stage duration intervals are extracted based on the matching results to obtain the stage matching results corresponding to each of the same type of fault evolution trajectories.

[0055] The stage matching results are filtered to determine the target fault evolution stage;

[0056] Extract the state change span and time change span corresponding to the target fault evolution stage, and determine the stage evolution rate based on the ratio of the state change span to the time change span.

[0057] By adopting the above technical solution, the fault state characterization sequence can be first identified to obtain stage transition features. Then, the evolution trajectories of the same type in the candidate evolution trajectory cluster can be segmented to obtain trajectory segmentation features. Subsequently, the stage transition features are matched with the trajectory segmentation features, and the corresponding stage duration intervals are extracted based on the matching results to form stage matching results corresponding to each type of fault evolution trajectory. The stage matching results are further filtered to determine the target fault evolution stage. By extracting the state change span and time change span corresponding to the target fault evolution stage, the stage evolution rate is determined based on the ratio of the state change span to the time change span. Thus, the identification process of the fault evolution stage can correspond to the stage change features in the historical fault evolution trajectory, improving the accuracy of determining the target fault evolution stage and the stage evolution rate.

[0058] The second objective of this invention is achieved through the following technical solution:

[0059] A multi-dimensional AI self-inspection and fault prediction device, the multi-dimensional AI self-inspection and fault prediction device comprising:

[0060] The multi-source detection data acquisition module is used to acquire multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer.

[0061] An abnormal change sequence generation module is used to input the multi-source detection data into a preset lightweight analysis model for analysis and processing, and to obtain environmental abnormal change sequences, security and fire protection abnormal change sequences, and communication abnormal change sequences.

[0062] The target self-test control module is used to determine target self-test control information based on the abnormal amplitude, duration and frequency of change in the environmental abnormal change sequence, the security and fire protection abnormal change sequence and the communication abnormal change sequence.

[0063] The target self-test execution module is used to control the corresponding detection equipment to perform target self-test according to the target self-test control information, and obtain self-test feedback data.

[0064] The target fault characterization module is used to jointly compare the self-test feedback data and the corresponding abnormal change sequence to obtain target fault characterization information.

[0065] The fault prediction module is used to determine the fault prediction result information based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data.

[0066] The alarm output module is used to output alarm information when the fault prediction result information meets preset conditions.

[0067] By adopting the above technical solution, multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer can be uniformly acquired and collaboratively analyzed. With the help of a lightweight analysis model, environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences are extracted from the multi-source detection data. Then, based on the anomaly amplitude, duration, and frequency of change, target self-test control information is determined. Then, the corresponding detection equipment is controlled to perform target self-test according to the target self-test control information. The joint comparison of self-test feedback data and anomaly change sequences forms target fault characterization information. Furthermore, based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, fault prediction result information is determined. Thus, a coherent processing link is formed between anomaly identification, self-test control, fault characterization, and fault prediction, improving the intelligence level of the detection process and the fault prediction capability.

[0068] The above-mentioned objective three of this application is achieved through the following technical solution:

[0069] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the aforementioned multi-dimensional AI self-testing and fault prediction method.

[0070] The fourth objective of this application is achieved through the following technical solution:

[0071] A computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the aforementioned multi-dimensional AI self-inspection and fault prediction method.

[0072] In summary, this application includes at least one of the following beneficial technical effects:

[0073] It can uniformly acquire and collaboratively analyze multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer. With the help of a lightweight analysis model, it extracts environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences from the multi-source detection data. Then, based on the anomaly amplitude, duration, and frequency of change, it determines the target self-test control information. Then, it controls the corresponding detection equipment to perform the target self-test according to the target self-test control information. The joint comparison of self-test feedback data and anomaly change sequences forms the target fault characterization information. Furthermore, based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, it determines the fault prediction result information. Thus, a coherent processing link is formed between anomaly identification, self-test control, fault characterization, and fault prediction, improving the intelligence level of the detection process and the fault prediction capability. Attached Figure Description

[0074] Figure 1This is a flowchart of a multi-dimensional AI self-inspection and fault prediction method in one embodiment of this application.

[0075] Figure 2 This is a principle block diagram of a multi-dimensional AI self-inspection and fault prediction device in one embodiment of this application. Detailed Implementation

[0076] The present application will be further described in detail below with reference to the accompanying drawings.

[0077] In one embodiment, such as Figure 1 As shown, this application discloses a multi-dimensional AI self-inspection and fault prediction method, which specifically includes the following steps:

[0078] S10: Acquire multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer.

[0079] In this embodiment, the environmental perception layer refers to the data source layer used to characterize the environmental state within the target area, the security and fire protection perception layer refers to the data source layer used to characterize the security and fire protection state within the target area, the communication connection layer refers to the data source layer used to characterize the connection state between each detection node, and the multi-source detection data refers to the detection data that comes from the environmental perception layer, the security and fire protection perception layer, and the communication connection layer, respectively.

[0080] Specifically, when acquiring multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer, the data source range corresponding to each layer is first determined according to the pre-set data source correspondence. Then, based on each data source range, the data content formed by the corresponding detection node within the current acquisition period is read. Specifically, when reading the data content corresponding to the environmental perception layer, the environmental status record content is extracted from the corresponding storage location according to the identification information of the environmental perception detection node. When reading the data content corresponding to the security and fire protection perception layer, the security status record content and fire protection status record content are extracted from the corresponding storage location according to the identification information of the security and fire protection detection nodes. When reading the data content corresponding to the communication connection layer, the connection status record content is extracted according to the communication identification relationship between the connection nodes. After completing the extraction of the corresponding data content for each layer, the data content corresponding to the environmental perception layer, the security and fire protection perception layer, and the communication connection layer are then organized according to the acquisition time to form multi-source detection data corresponding to the environmental perception layer, the security and fire protection perception layer, and the communication connection layer.

[0081] S20: Input the multi-source detection data into the preset lightweight analysis model for analysis and processing to obtain the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence.

[0082] In this embodiment, the lightweight analysis model refers to an analysis model that performs feature recognition and anomaly detection on multi-source detection data. The "lightweight" in the lightweight analysis model means that by controlling the number of model parameters, the number of computational levels, and the length of data processing paths, the lightweight analysis model can complete the analysis and processing of multi-source detection data under limited computing resources. The environmental anomaly change sequence refers to a time-series data set formed based on the multi-source detection data corresponding to the environmental perception layer, used to characterize the process of abnormal changes in environmental state. The security and fire protection anomaly change sequence refers to a time-series data set formed based on the multi-source detection data corresponding to the security and fire protection perception layers, used to characterize the process of abnormal changes in security and fire protection states. The communication anomaly change sequence refers to a time-series data set formed based on the multi-source detection data corresponding to the communication connection layer, used to characterize the process of abnormal changes in connection state.

[0083] Specifically, when inputting multi-source detection data into a preset lightweight analysis model for analysis and processing, the multi-source detection data is first organized according to the acquisition time sequence, so that the data content corresponding to the environmental perception layer, the security and fire protection perception layer, and the communication connection layer are respectively formed into continuous input data segments. Then, each continuous input data segment is sequentially written into the corresponding input position of the lightweight analysis model. The lightweight analysis model then performs identification processing on the data changes in each continuous input data segment according to preset analysis rules. During the identification processing, the data values ​​within the current time period and adjacent data are first read from each continuous input data segment. The data changes between time periods and the data differences between different detection objects in the same layer are analyzed. Based on the read data values, data changes, and data differences, corresponding abnormal fluctuation identification results are formed. Then, the abnormal fluctuation identification results are classified and organized according to the detection dimensions. The abnormal fluctuation identification results corresponding to the environmental perception layer are arranged in chronological order to form an environmental abnormal change sequence. The abnormal fluctuation identification results corresponding to the security and fire protection perception layer are arranged in chronological order to form a security and fire protection abnormal change sequence. The abnormal fluctuation identification results corresponding to the communication connection layer are arranged in chronological order to form a communication abnormal change sequence.

[0084] S30: Determine the target self-inspection control information based on the abnormal amplitude, duration, and frequency of change in the abnormal environmental change sequence, the abnormal security and fire protection change sequence, and the abnormal communication change sequence.

[0085] In this embodiment, abnormal amplitude refers to the magnitude of change between abnormal fluctuation identifier values ​​corresponding to adjacent time periods in the abnormal change sequence, duration refers to the length of time that the abnormal change sequence is continuously in an abnormal state, change frequency refers to the number of times the abnormal change sequence exhibits abnormal changes within a preset judgment interval, and target self-check control information refers to information used to characterize the self-check arrangement content corresponding to each detection dimension.

[0086] Specifically, when determining the target self-inspection control information based on the abnormal amplitude, duration, and frequency of changes in environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences, the abnormal fluctuation identifier values ​​corresponding to these sequences within a preset judgment interval are first read. Let the abnormal fluctuation identifier value corresponding to any anomaly change sequence within the preset judgment interval be represented in chronological order as follows: Where ai represents the abnormal fluctuation identifier value corresponding to the i-th time period, and n represents the number of time periods within the preset judgment interval. Subsequently, the abnormal change sequence is traversed. When calculating the abnormal amplitude, the change difference between the abnormal fluctuation identifier values ​​corresponding to adjacent time periods is first calculated. Then, the weighted summation of each change difference is performed to obtain the abnormal amplitude. ,in, This represents the temporal progressive weight corresponding to the i-th time period, used to distinguish the difference between changes in the preceding and following segments within a preset judgment interval. When calculating the duration, it first identifies the abnormal change sequence that continuously satisfies... The abnormal time period interval, among which, This represents the preset anomaly detection threshold. The duration corresponding to each anomaly time interval is then denoted as dj, where j represents the j-th consecutive anomaly interval. The maximum value among these durations is then taken as the duration of the anomaly. When calculating the frequency of change, for abnormal change sequences that satisfy... and The number of state transitions is counted to obtain the frequency of change. , among which, when hour, ,otherwise, After extracting the abnormal amplitude, duration, and frequency of change, the abnormal amplitude, duration, and frequency of change are combined and characterized to obtain the self-inspection strength value of the corresponding abnormal change sequence. Where Q represents the self-inspection intensity value of the corresponding detection dimension, and e represents the natural constant. Subsequently, the self-inspection intensity values ​​corresponding to the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence are classified and organized according to the source identifiers corresponding to each anomaly change sequence, forming the environmental self-inspection arrangement content, the security and fire protection self-inspection arrangement content, and the communication self-inspection arrangement content. Finally, the environmental self-inspection arrangement content, the security and fire protection self-inspection arrangement content, and the communication self-inspection arrangement content are merged and organized to form the target self-inspection control information.

