199 results about "Data differencing" patented technology

The invention relates to the technical field of distributed data processing systems, and discloses a data processing system supporting full localization deployment and a compliance guarantee method thereof.The data processing system supporting full localization deployment comprises an audit event collection module used for collecting and storing processing event hash; the event organization module is used for realizing logic time sequence management of events through topological sorting and causal relationship marking; the hash chain construction module is used for constructing a local-to-global double-layer hash chain by utilizing the constructed directed acyclic graph and event hash; the increment synchronization module is used for executing increment synchronization based on Hash tree difference identification; the conflict solving module is used for detecting and solving event conflicts in the distributed environment by applying a version vector technology aiming at data differences found in the incremental synchronization process; according to the method, the lightweight linear hash chain is used for replacing a traditional binary tree structure, high-efficiency and low-cost audit chain maintenance is realized, and the data integrity is guaranteed in an unstable network environment.

The invention relates to the technical field of data monitoring, in particular to an intelligent storage supervision system for biological samples. Obtaining a data anomaly degree according to data discrete features at the current moment in the monitoring data sequence; obtaining the data credibility according to the data difference characteristics of the current moment and the adjacent moment, the recent data distribution characteristics and the data difference characteristics of the adjacent sensors; according to the recent data change characteristics, the difference change characteristics of the recent data and the preset door opening time period data and the time interval characteristics of the current moment and the door opening moment, the external isolation degree is obtained; according to the data credibility and the external isolation degree, obtaining an abnormal feature confidence degree; and obtaining an enhancement coefficient according to the change trend characteristics of the monitoring data sequence. According to the invention, the abnormal score of the current moment is obtained according to the data anomaly degree, the abnormal feature confidence and the enhancement coefficient, and environmental anomaly early warning is carried out, so that the accuracy of environmental anomaly early warning is improved.

The invention relates to the technical field of intelligent data correction, and discloses an intelligent accounting data correction method and system based on data difference identification, and the method comprises the steps: S, synchronously executing data standardization and data portrait construction based on accounting data and a preset business rule, business rule injection type feature extraction is carried out on the standardized accounting data to obtain a difference composite feature vector, multi-level confidence classification is carried out based on the vector, difference type labels are generated, a cost-risk-efficiency optimization correction decision is carried out, an optimal correction strategy instruction and a traceability record are obtained, and the traceability record is obtained. Performing directional correction process management and control on the data, then searching similar cases, performing case auxiliary verification on the corrected accounting data, generating a correction verification result, and finally outputting a correction traceability report; according to the invention, the accuracy of intelligent accounting data correction based on data difference identification can be improved.

The invention relates to the technical field of on-line monitoring, in particular to a water level on-line detection method and system for water conservancy project waterproofing, and the method comprises the following steps: analyzing water level changes before and after an event, screening abnormal measurement points, recognizing an abnormal region in combination with an excitation signal, and adjusting transmission priority configuration according to data change characteristics. And judging a trend boundary to construct trend structure information, and analyzing a multi-cycle water level record to output a cycle gradient change result. According to the method, through combination of event attributes and the liquid level change relation, abnormal signal identification is optimized, data difference comparison before and after disturbance is utilized, the abnormal area judgment capacity is enhanced, jump trend capture is achieved in combination with the period slope change trend, and a dynamically-adjusted data uploading priority rule is constructed, so that the measuring point channel response has the real-time regulation and control capacity; the accuracy of trend continuation judgment is improved, the water level change direction and the periodic gradient offset state are subjected to association expression, and the dynamic tracking and grading response capability of water level monitoring is enhanced.