[0087] S40: Control the corresponding testing equipment to perform target self-test according to the target self-test control information and obtain self-test feedback data.

[0088] In this embodiment, self-test feedback data refers to the feedback result data generated by the corresponding detection device after performing target self-test based on the target self-test control information.

[0089] Specifically, according to the target self-inspection control information, the corresponding detection equipment is controlled to perform target self-inspection. When obtaining self-inspection feedback data, the target self-inspection control information is first read, and the environmental self-inspection arrangement content, security and fire protection self-inspection arrangement content, and communication self-inspection arrangement content are extracted from the target self-inspection control information. Then, the corresponding detection equipment is controlled sequentially according to the self-inspection trigger order contained in the target self-inspection control information. Specifically, when controlling the detection equipment corresponding to the environmental perception layer to perform environmental self-inspection, the target reading range and target acquisition number of the detection equipment corresponding to the environmental perception layer are first determined according to the environmental self-inspection arrangement content. Then, the output data of the detection equipment corresponding to the environmental perception layer is repeatedly read according to the target reading range, and the reading results are organized according to the acquisition sequence to form environmental self-inspection feedback data. When controlling the detection equipment corresponding to the security and fire protection perception layer to perform security and fire protection self-inspection, the target detection range and target response number of the detection equipment corresponding to the security and fire protection perception layer are first determined according to the security and fire protection self-inspection arrangement content. The system continuously reads the output data of the corresponding detection devices in the security and fire protection sensing layer according to the target detection interval, and organizes the reading results in the order of detection to form security and fire protection self-inspection feedback data. When the corresponding detection devices in the control communication connection layer perform communication self-inspection, the target connection interval and target interaction number of the corresponding detection devices in the communication connection layer are first determined according to the communication self-inspection arrangement. Then, the connection status data of the corresponding detection devices in the communication connection layer are read cyclically according to the target connection interval, and the reading results are organized in the order of interaction to form communication self-inspection feedback data. After acquiring the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data, the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are organized according to a unified time benchmark. Let the feedback result sequences corresponding to the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data be denoted as follows: , as well as Where ei represents the i-th environmental feedback result, sj represents the j-th security and fire protection feedback result, ck represents the k-th communication feedback result, and m, p, and q represent the number of corresponding feedback results. Then, each feedback result is mapped to its corresponding time position according to a unified time base, and then... Feedback results from the same time location are integrated, where R(t) represents the integrated feedback result at time location t, E(t) represents the environmental feedback result value at time location t, S(t) represents the security and fire protection feedback result value at time location t, and C(t) represents the communication feedback result value at time location t. , and This represents the feedback adjustment coefficient for the corresponding detection dimension, and satisfies... Finally, the integrated feedback results corresponding to each time position are arranged in chronological order to form self-inspection feedback data.

[0090] S50: The self-test feedback data and the corresponding abnormal change sequence are jointly compared to obtain the target fault characterization information.

[0091] In this embodiment, the target fault characterization information refers to the data content used to characterize abnormal state changes, formed by the correspondence between self-test feedback data and the corresponding abnormal change sequence.

[0092] Specifically, when jointly comparing the self-inspection feedback data and the corresponding abnormal change sequences to obtain the target fault characterization information, the self-inspection feedback data is first split and organized according to the detection dimensions. The self-inspection feedback data is then split into environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data. Next, the environmental self-inspection feedback data is correlated with environmental abnormal change sequences according to a unified time benchmark, the security and fire protection self-inspection feedback data is correlated with security and fire protection abnormal change sequences according to a unified time benchmark, and the communication self-inspection feedback data is correlated with communication abnormal change sequences according to a unified time benchmark. For any detection dimension, let the corresponding abnormal change sequence be... The corresponding self-test feedback data sequence is } where ai represents the abnormal change value corresponding to the i-th time position, bi represents the feedback change value corresponding to the i-th time position, and n represents the number of corresponding time positions. After completing the time mapping, the difference between the abnormal change value and the feedback change value at each time position is calculated to obtain the comparison deviation value of the corresponding time position. Then, the direction of each comparison deviation value is identified. When bi > ai, the corresponding time position is recorded as the anomaly reduction direction; when bi = ai, the corresponding time position is recorded as the anomaly maintenance direction; and when bi > ai, the corresponding time position is recorded as the anomaly increase direction. Then, residual anomalies are extracted based on the comparison deviation value and direction identification results corresponding to each time position. The residual anomalies are then processed through... To determine, where ri represents the residual outlier corresponding to the i-th time position. This represents the direction adjustment factor, which is applied when the corresponding time position is in the direction of anomaly reduction. When the corresponding time position is in the direction of abnormal maintenance, When the corresponding time position is in the direction of abnormal increase, After extracting the residual outliers corresponding to each detection dimension, a cross-combination analysis is performed on the residual outliers corresponding to each detection dimension. The cross-combination analysis refers to reading the residual outliers corresponding to environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data at the same time and location, and calculating the joint anomaly characterization value. Where Gi represents the joint anomaly representation value corresponding to the i-th time position, This represents the residual anomaly value of the environment corresponding to the i-th time location. This represents the residual anomaly value of security and fire protection corresponding to the i-th time position. This represents the residual anomaly value corresponding to the i-th time position. Finally, the joint anomaly characterization value corresponding to each time position, the direction recognition result of the corresponding detection dimension, and the residual anomaly value of the corresponding detection dimension are sorted in chronological order to form the target fault characterization information.

[0093] S60: Based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, determine the fault prediction result information.

[0094] In this embodiment, the similar fault evolution trajectory refers to the trajectory data formed in historical fault data that is consistent with the fault category of the target fault characterization information and can characterize the continuous change process of the fault state over time. The fault prediction result information refers to the data content corresponding to the fault occurrence probability and the expected fault time formed based on the correspondence between the target fault characterization information and the similar fault evolution trajectory.

[0095] Specifically, when determining the fault prediction result information based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, the joint anomaly characterization value, the direction recognition result of the corresponding detection dimension, and the residual anomaly value of the corresponding detection dimension are first extracted from the target fault characterization information in chronological order. These are then expanded and organized according to a unified time step to form a target fault state sequence, denoted as [the target fault state sequence is then described]. Where zi represents the target fault state value corresponding to the i-th time position, and n represents the number of time positions corresponding to the target fault state sequence. Then, multiple similar fault evolution trajectories consistent with the fault category of the target fault characterization information are read from historical fault data. Let any similar fault evolution trajectory be represented as... ,in, Let mk represent the historical fault state value corresponding to the j-th time position of the k-th similar fault evolution trajectory, and mk represent the number of time positions corresponding to the k-th similar fault evolution trajectory. Then, the target fault state sequence is aligned and compared with each similar fault evolution trajectory using a sliding window of the same length. For any k-th similar fault evolution trajectory, a continuous sub-trajectory of length n is extracted. Where u represents the starting position of the window in the k-th similar fault evolution trajectory, the trajectory deviation value between the target fault state sequence and the continuous sub-trajectories is then calculated. ,in, This represents a time-progressive adjustment term used to distinguish the state differences between the beginning and end of the target fault state sequence. Then, each trajectory deviation value is filtered, and the continuous sub-trajectory corresponding to the smallest trajectory deviation value is taken as the target's corresponding trajectory segment. The end position of the target's corresponding trajectory segment in the same type of fault evolution trajectory is determined as the current corresponding position. Subsequently, the remaining trajectory length between the current corresponding position and the preset failure state corresponding position is read to obtain the remaining evolution time. Where tf represents the historical time position corresponding to the preset failure state, and tc represents the historical time position corresponding to the current position. At the same time, the probability of failure is determined based on the trajectory deviation value. Where P represents the probability of a failure occurring. } represents the minimum trajectory deviation value. Represents the bias sensitivity coefficient. This represents the baseline value for probability conversion. After calculating the probability of failure and the remaining evolution time, the current time is converted to the remaining evolution time to determine the expected failure time. The probability of failure and the expected failure time are then organized according to a preset data format to form the failure prediction result information.

[0096] S70: When the fault prediction result information meets the preset conditions, output alarm information.

[0097] In this embodiment, alarm information refers to abnormal prompt data generated when the fault prediction result information meets preset conditions.

[0098] Specifically, when an alarm is output if the fault prediction result meets preset conditions, the fault occurrence probability and expected fault time are first read from the fault prediction result. The fault occurrence probability is then compared with a preset probability threshold, and the expected fault time is compared with a preset time threshold. The preset conditions refer to the fault occurrence probability being greater than or equal to the preset probability threshold, or the time interval between the expected fault time and the current time being less than or equal to the preset time threshold. Let the fault occurrence probability be P, the preset probability threshold be P0, the expected fault time be ty, the current time be tn, and the preset time threshold be... The decision function corresponding to the preset condition is expressed as follows: , among which, when hour, ,otherwise, ,when hour, ,otherwise, When J=1, it is determined that the fault prediction result information meets the preset conditions. After determining that the fault prediction result information meets the preset conditions, the fault occurrence probability and expected fault time in the fault prediction result information are formatted and written into the alarm data content according to the preset field order. Specifically, when writing, the probability field corresponding to the fault occurrence probability is written first, and then the time field corresponding to the expected fault time is written. Subsequently, the alarm data content with completed field writing is output as alarm information.

[0099] In one embodiment, in step S20, the multi-source detection data is input into a preset lightweight analysis model for analysis and processing to obtain an environmental anomaly change sequence, a security and fire protection anomaly change sequence, and a communication anomaly change sequence, including:

[0100] S201: The multi-source detection data is classified and organized according to the collection timestamp and detection dimension to obtain environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data.

[0101] In this embodiment, the acquisition timestamp refers to the time stamp information used to characterize the corresponding data acquisition time in the multi-source detection data; the detection dimension refers to the classification basis used to distinguish the data source category to which the multi-source detection data belongs; the environmental perception layer detection data refers to the data content corresponding to the environmental perception layer selected from the multi-source detection data; the security and fire protection perception layer detection data refers to the data content corresponding to the security and fire protection perception layer selected from the multi-source detection data; and the communication connection layer detection data refers to the data content corresponding to the communication connection layer selected from the multi-source detection data.