The invention belongs to the technical field of data security management, and discloses a smart city data processing method and platform based on artificial intelligence, and a storage medium. Comprising the following steps: monitoring real-time data between network entities in real time to obtain a real-time data stream; constructing a dynamic data blood relationship map by analyzing an instantaneous behavior mode and a content derivation relationship of the real-time data stream; constructing a node symbiont by analyzing the data difference degree between different nodes in the dynamic data blood relationship map; calculating an interaction risk of the node symbiont and an external entity, and constructing a minimum access control rule according to the interaction risk; and constructing a life cycle management model, dynamically managing an access control strategy, and managing and controlling real-time data. According to the method, the optimization effect from passive response to active perception is realized on the data stream security management level, the collaborative effect of the authority structure and system resource allocation is improved, and the intelligence and efficiency of the overall data security management are improved.

The invention discloses a transformer substation-oriented multi-source heterogeneous data checking method and system. The method comprises the following steps of obtaining various types of data sources, executing a preliminary data cleaning process and outputting standardized data; aiming at database systems of different manufacturers, designing a customized migration tool for each database; extracting specific attribute information contained by the special domain analysis module by constructing the special domain analysis module; sequentially verifying the source database and the target database by adopting a progressive consistency verification strategy, and recording data difference information detected in each stage; a change rationality verification module is configured to analyze an SCD file of the transformer substation, and change rationality judgment is performed on difference items from a source database to a target database; performing unified modeling and graphical display on the obtained difference information through a difference data visualization module; according to the method, the data consistency and correctness in the upgrading and transformation process of the transformer substation monitoring system database are comprehensively verified, so that the stability and the operation reliability of the system are improved.

The invention relates to the technical field of multi-source data fusion, in particular to a multi-source communication data fusion method and system and a storage medium, and the method comprises the steps: obtaining various data of communication equipment in real time; acquiring a monitoring deviation index of any two types of data in the current period according to data difference characteristics between any two types of data, and clustering all types of data; according to the data deviation degree between each type of data and all other types of data in the cluster in which the data is located in the current period, the deviation degree of each type of data relative to historical data, and the synchronization degree of the mutation time of each type of data, obtaining a cheating factor of each type of data in the current period; therefore, the initial credibility of various types of data in the current period is optimized, and then all types of data in the current period are fused. According to the method, the data credibility of each type of data is analyzed, the initial credibility of each type of data is optimized, and the accuracy of multi-source communication data fusion is improved.

The invention relates to an intelligent mold calibration optimization method based on big data analysis. The method comprises the following steps: firstly, collecting current state data of a mold, including shape, size, wear degree and production environment parameters, and simultaneously obtaining ideal state data of a product; and constructing a mold-product correlation model by utilizing a physical simulation algorithm, and simulating a product forming process under different production parameters. Key control parameters are determined based on simulation results, the parameters are optimized through a machine self-learning algorithm, and an intelligent calibration optimization strategy is generated. Actual product data is collected in real time and compared with ideal state data, a product data difference value is generated, and an optimization strategy and key control parameters are dynamically adjusted according to the product data difference value. And finally, the production parameters of the mold are automatically adjusted according to the optimization strategy, intelligent calibration optimization is achieved, and the product quality and the production efficiency are improved.

The invention relates to the technical field of data processing, in particular to a pressure vessel data monitoring method based on a high-precision sensor. Obtaining a data period according to data change characteristics of the monitoring data sequence; obtaining the data deviation degree of the current moment according to the data difference characteristics of the current moment and the same moment in the historical data cycle; obtaining a suspected abnormal moment according to the data deviation degree; obtaining comprehensive stability according to the data difference characteristics of the suspected abnormal moment at the same moment in the historical data cycle and the fluctuation difference characteristics of the suspected abnormal moment at the same moment in the historical data cycle; and according to the data deviation degree of the suspected abnormal moment, the comprehensive stability and the quantity characteristics of the suspected abnormal moments of other sensors at the same moment, obtaining the fault abnormity degree. According to the invention, the pressure vessel at the current moment is monitored according to the fault abnormity degree, and the accuracy of pressure vessel data monitoring is improved.

The invention provides a dynamic credit risk assessment and early warning system for a new citizen group, and relates to the technical field of risk early warning. The behavior sequence construction module is used for preprocessing the data and generating user behavior sequence data; the difference feature generation module is used for calculating a difference metric value between adjacent time slices according to the user behavior sequence data; the credit state calculation module is used for calculating credit variation data in combination with the historical credit evaluation record and generating initial credit state data; the dynamic correction module is used for determining a deviation correction parameter according to a matching result of the initial credit state data and an actual repayment record, and generating dynamic credit evaluation data; the risk early warning generation module is used for generating a risk early warning signal when the change rate or the change direction of the dynamic credit assessment data reaches a preset risk triggering condition; according to the invention, the autonomy and accuracy of dynamic credit risk assessment and early warning are improved.

The invention provides a spatial intelligent visual physical process inference method based on an implicit physical large model, and belongs to the field of spatial intelligent and artificial intelligence modeling calculation. The problems of low prediction precision and lack of physical consistency for complex physical scenes in the prior art are solved. The method comprises the following specific steps: acquiring multi-modal data of an environment, and representing the multi-modal data in a unified coordinate system; preprocessing the multi-modal data, designing a geometric coding model, a dynamic prediction model and an energy conservation constraint, introducing a space-time attention mechanism, predicting the motion state of an object according to the multi-modal data, and obtaining prediction data of the motion state of the object; acquiring real observation data of object motion, comparing the difference between the prediction data and the real observation data, and performing correction and weight updating on the prediction data of the model through a self-adaptive residual term; according to the method, visual information and implicit physical law modeling are fused, and the self-supervised physical consistency constraint is utilized, so that automatic learning and prediction of a potential physical process are realized.

The invention relates to the technical field of anomaly detection, in particular to an industrial Internet of Things equipment behavior pattern anomaly detection method. Clustering the data points according to the time domain features and the frequency domain features of the data points in the local time period, and obtaining an outlier degree and a target data point according to the number features of the data points in the data cluster and the distance features between the data points and the maximum data cluster; obtaining a first confidence coefficient according to the data difference of the target data point in other data points in a preset neighborhood range in the clustering space, the number ratio of other data clusters to which the other data points belong, and the discrete feature of the target data point; obtaining a second confidence coefficient according to the data difference between the target data point and the same historical moment, the trend difference between the target data point and the same local time period corresponding to the same historical moment, and the data fluctuation in the local time period of the target data point; and obtaining the abnormal degree of the target data point according to the comprehensive confidence coefficient and the outlier degree, and detecting the behavior state of the equipment, thereby improving the accuracy of anomaly detection.

The invention relates to the technical field of underground pump fault early warning, in particular to a side cloud collaborative diagnosis early warning system for a tower type pumping unit cloud platform. The system comprises an edge end module and a cloud platform module, wherein the edge end module comprises a data acquisition unit used for acquiring operation data of the tower type pumping unit and a local diagnosis unit used for determining a fault moment; the cloud platform unit comprises a remote diagnosis and decision-making unit used for obtaining a fault response value according to the correlation degree of the same type of operation data and the importance degree of the difference of each type of operation data in an analysis period before and after the fault moment; determining a response characteristic sequence based on the fault response value of the same type of operation data at the fault moment; and training a neural network by using the neural network to obtain an oil pumping unit fault diagnosis model, and further performing diagnosis and early warning on the tower type oil pumping unit. According to the invention, the accuracy of diagnosis and early warning of the fault of the oil pumping unit is effectively improved.