[0102] Specifically, when classifying and organizing multi-source detection data according to acquisition timestamps and detection dimensions to obtain environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data, the acquisition timestamp and detection dimension markers corresponding to each data record in the multi-source detection data are first read one by one. Then, the data category to which each data record belongs is determined based on the detection dimension markers. Specifically, when the detection dimension marker indicates that the corresponding data record originates from the environmental status detection range, the corresponding data record is classified as environmental perception layer detection data. When the detection dimension marker indicates that the corresponding data record originates from the security status detection range or the fire protection status detection range, the corresponding data record is classified as security and fire protection perception layer detection data. When the detection dimension marker indicates that the corresponding data record originates from the connection status detection range, the corresponding data record is classified as communication connection layer detection data. After completing the classification by detection dimension, each data record in the environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data is sorted according to the acquisition timestamp. Let the set of data records corresponding to any detection dimension be represented as... Where mi represents the i-th data record, and n represents the number of data records in the corresponding detection dimension. If the collection timestamp corresponding to the i-th data record is ti, then according to... The data records are arranged in the following order. After the arrangement is completed, the set of data records corresponding to the environmental perception layer is used as the environmental perception layer detection data, the set of data records corresponding to the security and fire protection perception layer is used as the security and fire protection perception layer detection data, and the set of data records corresponding to the communication connection layer is used as the communication connection layer detection data.

[0103] S202: The environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data are segmented according to a preset analysis time window to obtain multiple continuous detection data segments.

[0104] In this embodiment, the preset analysis time window refers to the time interval used to limit the time range covered by a single analysis, and the continuous detection data segment refers to the set of data records arranged continuously in chronological order of collection time within the range corresponding to the preset analysis time window.

[0105] Specifically, the environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data are segmented according to a preset analysis time window to obtain multiple continuous detection data segments. First, the acquisition timestamps corresponding to each data record in the environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data are read. Then, the segment boundaries are determined based on the start time length and sliding time length of the preset analysis time window. Here, the start time length of the preset analysis time window refers to the continuous time range covered by a single time window, and the sliding time length refers to the interval between the starting positions of two adjacent time windows. Let the set of data records corresponding to any detection dimension be represented as... The collection timestamp corresponding to each data record is represented as: Where mi represents the i-th data record, ti represents the collection timestamp corresponding to the i-th data record, and n represents the number of data records under the corresponding detection dimension. If the preset analysis time window length is denoted as L and the sliding time length is denoted as S, then the time interval corresponding to the k-th preset analysis time window is denoted as... Where t0 represents the starting acquisition timestamp in the data record corresponding to the current detection dimension, and k represents the time window number. Then, each preset analysis time window is traversed sequentially, and data records whose acquisition timestamps fall within the corresponding time interval are extracted and arranged in ascending order of acquisition timestamps to form the k-th continuous detection data segment. If the set of data records contained within a certain time interval is represented as... Then, the data record set Dk is recorded as a continuous detection data segment under the corresponding detection dimension. The detection data of the environmental perception layer, the security and fire protection perception layer, and the communication connection layer are extracted by window sliding and arranged in time order according to the above method, and finally multiple continuous detection data segments are obtained.

[0106] S203: Input each continuous detection data segment into the lightweight analysis model to obtain the corresponding offset feature value and difference feature value.

[0107] In this embodiment, the offset feature value refers to the feature value output by the lightweight analysis model based on the degree of deviation of each data record within the continuous detection data segment from the overall change center of the continuous detection data segment, and the difference feature value refers to the feature value output by the lightweight analysis model based on the degree of difference between adjacent data records within the continuous detection data segment.

[0108] Specifically, when inputting each continuous detection data segment into the lightweight analysis model to obtain the corresponding offset feature value and difference feature value, the data records in each continuous detection data segment are read segment by segment, and the detection values ​​corresponding to each data record are extracted in chronological order of acquisition time. Let the sequence of detection values ​​corresponding to any continuous detection data segment be represented as... Where xi represents the detection value corresponding to the i-th data record, and n represents the number of data records contained in the continuous detection data segment. Then, the overall change center of the continuous detection data segment is calculated based on the detection value sequence, and the overall change center is represented as... ,in, , This represents the position weight corresponding to the i-th data record. After obtaining the overall change center, the deviation of each detected value relative to the overall change center is calculated one by one, and the deviation is expressed as... Then calculate the difference between adjacent detection values, and the difference is expressed as... ,in, and order After extracting the deviation and difference, the deviation sequence is... and difference sequence The values ​​are written into the input positions of the lightweight analysis model according to their positional correspondences. The lightweight analysis model then performs parameter mapping operations on the deviation sequence and the difference sequence, respectively. The mapping result corresponding to the offset feature value is expressed as follows: The mapping result corresponding to the difference feature values ​​is represented as ,in, This represents the offset parameter corresponding to the i-th deviation. Let represent the difference parameter corresponding to the i-th difference quantity, bp represent the offset mapping base value, and bq represent the difference mapping base value. After completing the parameter mapping operation, compression calculations are performed on the mapping results to obtain the offset eigenvalues. and difference eigenvalues Finally, the offset feature value and difference feature value corresponding to any continuous detection data segment are used as the analysis result corresponding to that continuous detection data segment.

[0109] S204: Combine the offset feature value and the difference feature value to obtain the abnormal fluctuation identification value.

[0110] Specifically, first, the offset feature value and the difference feature value corresponding to any continuous detection data segment are read, and the offset feature value is denoted as fp and the difference feature value as fq. Then, the coupling degree between the offset feature value and the difference feature value is calculated, and the coupling degree is expressed as: Where g represents the synchronous change of the offset feature value and the difference feature value in the same continuous detection data segment. After obtaining the coupling degree, a combination operation is performed on the offset feature value, the difference feature value, and the coupling degree. The combination operation is expressed as follows: Where h represents the intermediate value of the combination, This represents the coupling adjustment coefficient, used to adjust the degree of coupling in the combinatorial operation. After obtaining the intermediate combinatorial value, a deviation correction is applied to the intermediate combinatorial value based on the deviation relationship between the offset eigenvalue and the difference eigenvalue. The deviation correction is expressed as... Where r represents the deviation correction amount, the combined median value and the deviation correction amount are then jointly calculated to obtain the abnormal fluctuation indicator value. Where y represents the abnormal fluctuation identifier value. After the abnormal fluctuation identifier value is calculated, it is written into the identifier position of the corresponding continuous detection data segment as the abnormal characterization result of the continuous detection data segment.

[0111] S205: Arrange the abnormal fluctuation identifier values ​​of each continuous detection data segment corresponding to the environmental perception layer detection data in chronological order to obtain the environmental abnormal change sequence; arrange the abnormal fluctuation identifier values ​​of each continuous detection data segment corresponding to the security and fire protection perception layer detection data in chronological order to obtain the security and fire protection abnormal change sequence; arrange the abnormal fluctuation identifier values ​​of each continuous detection data segment corresponding to the communication connection layer detection data in chronological order to obtain the communication abnormal change sequence.

[0112] In this embodiment, the abnormal fluctuation identifier value corresponding to each continuous detection data segment of the environmental perception layer detection data refers to the abnormal fluctuation identifier value determined separately from the multiple continuous detection data segments formed by the environmental perception layer detection data; the abnormal fluctuation identifier value corresponding to each continuous detection data segment of the security and fire protection perception layer detection data refers to the abnormal fluctuation identifier value determined separately from the multiple continuous detection data segments formed by the security and fire protection perception layer detection data; and the abnormal fluctuation identifier value corresponding to each continuous detection data segment of the communication connection layer detection data refers to the abnormal fluctuation identifier value determined separately from the multiple continuous detection data segments formed by the communication connection layer detection data.

[0113] Specifically, firstly, the continuous detection data segments corresponding to the environmental perception layer detection data, the security and fire protection perception layer detection data, and the communication connection layer detection data are read respectively, and the starting acquisition timestamp and abnormal fluctuation identifier value corresponding to each continuous detection data segment are extracted. Then, the correspondence between the abnormal fluctuation identifier value and the starting acquisition timestamp is established according to the detection dimension. For any detection dimension, if the starting acquisition timestamps corresponding to each continuous detection data segment are sequentially... The abnormal fluctuation flag values ​​corresponding to each continuous detection data segment are as follows: First, sort the starting collection timestamps according to a consistent size order to ensure that the following conditions are met. The order of the data is then determined, and the position order of the corresponding abnormal fluctuation marker values ​​is synchronously adjusted according to the starting collection timestamp after sorting, forming an abnormal fluctuation marker value arrangement result corresponding to the time sequence. Subsequently, the abnormal fluctuation marker value arrangement result corresponding to the environmental perception layer detection data is taken as the environmental abnormal change sequence, the abnormal fluctuation marker value arrangement result corresponding to the security and fire protection perception layer detection data is taken as the security and fire protection abnormal change sequence, and the abnormal fluctuation marker value arrangement result corresponding to the communication connection layer detection data is taken as the communication abnormal change sequence.

[0114] In one embodiment, in step S30, based on the abnormal amplitude, duration, and frequency of change in the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence, the target self-test control information is determined, including:

[0115] S301: Within a preset judgment period, extract the abnormal amplitude, duration and frequency of change corresponding to the abnormal environmental change sequence, the abnormal security and fire protection change sequence, and the abnormal communication change sequence.

[0116] In this embodiment, the preset judgment period refers to a pre-set time range used for statistical analysis of abnormal changes.

[0117] Specifically, firstly, abnormal fluctuation marker values ​​and corresponding time positions falling within a preset judgment period are read from environmental abnormal change sequences, security and fire protection abnormal change sequences, and communication abnormal change sequences. Then, for any abnormal change sequence, each abnormal fluctuation marker value is traversed in chronological order. When extracting the abnormal amplitude, the change difference between abnormal fluctuation marker values ​​corresponding to adjacent time positions is first calculated, and the maximum value among the change differences is determined as the abnormal amplitude of the corresponding abnormal change sequence within the preset judgment period. When extracting the duration, the time intervals in which the abnormal fluctuation marker values ​​are continuously greater than or equal to the preset abnormal judgment threshold are first identified, and then the duration corresponding to each time interval is calculated. The system first determines the duration of the time interval with the longest duration. When extracting the frequency of change, it first determines whether there is a state transition between adjacent time positions where the abnormal fluctuation indicator value changes from less than the preset abnormal judgment threshold to greater than or equal to the preset abnormal judgment threshold. Then, it accumulates the number of times the state transition occurs and determines the frequency of change. After extracting the abnormal amplitude, duration, and frequency of change for any abnormal change sequence, it processes the environmental abnormal change sequence, the security and fire protection abnormal change sequence, and the communication abnormal change sequence in the same way to obtain the abnormal amplitude, duration, and frequency of change for each abnormal change sequence.