The invention relates to the technical field of data processing, in particular to an industrial electromagnetic environment multi-dimensional health degree assessment model, which comprises a processor and a memory, and the processor executes a computer program of the memory to implement the following steps: acquiring a multi-dimensional electromagnetic environment data sequence of any equipment in an industrial electromagnetic environment in a current time period, if the abnormal fluctuation time period exists in the current time period, obtaining the abnormal performance degree of each abnormal fluctuation time period according to the data difference of each abnormal fluctuation time period; according to the data fluctuation characteristics of the multi-dimensional electromagnetic environment data sequence and the change trend of the multi-dimensional electromagnetic environment data of each reference device, the interference degree of any device in each abnormal fluctuation time period is obtained; according to the abnormal performance degree of each abnormal fluctuation time period and the interfered degree of any device, the affected degree of any device is obtained, the health degree of the industrial electromagnetic environment is obtained by using the affected degree of each device, and the accuracy of a health degree evaluation result is improved.

The invention relates to a distributed data consistency characterization and credibility verification method based on a block chain, and the method comprises the steps: collecting multi-source power grid operation data in real time among a dispatching master station, a transformer substation and edge equipment, carrying out the unified structural processing of the multi-source power grid operation data, and extracting a hash abstract to guarantee the normalization of a data source; hash abstracts generated by all the nodes are uploaded to a consistency comparison engine, a multi-dimensional consistency calculation model is established based on key parameters including timestamps, voltages and currents, and data differences and conflicts are identified; state coding is carried out on a comparison result, a power grid data consistency state identifier is generated, and the state identifier is written into a block chain in combination with a device ID and a time label; and the system constructs an index mapping mechanism based on the on-chain state identifier, so that the master station and the dispatching center can perform cross-system consistency verification. According to the method, automatic consistency judgment and credible evidence storage of the distributed power grid data can be realized, and the accuracy of cross-system data sharing, the real-time performance of verification and the safety of system operation are improved.

The invention relates to the technical field of automobile collision safety, and discloses a passenger posture prediction method and system based on active safety early warning, and the method comprises the steps: synchronously collecting vehicle dynamics data, passenger target muscle group surface electromyogram signals and kinematics states under a simulated or real AEB working condition; after preprocessing the data, constructing a myoelectricity-attitude prediction model by using a CNN and LSTM hybrid model; training the model by using a large amount of data to learn the internal relationship between the characteristics and the passenger kinematics state change, and predicting the kinematics state of each section of the passenger body at the collision moment; acquiring a vehicle deceleration curve, inputting characteristics to predict the posture of a passenger, converting into a simulation software instruction to complete initialization of a virtual dummy, and running collision simulation; the problem that the initial position of a traditional collision simulation model cannot reflect the actual posture of a passenger under the emergency braking working condition, and consequently data difference is large is solved.

The application discloses a kind of based on data bias feedback electric energy meter data intelligent analysis method and system, it is related to electric power fault detection technical field, including obtaining electric power system information, according to electric power system information, with electric power system electrical circuit as foundation, electric energy meter is regarded as topological node, obtains electric energy meter topological relationship graph, based on electric power system information, obtains electric energy meter historical data.The application is through data difference coefficient, on the basis of considering the influence of electric energy meter association, accurately assesses the data variation range of electric energy meter, provides data basis for subsequent abnormal data identification, establishes electric energy meter association feature chain based on electric energy meter topological relationship graph to classify electric power system area, improves electric power fault detection efficiency, detects electric power fault through electric energy meter real-time data and data difference coefficient, improves the accuracy and timeliness of electric power fault detection, ensures the stability of electric power system.

The application discloses a method and system for predicting water inrush of sandified dolomite stratum under water-rich conditions, comprising: obtaining first detection data and second detection data; extracting first cross-section data and second cross-section data; the cross-section data is position data of a cross-section with discontinuity; comparing the first cross-section data and the second cross-section data, and taking a region with a difference greater than a preset value between the first cross-section data and the second cross-section data as a target drilling region; obtaining third cross-section data; forming cross-section fitting data of the target cross-section; and predicting construction water inrush risk. The method and system for predicting water inrush of sandified dolomite stratum under water-rich conditions effectively improve the advanced detection capability of sandified dolomite tunnel construction, improve construction safety and reduce construction risk by data fusion on the detection data in front of a tunnel face and accurate data acquisition and completion in combination with the advanced drilling technology.