[0118] S302: Combine the abnormal amplitude, duration and frequency of change to obtain the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value and the communication self-inspection intensity value.

[0119] In this embodiment, the environmental self-inspection strength value refers to a value formed by combining and characterizing the abnormal amplitude, duration, and frequency of change corresponding to the abnormal environmental change sequence, and is used to characterize the strength of the self-inspection arrangement corresponding to the environmental perception layer. The security and fire protection self-inspection strength value refers to a value formed by combining and characterizing the abnormal amplitude, duration, and frequency of change corresponding to the abnormal security and fire protection change sequence, and is used to characterize the strength of the self-inspection arrangement corresponding to the security and fire protection perception layer. The communication self-inspection strength value refers to a value formed by combining and characterizing the abnormal amplitude, duration, and frequency of change corresponding to the abnormal communication change sequence, and is used to characterize the strength of the self-inspection arrangement corresponding to the communication connection layer.

[0120] Specifically, when combining and characterizing abnormal amplitude, duration, and frequency of change to obtain environmental self-inspection intensity values, security and fire protection self-inspection intensity values, and communication self-inspection intensity values, the abnormal amplitude, duration, and frequency of change corresponding to the abnormal environmental change sequence, security and fire protection change sequence, and communication change sequence are first read. For any abnormal change sequence, the abnormal amplitude is denoted as F, the duration as T, and the frequency of change as C. Then, a combination operation is performed on the abnormal amplitude, duration, and frequency of change. To ensure that the duration and frequency of change progressively enhance the combination result when the abnormal amplitude is large, the combination characterization result is expressed as follows: Where Q represents the self-test strength value of the corresponding anomalous change sequence, This represents the duration characteristic after compression of the duration. This represents the frequency representation term after normalizing the frequency of changes. After completing the combination operation, Q, calculated based on the abnormal amplitude, duration, and frequency of change corresponding to the abnormal environmental change sequence, is determined as the environmental self-inspection intensity value. Q, calculated based on the abnormal amplitude, duration, and frequency of change corresponding to the abnormal security and fire protection change sequence, is determined as the security and fire protection self-inspection intensity value. Q, calculated based on the abnormal amplitude, duration, and frequency of change corresponding to the abnormal communication change sequence, is determined as the communication self-inspection intensity value.

[0121] S303: Perform time-series alignment on the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence, and perform synchronization fluctuation analysis on each aligned anomaly change sequence to obtain the dimension correlation correction value.

[0122] In this embodiment, the dimension correlation correction value refers to the correction value formed based on the synchronous change relationship of the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence at the same time location.

[0123] Specifically, the time positions corresponding to each abnormal fluctuation identifier value in the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence are first read separately. A time interval commonly covered by these three anomaly change sequences is selected as an alignment interval. Then, multiple aligned time positions are established within the alignment interval according to a uniform time step. If any anomaly change sequence does not have an original abnormal fluctuation identifier value at a certain aligned time position, the abnormal fluctuation identifier value closest to the aligned time position is selected from the abnormal fluctuation identifier values ​​corresponding to the adjacent time positions before and after that aligned time position as the mapping value. After mapping, aligned environmental anomaly change sequences, aligned security and fire protection anomaly change sequences, and aligned communication anomaly change sequences are formed respectively. Let the aligned environmental anomaly change sequence be represented as... The aligned sequence of security and fire safety anomalies is represented as follows: The aligned communication anomaly change sequence is represented as follows: Where ei represents the environmental anomaly fluctuation flag value corresponding to the i-th aligned time position, si represents the security and fire protection anomaly fluctuation flag value corresponding to the i-th aligned time position, ci represents the communication anomaly fluctuation flag value corresponding to the i-th aligned time position, and n represents the number of aligned time positions. After completing the time alignment process, the change between adjacent aligned time positions is calculated to obtain the environmental change amount. Changes in security and fire protection and communication changes ,in, Subsequently, a synchronous fluctuation analysis was performed on the environmental changes, security and fire protection changes, and communication changes corresponding to each alignment time position. The synchronous fluctuation analysis refers to determining whether the environmental changes, security and fire protection changes, and communication changes at the same alignment time position are simultaneously greater than their respective change judgment thresholds, and calculating the synchronous fluctuation intensity value. ,in, Represents the synchronization decision factor, when , and When both exceed their respective change judgment thresholds ,otherwise, After calculating the synchronization fluctuation intensity values ​​corresponding to each alignment time position, the synchronization fluctuation intensity values ​​are then summed and averaged to obtain the dimension correlation correction value: , where R represents the dimension correlation correction value.

[0124] S304: Based on the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value, the communication self-inspection intensity value, and the dimension correlation correction value, determine the self-inspection triggering order and self-inspection execution method corresponding to each detection dimension.

[0125] In this embodiment, the self-test triggering order refers to the order in which the environmental perception layer, the security and fire protection perception layer, and the communication connection layer perform self-tests, and the self-test execution method refers to the execution form adopted by the corresponding detection dimension when performing a self-test.

[0126] Specifically, the environmental self-inspection strength value, security and fire protection self-inspection strength value, communication self-inspection strength value, and dimension correlation correction value are first read separately. The environmental self-inspection strength value is denoted as Qe, the security and fire protection self-inspection strength value as Qs, the communication self-inspection strength value as Qc, and the dimension correlation correction value as R. Then, a correlation correction operation is performed on the self-inspection strength values ​​corresponding to each detection dimension to obtain the order judgment value corresponding to each detection dimension. The order judgment value corresponding to the environmental perception layer is represented as follows: The sequence determination value corresponding to the security and fire protection sensing layers is expressed as follows: The order determination value corresponding to the communication connection layer is represented as follows: After calculating the sequence judgment value, Ue, Us, and Uc are sorted from largest to smallest. Detection dimensions with larger values ​​correspond to earlier self-check triggering sequences, and those with smaller values ​​correspond to later self-check triggering sequences. If two detection dimensions have the same sequence judgment value, the original self-check intensity values ​​for both dimensions are read, and the detection dimension with the larger original self-check intensity value is determined as the earlier self-check triggering sequence. After determining the self-check triggering sequence, the corresponding self-check execution method is determined based on the combination relationship between the sequence judgment value and the dimension association correction value for each detection dimension. If the sequence judgment value for any detection dimension is greater than or equal to a preset high-intensity threshold, the self-check execution method for that detection dimension is determined as a high-frequency execution method. If the order judgment value corresponding to any detection dimension is less than the preset high-intensity threshold but greater than or equal to the preset medium-intensity threshold, the self-check execution mode of the corresponding detection dimension is determined to be the normal execution mode. If the order judgment value corresponding to any detection dimension is less than the preset medium-intensity threshold, the self-check execution mode of the corresponding detection dimension is determined to be the low-frequency execution mode. When the dimension association correction value is greater than or equal to the preset association threshold, the self-check execution mode corresponding to the top two detection dimensions is upgraded by one level. Among them, the low-frequency execution mode is upgraded to the normal execution mode, and the normal execution mode is upgraded to the high-frequency execution mode. Finally, the self-check triggering order and the self-check execution mode corresponding to each detection dimension are sorted together to form the self-check triggering order and self-check execution mode corresponding to each detection dimension.

[0127] S305: Combine the self-test triggering sequence and self-test execution method corresponding to each detection dimension to obtain the target self-test control information.

[0128] Specifically, the self-test triggering sequence and execution method corresponding to the environmental perception layer, the security and fire protection perception layer, and the communication connection layer are read separately. Then, the correspondence between the sequence content and the execution content is established item by item according to the detection dimension. For example, if the self-test triggering sequence corresponding to the environmental perception layer is represented as Oe and the self-test execution method is represented as Me, the self-test triggering sequence corresponding to the security and fire protection perception layer is represented as Os and the self-test execution method is represented as Ms, and the self-test triggering sequence corresponding to the communication connection layer is represented as Oc and the self-test execution method is represented as Mc, then the sequence execution combination corresponding to the environmental perception layer is formed respectively. The sequential execution combination corresponding to the security and fire protection sensing layer and the sequential execution combination corresponding to the communication connection layer. After establishing the sequential execution combinations corresponding to each detection dimension, the sequential execution combinations are sorted and arranged according to the self-test trigger order, so that the sequential execution combinations with the earlier self-test trigger order are arranged first, and the sequential execution combinations with the later self-test trigger order are arranged last. Then, the sorted sequential execution combinations are written into the preset control fields in sequence. When writing, the detection dimension identifier content is written first, then the corresponding self-test trigger order content is written, and finally the corresponding self-test execution method content is written. After all the fields of the sequential execution combinations are written, the combined result of the writing is used as the target self-test control information.

[0129] In one embodiment, in step S40, the corresponding detection device is controlled to perform a target self-test according to the target self-test control information to obtain self-test feedback data, including:

[0130] S401: Analyze the target self-inspection control information to obtain the environmental self-inspection control content corresponding to the environmental perception layer, the security and fire protection self-inspection control content corresponding to the security and fire protection perception layer, and the communication self-inspection control content corresponding to the communication connection layer.

[0131] In this embodiment, the environmental self-inspection control content refers to the self-inspection arrangement content corresponding to the environmental perception layer, which is parsed from the target self-inspection control information; the security and fire protection self-inspection control content refers to the self-inspection arrangement content corresponding to the security and fire protection perception layer, which is parsed from the target self-inspection control information; and the communication self-inspection control content refers to the self-inspection arrangement content corresponding to the communication connection layer, which is parsed from the target self-inspection control information.

[0132] Specifically, the process first reads the content of each field in the target self-inspection control information, arranged in the order of preset control fields. Then, it identifies the detection dimension identifier, self-inspection trigger sequence, and self-inspection execution method according to the order of the fields. Sequence identification refers to determining the category of each field based on its position in the preset control fields. After determining the category of each field, the detection dimension identifier is associated with the self-inspection trigger sequence and self-inspection execution method in the same arrangement unit. When the detection dimension identifier indicates that the corresponding arrangement unit belongs to the environmental perception layer, the self-inspection trigger sequence and self-inspection execution method in that arrangement unit are then associated. The content is organized into environmental self-inspection control content. When the detection dimension identifier indicates that the corresponding arrangement unit belongs to the security and fire protection perception layer, the self-inspection trigger sequence content and self-inspection execution method content in that arrangement unit are organized into security and fire protection self-inspection control content. When the detection dimension identifier indicates that the corresponding arrangement unit belongs to the communication connection layer, the self-inspection trigger sequence content and self-inspection execution method content in that arrangement unit are organized into communication self-inspection control content. After completing the classification and organization of the content corresponding to each arrangement unit, environmental self-inspection control content corresponding to the environmental perception layer, security and fire protection self-inspection control content corresponding to the security and fire protection perception layer, and communication self-inspection control content corresponding to the communication connection layer are formed respectively.

[0133] S402: According to the self-test triggering sequence corresponding to the target self-test control information, control the detection equipment corresponding to the environmental perception layer to perform environmental self-test according to the environmental self-test control content and obtain environmental self-test feedback data; control the detection equipment corresponding to the security and fire protection perception layer to perform security and fire protection self-test according to the security and fire protection self-test control content and obtain security and fire protection self-test feedback data; control the detection equipment corresponding to the communication connection layer to perform communication self-test according to the communication self-test control content and obtain communication self-test feedback data.

[0134] Specifically, the self-test trigger sequence corresponding to each detection dimension is read from the target self-test control information, and an execution sequence is formed from front to back according to the self-test trigger sequence. Then, control commands are issued sequentially according to the execution sequence. When the execution sequence points to the environmental perception layer, the self-test execution method and corresponding arrangement position in the environmental self-test control content are read first. Then, the start time, end time, and number of reads of the corresponding detection device in the environmental perception layer are determined according to the environmental self-test control content. Subsequently, the output data of the corresponding detection device in the environmental perception layer is repeatedly read at uniform intervals between the start time and the end time, and the results of each read are organized in chronological order to form environmental self-test feedback data. When the execution sequence points to the security and fire protection perception layer, the self-test execution method and corresponding arrangement position in the security and fire protection self-test control content are read first. Then, the start time, end time, and number of reads of the corresponding detection device in the security and fire protection perception layer are determined according to the security and fire protection self-test control content. Subsequently, the output data of the corresponding detection device in the security and fire protection perception layer is repeatedly read at uniform intervals between the start time and the end time. The system continuously reads the output data of the corresponding detection devices in the security and fire protection sensing layer at intervals, and organizes the reading results in chronological order to form security and fire protection self-inspection feedback data. When the execution sequence points to the communication connection layer, it first reads the self-inspection execution mode and corresponding arrangement position in the communication self-inspection control content, and then determines the connection start time, connection end time, and number of interactions of the corresponding detection device in the communication connection layer according to the communication self-inspection control content. Subsequently, it cyclically reads the connection status data of the corresponding detection device in the communication connection layer at even intervals between the connection start time and the connection end time, and organizes the reading results in chronological order to form communication self-inspection feedback data. After acquiring the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data, it retains the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data in their respective feedback result positions as environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data for subsequent processing.

[0135] S403: Compare the environmental self-inspection feedback data with the environmental anomaly change sequence, compare the security and fire protection self-inspection feedback data with the security and fire protection anomaly change sequence, and compare the communication self-inspection feedback data with the communication anomaly change sequence to obtain the anomaly continuation result.

[0136] Specifically, first read the time positions corresponding to each feedback result in the environmental self-check feedback data, security and fire protection self-check feedback data, and communication self-check feedback data respectively. At the same time, read the time positions corresponding to each abnormal fluctuation identification value in the environmental abnormal change sequence, security and fire protection abnormal change sequence, and communication abnormal change sequence respectively, and then make one-to-one correspondence according to the same time position. Among them, when the time position corresponding to any feedback result is not completely consistent with the time position in the corresponding abnormal change sequence, select the abnormal fluctuation identification value with the closest distance to the time position corresponding to the feedback result as the corresponding comparison value. After completing the time position correspondence, for any detection dimension, let the corresponding feedback result sequence be denoted as , and the corresponding abnormal change sequence be denoted as . Among them, bi represents the feedback result value corresponding to the i-th time position, ai represents the abnormal fluctuation identification value corresponding to the i-th time position, and n represents the number of comparisons after completing the time position correspondence. Subsequently, perform difference calculation on each time position to obtain the continuation deviation value of the corresponding time position . Then, identify the abnormal change direction based on the magnitude relationship between the feedback result value and the abnormal fluctuation identification value. When bi < ai, determine that the corresponding time position is in the abnormal weakening state. When , determine that the corresponding time position is in the abnormal maintaining state. When bi > ai, determine that the corresponding time position is in the abnormal strengthening state. After completing the identification of the continuation deviation value and abnormal change direction of each time position, perform abnormal continuation determination on each time position. Among them, when and the corresponding time position is in the abnormal maintaining state, record the corresponding time position as the abnormal continuation position. When and the corresponding time position is in the abnormal strengthening state, also record the corresponding time position as the abnormal continuation position. Among them, represents the preset continuation deviation threshold. Subsequently, count the number of abnormal continuation positions under any detection dimension, and sort them in combination with the abnormal change direction corresponding to each abnormal continuation position to form the abnormal continuation result corresponding to this detection dimension. Finally, retain the abnormal continuation results corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer respectively as the abnormal continuation results

[0137] S404: In the case where any abnormal continuation result indicates that there is an abnormal continuation in the corresponding detection dimension, adjust the self-check control content corresponding to the remaining detection dimensions to obtain the adjusted self-check control content, and control the corresponding detection device to perform the target self-check according to the adjusted self-check control content

[0138] Specifically, the number of abnormal continuation positions and the direction of abnormal change in the corresponding detection dimension are read from the abnormal continuation results. The number of abnormal continuation positions is denoted as N, and the cumulative value of the direction corresponding to the abnormal change direction is denoted as V. When the abnormal change direction is in an abnormal maintenance state, the corresponding cumulative value of the direction is 1, and when the abnormal change direction is in an abnormal enhancement state, the corresponding cumulative value of the direction is 2. Then, the number of abnormal continuation positions and the cumulative value of the direction are combined and calculated to obtain the abnormal continuation adjustment amount. Then, the self-check trigger order and self-check execution method contents contained in the self-check control contents corresponding to the other detection dimensions are read, and the self-check control contents corresponding to the other detection dimensions are adjusted according to the abnormal continuation adjustment amount. Specifically, when adjusting the self-check trigger order, the order position of the other detection dimensions in the original arrangement is read first. If the order position of the corresponding detection dimension in the original arrangement is recorded as 0, then... The update process yields the adjusted self-check trigger sequence, where O' represents the adjusted sequence position and L_s represents the sequence adjustment amount calculated from the exception continuation adjustment amount. When adjusting the self-check execution method, first read the execution level corresponding to the original self-check execution methods of the other detection dimensions. If the execution levels are divided into low-frequency execution mode, normal execution mode, and high-frequency execution mode, and denoted as 1, 2, and 3 respectively, then proceed according to... The update process is performed to obtain the adjusted execution level, where M represents the original execution level, M' represents the adjusted execution level, and Lm represents the mode adjustment amount converted from the exception continuation adjustment amount. After updating the sequence position and execution level corresponding to the other detection dimensions, the updated self-test trigger sequence content and the updated self-test execution mode content are written into the control fields of the corresponding detection dimensions to form the adjusted self-test control content. Then, control commands are issued sequentially according to the adjusted self-test trigger sequence content, and the number of repeated reads, continuous read intervals, or cyclic reads of the corresponding detection device are determined according to the adjusted self-test execution mode content. Finally, the corresponding detection device is controlled to perform the target self-test according to the updated read arrangement.

[0139] S405: Integrate environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data to obtain self-inspection feedback data.

[0140] Specifically, firstly, the time position and feedback result value corresponding to each feedback result in the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are read respectively. Then, the time interval commonly covered by the three types of feedback results is selected as the integration interval, and multiple integration time positions are established within the integration interval according to a unified time step. Subsequently, the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are mapped to each integration time position respectively. When there is no original feedback result value for the corresponding detection dimension at any integration time position, the feedback result value closest to that integration time position is selected as the mapping result. After completing the mapping, the environmental feedback result value, security and fire protection feedback result value, and communication feedback result value at each integration time position are organized according to their positions to form the feedback result group corresponding to each integration time position. Let the environmental feedback result value, security and fire protection feedback result value, and communication feedback result value corresponding to the i-th integration time position be denoted as ei, si, and ci respectively. Then, an integration calculation is performed on any integration time position, and the integration calculation is expressed as follows: Where ri represents the integration feedback result value corresponding to the i-th integration time position. , and These represent the integrated weights corresponding to the environmental feedback result values, security and fire protection feedback result values, and communication feedback result values, respectively. After calculating the integrated feedback result values ​​corresponding to each integrated time position, the integrated feedback result values ​​corresponding to each integrated time position are arranged in chronological order to form self-inspection feedback data.

[0141] In one embodiment, in step S50, the self-test feedback data and the corresponding abnormal change sequence are jointly compared to obtain target fault characterization information, including:

[0142] S501: Classify and analyze the self-inspection feedback data to obtain environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data.

[0143] Specifically, first, the feedback result groups corresponding to each integrated time position in the self-inspection feedback data are read, and the detection dimension identifier content, time position content, and feedback result value contained in each feedback result group are extracted item by item. Then, each feedback result value is classified according to the detection dimension identifier content. When the detection dimension identifier content indicates that the corresponding feedback result value belongs to the environmental perception layer, the corresponding feedback result value and corresponding time position content are written into the environmental feedback result set. When the detection dimension identifier content indicates that the corresponding feedback result value belongs to the security and fire protection perception layer, the corresponding feedback result value and corresponding time position content are written into the security and fire protection feedback result set. When the corresponding feedback result value belongs to the communication connection layer, the corresponding feedback result value and corresponding time position content are written into the communication feedback result set. After the classification is completed, the feedback result values ​​in the environmental feedback result set, the security and fire protection feedback result set, and the communication feedback result set are sorted and organized according to time position content, so that the feedback result values ​​in the same set are arranged in chronological order. Then, the sorted environmental feedback result set is determined as environmental self-inspection feedback data, the sorted security and fire protection feedback result set is determined as security and fire protection self-inspection feedback data, and the sorted communication feedback result set is determined as communication self-inspection feedback data.

[0144] S502: Compare the environmental self-inspection feedback data with the environmental anomaly change sequence before and after to determine the environmental anomaly reduction result; compare the security and fire protection self-inspection feedback data with the security and fire protection anomaly change sequence before and after to determine the security and fire protection anomaly reduction result; compare the communication self-inspection feedback data with the communication anomaly change sequence before and after to determine the communication anomaly reduction result.