The invention discloses a method for dynamically putting a questionnaire in a third-party application through an SDK, and belongs to the technical field of computer data process.The method comprises the steps that six types of data are collected and subjected to multi-factor weighted decision making, a multi-platform exclusive integration scheme is provided, a questionnaire display and no-disturbing strategy is configured, the display effect is intelligently triggered and optimized, and answer data are verified in real time. Differentiated submission and multi-mode synchronous high-efficiency circulation are realized. According to the method for dynamically putting the questionnaire in the third-party application through the SDK, the compatibility and flexibility of questionnaire putting are improved through the multi-dimensional factor weighting decision model, and the method is suitable for digital operation scenes such as user feedback collection and demand investigation of various third-party applications.

A metal modified graphene structure model is applied to predict hydrogen storage materials, comprising a structure simulation unit, a simulation data unit, a preparation measurement unit, a data processing unit and a parameter feedback unit; the structure simulation unit includes a graphene base structure composed of carbon atoms and a plurality of modified structure models pre-prepared for metal modification of the graphene base, combined into hydrogen storage materials; the simulation data unit calculates the structure of H2 molecules adsorbed on the hydrogen storage materials and calculates the simulation hydrogen storage data; the preparation measurement unit actually prepares the hydrogen storage materials and measures the actual hydrogen storage data; when the actual hydrogen storage rate data is greater than or equal to the first threshold value, the data processing unit calculates the difference percentage of the actual hydrogen storage data and the simulation hydrogen storage data, and outputs the data difference degree; otherwise, an abnormal warning signal is output; when the data difference degree is less than or equal to the second threshold value, the parameter feedback unit outputs that the hydrogen storage data of the preparation measurement unit is qualified, otherwise, an abnormal warning signal is output and feedback is performed, and data optimization is performed.

The invention belongs to the technical field of advertisement monitoring, particularly discloses an advertisement effect conversion tracking method, system and program product based on a front-end burying point, aims to solve the problems of inaccurate advertisement service data and conversion attribution breakage, and designs a systematic scheme with front and rear ends cooperating and data streams penetrating through a user complete operation link. Through an innovative uniform resource locator cross-page transmission mechanism, all paths of the user from first contact with the advertisement elements to final conversion can be completely reconstructed, efficient advertisement conversion attribution of the advertisement service data is realized, and a data basis is provided for accurate evaluation of the advertisement ROI. Meanwhile, through a behavior bottom exposure mechanism and a unified identifier system, the consistency of data sources on which front-end advertisement exposure and back-end transaction logic depend is ensured, and the data difference is reduced fundamentally.

The invention relates to the technical field of industrial data processing, in particular to a computer-aided transformer substation three-dimensional model data dynamic mapping method. Transformer substation monitoring data and model data after three-dimensional model mapping are obtained, and multi-dimensional monitoring can be achieved; however, if the three-dimensional model mapping is not timely and accurately refreshed, the virtual and actual physical states are deviated; therefore, in dynamic mapping, based on electric quantity fluctuation, control response data time sequence interval and multi-data source deviation joint analysis, a dynamic mapping distortion index is constructed, and the matching degree of a quantitative model and an actual state is quantified; in view of continuous existence of delay in operation of the transformer substation, determining a model error change index and representing an error change accumulation degree by analyzing an electrical quantity time sequence, a model data difference change rule and an alarm log change trend; and finally, dynamically adjusting the model mapping refresh frequency according to the two indexes, so that the refresh frequency adapts to a data state, and the balance of efficiency and precision is realized.