[0145] In this embodiment, the environmental anomaly reduction result refers to the environmental anomaly change status data formed based on the difference between the environmental self-inspection feedback data and the environmental anomaly change sequence. The security and fire protection anomaly reduction result refers to the security and fire protection anomaly change status data formed based on the difference between the security and fire protection self-inspection feedback data and the security and fire protection anomaly change sequence. The communication anomaly reduction result refers to the communication anomaly change status data formed based on the difference between the communication self-inspection feedback data and the communication anomaly change sequence.

[0146] Specifically, firstly, the time positions corresponding to each feedback result value in the environmental self-inspection feedback data, security and fire protection self-inspection feedback data, and communication self-inspection feedback data are read respectively. Then, the time positions corresponding to each abnormal fluctuation indicator value in the environmental abnormal change sequence, security and fire protection abnormal change sequence, and communication abnormal change sequence are read respectively. Each feedback result value and each abnormal fluctuation indicator value are matched one-to-one according to the same time position. Where the time position corresponding to any feedback result value does not have a completely identical time position in the corresponding abnormal change sequence, the abnormal fluctuation indicator value closest to that time position is selected as the corresponding comparison value. After completing the time position matching, for any detection dimension, let the corresponding abnormal change sequence be represented as... The corresponding self-test feedback data is represented as follows: Where ai represents the abnormal fluctuation flag value corresponding to the i-th time position, bi represents the feedback result value corresponding to the i-th time position, and n represents the number of data points after completing the corresponding time position. Subsequently, the difference before and after each time position is calculated. And based on the difference before and after, it identifies abnormal changes at each time point. When, the corresponding time position is determined as an abnormal weakening state, when When, the corresponding time position is determined as an abnormal maintenance state, when At that time, the corresponding time position is determined as an abnormal enhancement state, where, This indicates a preset threshold for anomaly detection. After identifying abnormal changes at each time point, the differences and abnormal changes at all time points under the same detection dimension are sequentially organized to form anomaly reduction results for the corresponding detection dimension. Specifically, the organized result formed by environmental self-inspection feedback data and environmental abnormal change sequence is determined as the environmental anomaly reduction result; the organized result formed by security and fire protection self-inspection feedback data and security and fire protection abnormal change sequence is determined as the security and fire protection anomaly reduction result; and the organized result formed by communication self-inspection feedback data and communication abnormal change sequence is determined as the communication anomaly reduction result.

[0147] S503: Based on the results of environmental anomaly reduction, security and fire protection anomaly reduction, and communication anomaly reduction, extract residual anomaly values ​​and anomaly reduction directions respectively.

[0148] In this embodiment, residual outliers refer to values ​​that still represent the degree of residual abnormality after the comparison before and after the corresponding detection dimension is completed, and the direction of abnormality reduction refers to the trend of abnormal change represented by the corresponding detection dimension after the comparison before and after.

[0149] Specifically, first, the differences and abnormal change states at each time position in the environmental anomaly reduction results, security and fire protection anomaly reduction results, and communication anomaly reduction results are read respectively. Then, for any detection dimension, the differences at each time position are traversed in chronological order. If the difference at the i-th time position is denoted as di, the absolute value of the difference is processed to obtain the residual anomaly value at the i-th time position. Where ri represents the residual outlier corresponding to the i-th time position. After extracting the residual outlier corresponding to each time position, the direction of anomaly reduction is determined according to the anomaly change state of the corresponding time position. When the anomaly change state is the anomaly weakening state, the anomaly reduction direction of the corresponding time position is determined as the reduction direction. When the anomaly change state is the anomaly maintaining state, the anomaly reduction direction of the corresponding time position is determined as the maintenance direction. When the anomaly change state is the anomaly strengthening state, the anomaly reduction direction of the corresponding time position is determined as the strengthening direction. Then, the same processing is performed on the environmental anomaly reduction results, the security and fire protection anomaly reduction results, and the communication anomaly reduction results. The residual outliers and anomaly reduction directions of each detection dimension corresponding to each time position are sorted in chronological order to form the residual outliers and anomaly reduction directions corresponding to each detection dimension.

[0150] S504: Cross-combination analysis of each residual outlier and each outlier reduction direction is performed to obtain the correlation comparison results between each detection dimension.

[0151] In this embodiment, the correlation comparison result refers to the correlation analysis data formed based on the correspondence between residual outliers and outlier reduction directions at the same time position for different detection dimensions.

[0152] Specifically, firstly, the residual anomalies and anomaly reduction directions of the environmental perception layer, security and fire protection perception layer, and communication connection layer at each time position are read separately. Then, the data content corresponding to each detection dimension is aligned and organized according to the same time position. Here, if the environmental residual anomaly, security and fire protection residual anomaly, and communication residual anomaly corresponding to the i-th time position are respectively denoted as... , and The corresponding anomaly reduction directions are denoted as follows: , and First, the anomaly reduction directions are numerically processed. A value of 1 is assigned when the anomaly reduction direction is a reduction direction, a value of 2 when the anomaly reduction direction is a maintenance direction, and a value of 3 when the anomaly reduction direction is a reinforcement direction. After numerical processing, environmental direction values, security and fire protection direction values, and communication direction values ​​are generated respectively. Then, a cross-combination calculation is performed on the residual anomaly values ​​at the same time location to obtain the residual cross-values. Here, gi represents the residual cross value corresponding to the i-th time position. Then, cross-consistency calculation is performed on the direction values ​​at the same time position to obtain the direction combination value. Where hi represents the direction combination value corresponding to the i-th time position, after obtaining the residual cross value and the direction combination value, the two are jointly calculated to obtain the association comparison value corresponding to the i-th time position. , where ki represents the correlation comparison value corresponding to the i-th time position. Then, the correlation comparison values ​​corresponding to each time position are arranged in chronological order, and combined with the residual outliers and outlier reduction directions corresponding to each time position, the correlation comparison results between each detection dimension are formed.

[0153] S505: Perform joint mapping on the correlation comparison results to obtain the target fault characterization information.

[0154] Specifically, first, the correlation comparison value, residual outlier, and anomaly reduction direction corresponding to each time position in each correlation comparison result are read separately, and a correspondence is established according to a unified time position. Then, for any given time position, the correlation comparison value corresponding to that time position is denoted as ki, and the environmental residual outlier, security and fire protection residual outlier, and communication residual outlier corresponding to that time position are denoted as ki respectively. , and The directions for mitigating environmental anomalies, security and fire safety anomalies, and communication anomalies corresponding to this time and location are respectively denoted as... , and Then, a convergence calculation is performed on the residual outliers corresponding to the three detection dimensions to obtain the residual convergence value. Then, perform directional aggregation calculation on the anomaly reduction directions corresponding to the three detection dimensions to obtain the directional aggregation value. The directional aggregation value is used to characterize the concentration of abnormal change trends in each detection dimension at the same time location. After calculating the residual aggregation value and the directional aggregation value, the correlation comparison value, residual aggregation value, and directional aggregation value are jointly mapped. The joint mapping is expressed as follows: Where gi represents the fault characterization value corresponding to the i-th time position. After obtaining the fault characterization values ​​corresponding to each time position, the fault characterization values, correlation comparison values, residual aggregation values ​​and directional aggregation values ​​corresponding to each time position are then sorted in chronological order to form the target fault characterization information.

[0155] In one embodiment, step S60, namely, determining fault prediction result information based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, includes:

[0156] S601: Expand the target fault characterization information to obtain the fault state characterization sequence.

[0157] In this embodiment, the fault state representation sequence refers to the continuous state data sequence formed by unfolding the fault representation values ​​and related representation contents corresponding to each time position in the target fault representation information in chronological order.

[0158] Specifically, first, the fault characterization value, correlation comparison value, residual convergence value, and directional convergence value corresponding to each time position in the target fault characterization information are read, and a one-to-one correspondence is established according to the chronological order of each time position. Then, for any time position, the corresponding fault characterization value is denoted as gi, the corresponding correlation comparison value as ki, the corresponding residual convergence value as mi, and the corresponding directional convergence value as ni, where i represents the time position number. After establishing the correspondence, the fault characterization value, correlation comparison value, residual convergence value, and directional convergence value corresponding to each time position are expanded and combined. The expansion and combination process refers to compressing multiple characterization contents corresponding to the same time position into a single state characterization value. The compression calculation is represented as follows: }, where zi represents the state representation value corresponding to the i-th time position. , and Let these represent the correlation adjustment coefficient, residual adjustment coefficient, and directional adjustment coefficient, respectively. After obtaining the state characterization values ​​corresponding to each time position, these values ​​are arranged sequentially from front to back according to their time positions, forming... Where Z represents the fault state representation sequence, and n represents the number of time positions corresponding to the target fault representation information. Then, the state representation values ​​corresponding to each time position are matched with the corresponding time position content to form the fault state representation sequence.

[0159] S602: The fault state representation sequence is compared with the evolution trajectories of similar faults in historical fault data to obtain a cluster of candidate evolution trajectories.

[0160] In this embodiment, the candidate evolution trajectory cluster refers to the set of trajectories selected from the evolution trajectories of the same type of fault in historical fault data that have a high degree of similarity to the fault state representation sequence.

[0161] Specifically, first, read the state representation values ​​corresponding to each time position in the fault state representation sequence, and then record the fault state representation sequence as follows: Where zi represents the state representation value corresponding to the i-th time position, and n represents the number of time positions corresponding to the fault state representation sequence. Then, the evolution trajectories of similar faults are read from historical fault data. Let the j-th similar fault evolution trajectory be represented as... ,in, This represents the historical state value corresponding to the j-th similar fault evolution trajectory at time t. Let represent the number of time positions corresponding to the j-th type of fault evolution trajectory. Then, for any given type of fault evolution trajectory, a sliding cut is performed with the same length as the fault state representation sequence, sequentially extracting multiple continuous trajectory segments from the type of fault evolution trajectory. If the continuous trajectory segment corresponding to the u-th starting position in the j-th type of fault evolution trajectory is denoted as , then ... Then, the degree of difference between the fault state representation sequence and the continuous trajectory segment is calculated, and the degree of difference is expressed as... ,in, This represents the sequence difference value between the fault state characterization sequence and the u-th continuous trajectory segment in the j-th type of fault evolution trajectory. This represents a time-position progression adjustment term, used to reduce the equal impact of earlier and later time positions on the cumulative difference. After calculating the sequence difference values ​​corresponding to each continuous trajectory segment, the minimum value is selected from the sequence difference values ​​corresponding to the same type of fault evolution trajectory, and this minimum value is used as the trajectory similarity characterization value for that type of fault evolution trajectory. If the trajectory similarity characterization value corresponding to the j-th type of fault evolution trajectory is denoted as... Then the trajectory similarity representation value will be compared with the preset similarity screening threshold. When comparing, When, the j-th similar fault evolution trajectory is determined as a candidate trajectory, when When the j-th type of fault evolution trajectory is discarded, after comparing and filtering all types of fault evolution trajectories, all candidate trajectories are arranged in ascending order of their corresponding trajectory similarity values, and the set of arranged candidate trajectories is taken as the candidate evolution trajectory cluster.

[0162] S603: Perform stage matching on each similar fault evolution trajectory in the candidate evolution trajectory cluster to determine the target fault evolution stage and stage evolution rate.

[0163] In this embodiment, the target fault evolution stage refers to the position of the fault development stage corresponding to each similar fault evolution trajectory in the candidate evolution trajectory cluster of the fault state characterization sequence, and the stage evolution rate refers to the speed at which the fault state changes over time within the target fault evolution stage.

[0164] Specifically, first, read the evolution trajectories of each type of fault in the candidate evolution trajectory cluster and the fault state representation sequence, and denote the fault state representation sequence as... The evolution trajectory of any similar fault is denoted as Subsequently, the evolution trajectory of any similar fault is divided into stages according to the abrupt changes in the magnitude of adjacent state changes. Here, the state difference between the t-th time position and the (t-1)-th time position is represented as... Then in If the time interval exceeds a preset stage transition threshold, the t-th time position is determined as the stage boundary position. Based on each stage boundary position, the corresponding fault evolution trajectory is divided into multiple stage intervals. Then, continuous stage segments with the same length as the fault state representation sequence are extracted from each stage interval, and the stage matching value between the fault state representation sequence and each continuous stage segment is calculated. The stage matching value is expressed as... ,in, This represents the state value at the i-th time position of the v-th consecutive stage segment in the j-th similar fault evolution trajectory. This represents the stage matching value between the fault state characterization sequence and the v-th consecutive stage segment in the j-th type of fault evolution trajectory. After calculating the stage matching values ​​for each stage, the stage interval containing the consecutive stage segment with the smallest stage matching value is selected as the matching stage corresponding to the same type of fault evolution trajectory, and the stage number corresponding to this matching stage is denoted as Gj. Then, the matching stages corresponding to each type of fault evolution trajectory are filtered according to the stage matching value from smallest to largest, and the stage number corresponding to the matching stage with the smallest stage matching value is determined as the target fault evolution stage. Subsequently, the start time position and end time position of the stage interval corresponding to the target fault evolution stage are read, and the state values ​​corresponding to the first and last time positions in the corresponding stage interval are denoted as hs and he, respectively, and the start time position and end time position of the corresponding stage interval are denoted as ts and te, respectively. Then, based on... Calculate the stage evolution rate, where V represents the stage evolution rate.

[0165] S604: Based on the target failure evolution stage and the stage evolution rate, the remaining evolution time to reach the preset failure state is estimated.

[0166] In this embodiment, the preset failure state refers to the position of the failure state pre-marked in the historical failure data, and the remaining evolution time refers to the remaining time required for the target failure evolution stage to develop from the current evolution position to the preset failure state.

[0167] Specifically, first, the position of the target fault evolution stage within the corresponding fault evolution trajectory is read, and the stage evolution rate and the failure state value corresponding to the preset failure state are read. The end time position within the target fault evolution stage that best matches the fault state representation sequence is determined as the current evolution position. If the state value corresponding to the current evolution position is denoted as h_c, the failure state value corresponding to the preset failure state is denoted as h_f, and the stage evolution rate is denoted as V, then the state difference ∂Deltah = |h_f - h_c| between the current evolution position and the preset failure state is calculated first. Then, the remaining evolution time is calculated based on ∂Deltah = ∂V + ∂V / ∂f, where ∂Deltah / ∂f represents the remaining evolution time, and ∂V / ∂f represents the remaining evolution time. arepsilon represents the rate correction term, used to avoid division by zero when the stage evolution rate is too small. After the initial calculation of the remaining evolution time is completed, consistency correction is performed based on the stage distribution of the target fault evolution stage in the candidate evolution trajectory cluster. If the number of stages in the candidate evolution trajectory cluster that are the same as the target fault evolution stage is denoted as N_g, and the total number of trajectories in the candidate evolution trajectory cluster is denoted as N_c, then the stage consistency coefficient K=\frac{N_g}{N_c} is calculated. Then, the remaining evolution time is corrected according to \Deltat'=\frac{\Deltat}{1+K}, where \Deltat' represents the corrected remaining evolution time. Subsequently, the corrected remaining evolution time is determined as the remaining evolution time to reach the preset failure state.

[0168] S605: Based on the candidate evolution trajectory clusters, the target fault evolution stage, and the remaining evolution time, determine the probability of fault occurrence and the expected fault time to obtain fault prediction results.

[0169] In this embodiment, the probability of failure refers to the likelihood that the target object will develop into a preset failure state during subsequent operation, and the estimated failure time refers to the time point at which the failure occurs, calculated based on the remaining evolution time.

[0170] Specifically, first, the stage matching value, the stage number corresponding to the target fault evolution stage, and the remaining evolution time of each similar fault evolution trajectory in the candidate evolution trajectory cluster are read. Then, the stage matching value of each similar fault evolution trajectory in the candidate evolution trajectory cluster is statistically analyzed. If the stage matching value corresponding to the j-th similar fault evolution trajectory is denoted as Mj, and the total number of trajectories in the candidate evolution trajectory cluster is denoted as Nc, then the average stage matching value is calculated first. _j, then based on Calculate the probability of failure occurring, where P represents the probability of failure occurring. Represents the matching sensitivity coefficient. This represents the baseline value for probability conversion. After calculating the probability of failure, the current time is read and recorded as [value]. The remaining evolution time is denoted as Subsequently based on Calculate the expected failure time, where ty represents the expected failure time. After determining the failure probability and the expected failure time, organize the failure probability, the expected failure time, the stage number corresponding to the target failure evolution stage, and the remaining evolution time according to the preset field order to form the failure prediction result information.

[0171] In one embodiment, step S603, namely, performing stage matching on each similar fault evolution trajectory in the candidate evolution trajectory cluster to determine the target fault evolution stage and stage evolution rate, includes:

[0172] S6031: Perform transition identification on the fault state representation sequence to obtain stage transition features.

[0173] In this embodiment, the stage transition feature refers to the positional and amplitude features of a significant shift in the trend of state change within the fault state characterization sequence.

[0174] Specifically, first read the state representation values ​​corresponding to each time position in the fault state representation sequence. Let the fault state representation sequence be represented as... Where zi represents the state representation value corresponding to the i-th time position, n represents the number of time positions, and the state change between adjacent time positions is calculated in chronological order. The state change is expressed as: ,in, Then, the difference in change between two adjacent state changes is calculated, and the difference in change is expressed as... ,in, After calculating the difference in change at each time point, the difference in change is compared with a preset threshold for determining a turning point. When the i-th time position is determined as a candidate turning point position, then, This indicates a preset threshold for determining a turning point. Subsequently, the absolute values ​​and differences in state changes corresponding to each candidate turning point position are jointly processed to form... The corresponding content is then arranged in chronological order to obtain the characteristics of the stage transition.

[0175] S6032: Segment the evolution trajectories of the same type of fault in the candidate evolution trajectory cluster to obtain trajectory segmentation features.

[0176] In this embodiment, the trajectory segmentation feature refers to the starting position, ending position, and state change characteristics of each stage formed after the segmentation of the evolution trajectory of each type of fault.

[0177] Specifically, first, read each fault evolution trajectory of the same type in the candidate evolution trajectory cluster one by one. Let any fault evolution trajectory of the same type be represented as... ,in, This represents the historical state value corresponding to the j-th similar fault evolution trajectory at time t. This represents the number of time positions corresponding to the j-th type of fault evolution trajectory. Then, the trajectory change between adjacent time positions is calculated in chronological order, and the trajectory change is expressed as... ,in, Then calculate the difference in trajectory change between two adjacent trajectory changes, expressed as: ,in, After calculating the trajectory change difference corresponding to each time position, the trajectory change difference is compared with the preset segmentation judgment threshold. When, the t-th time position is determined as the segment position, where, The preset segmentation threshold is used to divide the j-th type of fault evolution trajectory into multiple continuous stage intervals based on the position of each segment. The start time position, end time position, start state value, and end state value are extracted for each continuous stage interval. Then, the start time position, end time position, start state value, and end state value corresponding to each continuous stage interval are arranged in the order of the stages to obtain the trajectory segmentation features corresponding to the j-th type of fault evolution trajectory.

[0178] S6033: Match the stage transition features with the segmentation features of each trajectory, and extract the corresponding stage duration interval based on the matching results to obtain the stage matching results corresponding to each type of fault evolution trajectory.

[0179] Specifically, first, the time position, state change amount, and change difference corresponding to each candidate position in the stage transition features are read. Then, the start time position, end time position, start state value, and end state value corresponding to each stage interval in the trajectory segmentation features corresponding to any similar fault evolution trajectory are read. Subsequently, for any stage interval, the stage change span and stage change slope corresponding to the stage interval are calculated. Wherein, if the start state value corresponding to the stage interval is denoted as h_s, the end state value as he, the start time position as ts, and the end time position as te, then the stage change span is expressed as... The slope of the stage change is expressed as After calculating the stage change span and stage change slope corresponding to each stage interval, the difference between the change at each candidate position in the stage transition feature and the stage change span corresponding to each stage interval are compared. Similarly, the difference between the state change at each candidate position in the stage transition feature and the stage change slope corresponding to each stage interval is compared to form a stage matching value, which is expressed as follows: ,in, This represents the stage matching value corresponding to the v-th stage interval in the j-th similar fault evolution trajectory. and The matching adjustment coefficient is represented by the stage matching value. After calculating the stage matching value corresponding to each stage interval, the stage interval with the smallest stage matching value in the same type of fault evolution trajectory is selected as the matching stage interval. The start time position and end time position corresponding to the matching stage interval are extracted as the stage duration interval. Then, the stage matching value and the stage duration interval are sorted together to obtain the stage matching result of the corresponding type of fault evolution trajectory.

[0180] S6034: Filter the stage matching results to determine the target fault evolution stage.

[0181] Specifically, firstly, the stage matching values ​​and stage duration intervals contained in the stage matching results corresponding to the stage evolution trajectories of the same type are read, and sorted according to the stage matching values ​​from smallest to largest. Then, a preset number of stage matching results with the highest ranking are selected as candidate stage results. Next, the stage interval number corresponding to each candidate stage result is read, and the number of times the same stage interval number appears is counted. If any stage interval number is denoted as g, and the corresponding number of occurrences is denoted as Ng, then the number of occurrences corresponding to each stage interval number is compared, and the stage interval number with the largest number of occurrences is determined as the target stage number. When there are multiple stage interval numbers with the same number of occurrences, the average value of the stage matching values ​​corresponding to multiple stage interval numbers is read, and the stage interval number with the smallest average value of stage matching values ​​is determined as the target stage number. Then, the stage interval corresponding to the target stage number is determined as the target fault evolution stage.

[0182] S6035: Extract the state change span and time change span corresponding to the target fault evolution stage, and determine the stage evolution rate based on the ratio of the state change span to the time change span.

[0183] Specifically, first, the initial state value, final state value, initial time position, and final time position corresponding to the target fault evolution stage are read, and the initial state value is denoted as hs, the final state value as he, the initial time position as ts, and the final time position as te. Then, the state change span is calculated based on the initial and final state values, and the state change span is expressed as... Then, based on the start and end time positions, the time span is calculated, and the time span is expressed as... After calculating the state change span and time change span, the stage evolution rate is determined based on the ratio of the state change span to the time change span. The stage evolution rate is expressed as... Where V represents the stage evolution rate, Indicates the time correction item.

[0184] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0185] In one embodiment, a multi-dimensional AI self-checking and fault prediction device is provided, which corresponds one-to-one with the multi-dimensional AI self-checking and fault prediction method in the above embodiments. For example... Figure 2 As shown, the multi-dimensional AI self-inspection and fault prediction device includes a multi-source detection data acquisition module, an abnormal change sequence generation module, a target self-inspection control module, a target self-inspection execution module, a target fault characterization module, a fault prediction module, and an alarm output module.

[0186] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0187] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A multi-dimensional AI self-inspection and fault prediction method, characterized in that, The multi-dimensional AI self-inspection and fault prediction method includes: Acquire multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer; The multi-source detection data is input into a preset lightweight analysis model for analysis and processing to obtain environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences. Based on the abnormal amplitude, duration, and frequency of change in the environmental abnormal change sequence, the security and fire protection abnormal change sequence, and the communication abnormal change sequence, the target self-inspection control information is determined. According to the target self-inspection control information, the corresponding detection equipment is controlled to perform the target self-inspection to obtain self-inspection feedback data; The self-test feedback data and the corresponding abnormal change sequence are jointly compared to obtain the target fault characterization information. Based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data, the fault prediction result information is determined. When the fault prediction result information meets the preset conditions, an alarm message is output.

2. The multi-dimensional AI self-inspection and fault prediction method according to claim 1, characterized in that, The process involves inputting the multi-source detection data into a preset lightweight analysis model for analysis and processing, resulting in environmental anomaly change sequences, security and fire protection anomaly change sequences, and communication anomaly change sequences, including: The multi-source detection data are classified and organized according to the collection timestamp and detection dimension to obtain environmental perception layer detection data, security and fire protection perception layer detection data, and communication connection layer detection data. The environmental perception layer detection data, the security and fire protection perception layer detection data, and the communication connection layer detection data are segmented according to a preset analysis time window to obtain multiple continuous detection data segments. Each of the continuous detection data segments is input into the lightweight analysis model to obtain the corresponding offset feature value and difference feature value; The offset feature value and the difference feature value are combined to represent the abnormal fluctuation identification value; The abnormal fluctuation identifier values ​​of the environmental perception layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain an environmental abnormal change sequence. The abnormal fluctuation identifier values ​​of the security and fire protection perception layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain a security and fire protection abnormal change sequence. The abnormal fluctuation identifier values ​​of the communication connection layer detection data corresponding to each continuous detection data segment are arranged in chronological order to obtain a communication abnormal change sequence.

3. The multi-dimensional AI self-inspection and fault prediction method according to claim 1, characterized in that, The determination of target self-inspection control information based on the abnormal amplitude, duration, and frequency of changes in the environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence includes: Within a preset judgment period, the abnormal amplitude, duration, and frequency of change of the environmental abnormal change sequence, the security and fire protection abnormal change sequence, and the communication abnormal change sequence are extracted respectively. By combining and characterizing the abnormal amplitude, the duration, and the frequency of change, we can obtain the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value, and the communication self-inspection intensity value. The environmental anomaly change sequence, the security and fire protection anomaly change sequence, and the communication anomaly change sequence are time-series aligned, and synchronization fluctuation analysis is performed on each aligned anomaly change sequence to obtain dimension correlation correction values. Based on the environmental self-inspection intensity value, the security and fire protection self-inspection intensity value, the communication self-inspection intensity value, and the dimension correlation correction value, the self-inspection triggering order and self-inspection execution method corresponding to each detection dimension are determined; The self-test triggering sequence and self-test execution method corresponding to each detection dimension are combined to obtain the target self-test control information.

4. The multi-dimensional AI self-inspection and fault prediction method according to claim 1, characterized in that, The step of controlling the corresponding detection equipment to perform target self-testing according to the target self-testing control information, and obtaining self-testing feedback data, includes: The target self-inspection control information is parsed to obtain the environmental self-inspection control content corresponding to the environmental perception layer, the security and fire protection self-inspection control content corresponding to the security and fire protection perception layer, and the communication self-inspection control content corresponding to the communication connection layer. According to the self-test triggering sequence corresponding to the target self-test control information, the detection device corresponding to the control environment perception layer performs environmental self-test according to the environmental self-test control content and obtains environmental self-test feedback data; the detection device corresponding to the control security and fire protection perception layer performs security and fire protection self-test according to the security and fire protection self-test control content and obtains security and fire protection self-test feedback data; the detection device corresponding to the control communication connection layer performs communication self-test according to the communication self-test control content and obtains communication self-test feedback data. The environmental self-inspection feedback data is compared with the environmental abnormal change sequence, the security and fire protection self-inspection feedback data is compared with the security and fire protection abnormal change sequence, and the communication self-inspection feedback data is compared with the communication abnormal change sequence to obtain the abnormal continuation result; If any of the above-mentioned abnormal continuation results indicate that there is abnormal continuation in the corresponding detection dimension, adjust the self-inspection control content corresponding to the other detection dimensions to obtain the adjusted self-inspection control content, and control the corresponding detection device to perform target self-inspection according to the adjusted self-inspection control content; The self-inspection feedback data of the environment, the self-inspection feedback data of the security and fire protection, and the self-inspection feedback data of the communication are integrated to obtain the self-inspection feedback data.

5. The multi-dimensional AI self-inspection and fault prediction method according to claim 1, characterized in that, The step of jointly comparing the self-test feedback data and the corresponding abnormal change sequence to obtain target fault characterization information includes: The self-test feedback data is classified and analyzed to obtain the environmental self-test feedback data, the security and fire protection self-test feedback data, and the communication self-test feedback data. The environmental self-inspection feedback data is compared with the environmental anomaly change sequence to determine the environmental anomaly reduction result; the security and fire protection self-inspection feedback data is compared with the security and fire protection anomaly change sequence to determine the security and fire protection anomaly reduction result; the communication self-inspection feedback data is compared with the communication anomaly change sequence to determine the communication anomaly reduction result. Based on the environmental anomaly reduction results, the security and fire protection anomaly reduction results, and the communication anomaly reduction results, residual anomaly values ​​and anomaly reduction directions are extracted respectively. By performing cross-combination analysis on each of the residual outliers and each of the outlier reduction directions, the correlation comparison results between each detection dimension are obtained. The target fault characterization information is obtained by jointly mapping the correlation comparison results.

6. The multi-dimensional AI self-inspection and fault prediction method according to claim 1, characterized in that, The process of determining fault prediction results based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data includes: The target fault characterization information is expanded to obtain a fault state characterization sequence; The fault state representation sequence is compared with the evolution trajectories of the same type of fault in the historical fault data to obtain a cluster of candidate evolution trajectories; Stage matching is performed on each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster to determine the target fault evolution stage and the stage evolution rate; Based on the target failure evolution stage and the stage evolution rate, the remaining evolution time to reach the preset failure state is estimated. Based on the candidate evolution trajectory cluster, the target fault evolution stage, and the remaining evolution duration, the probability of fault occurrence and the expected fault time are determined, and the fault prediction result information is obtained.

7. The multi-dimensional AI self-inspection and fault prediction method according to claim 6, characterized in that, The step of performing stage matching on each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster to determine the target fault evolution stage and stage evolution rate includes: The fault state representation sequence is subjected to transition identification to obtain stage transition features; Each of the same type of fault evolution trajectories in the candidate evolution trajectory cluster is segmented to obtain trajectory segmentation features; The stage transition features are matched with the trajectory segmentation features, and the corresponding stage duration intervals are extracted based on the matching results to obtain the stage matching results corresponding to each of the same type of fault evolution trajectories. The stage matching results are filtered to determine the target fault evolution stage; Extract the state change span and time change span corresponding to the target fault evolution stage, and determine the stage evolution rate based on the ratio of the state change span to the time change span.

8. A multi-dimensional AI self-inspection and fault prediction device, characterized in that, The multi-dimensional AI self-inspection and fault prediction device includes: The multi-source detection data acquisition module is used to acquire multi-source detection data corresponding to the environmental perception layer, security and fire protection perception layer, and communication connection layer. An abnormal change sequence generation module is used to input the multi-source detection data into a preset lightweight analysis model for analysis and processing, and to obtain environmental abnormal change sequences, security and fire protection abnormal change sequences, and communication abnormal change sequences. The target self-test control module is used to determine target self-test control information based on the abnormal amplitude, duration and frequency of change in the environmental abnormal change sequence, the security and fire protection abnormal change sequence and the communication abnormal change sequence. The target self-test execution module is used to control the corresponding detection equipment to perform target self-test according to the target self-test control information, and obtain self-test feedback data. The target fault characterization module is used to jointly compare the self-test feedback data and the corresponding abnormal change sequence to obtain target fault characterization information. The fault prediction module is used to determine the fault prediction result information based on the target fault characterization information and the evolution trajectory of similar faults in historical fault data. The alarm output module is used to output alarm information when the fault prediction result information meets preset conditions.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the multi-dimensional AI self-testing and fault prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the multi-dimensional AI self-inspection and fault prediction method as described in any one of claims 1 to 7.

Images