This invention relates to the field of data monitoring technology, specifically to an intelligent control method for workshop air purification based on environmental data. The method involves: obtaining upward trend points and concentration inflection points based on changes in concentration data sequences; identifying concentration rise periods and concentration fall periods based on the distribution characteristics of concentration inflection points; identifying suspected abnormal data points based on the differences in the trends of concentration rise and fall periods and the local data differences at the upward trend points; determining the air pollution risk level based on the quantity, interval, and concentration change characteristics of the suspected abnormal data points; and adaptively adjusting a preset concentration anomaly threshold based on the air pollution risk level to obtain an adaptive threshold. This invention adjusts ventilation power based on the adaptive threshold, thereby reducing the power consumption of the air ventilation system while improving the efficiency of air purification in the workshop.

The invention discloses a data self-adaptive correction and acquisition method and device for dynamic environment monitoring. The method comprises the following steps: configuring key data item test points, classifying the priorities of the test points, setting update levels and weights, allocating polling resources to each level of test point, and acquiring dynamic environment monitoring data of each data item; creating a task in the power and environment monitoring system, performing field measurement, calculating the difference between power and environment monitoring data and actually acquired data, and analyzing a difference mode; dynamically adjusting the parameters of the dynamic environment monitoring system through an adaptive correction algorithm according to the difference mode to obtain corrected data; a selective polling strategy is adopted, polling priority weight and polling frequency are dynamically adjusted, and polling resources are updated and distributed; constructing a difference data prediction model to predict future differences, and judging whether the future differences exceed a preset safety threshold value or not to perform error correction; according to the invention, difference modes can be identified in real time, environment and equipment state changes can be handled, and monitoring data accuracy and system response timeliness can be improved.

The invention discloses a system for rapidly identifying data difference through vehicle CFD analysis, and relates to the technical field of vehicle data processing, and the system comprises a file analysis module which is used for calling a simulation result file; the parameter extraction module is used for capturing key parameters according to the simulation result file and a category rule; the geometric quantization module is used for constructing a geometric difference degree; the grid evaluation module is used for determining grid quality parameters and constructing geometric grid association mapping; the topology construction module is used for constructing a parameter topology tree according to the key parameters; the data management module is used for constructing a simulation data table based on the key parameters, the geometric difference degree, the geometric grid association mapping and the parameter topology tree so as to generate a simulation management library; the difference marking module is used for formulating a data tolerance strategy and performing difference marking on the simulation data table to obtain a marked data table; and the data output module is used for outputting the marked data table. According to the invention, accurate difference identification and marking of the vehicle CDF data are realized.

The invention discloses an image target detection enhancement method and system based on controllable synthesis and reprojection correction, and the method comprises the steps: injecting noise into an input image, obtaining a disturbance image, carrying out the single-step denoising of the disturbance image through an image generator, and generating a first synthesis image; a target detector is adopted to map the first composite image back to a layout space, and a prediction layout of the input image is obtained; according to the method, the functions of adopting a layout detection reprojection correction item to take the input layout as a geometric and semantic anchor point explicit press-back detection learning process and inhibiting layout offset and annotation noise caused by controllable synthesis or data difference are realized; the triggering frequency at the alignment moment can be effectively improved by combining one-step reduction of noise injection with time-step weighted sampling, and perception-level lightweight alignment can be carried out with low weight by introducing synthetic observation appearance correction, and the adverse effect of style difference on detection learning is relieved; and the positioning and classification accuracy of target detection in a complex scene is remarkably improved.

The invention relates to the technical field of data processing, in particular to a massage therapy patient physiological index data monitoring and analyzing method, which comprises the following steps of: acquiring a multi-dimensional physiological index data sequence of a patient in a current monitoring period, constructing a difference matrix of the multi-dimensional physiological index data sequence, and acquiring the structural complexity of the difference matrix; if the structural complexity is greater than a preset structural complexity threshold value, acquiring a data change time period and a continuous characteristic index thereof, dividing the multi-dimensional physiological index data sequence into at least two subsequences, acquiring a residual component corresponding to any round of iterative decomposition in the STL decomposition process of any subsequence, and obtaining a residual component corresponding to any round of iterative decomposition; and according to the data difference between the residual component in the current monitoring period and the data change time period and the continuous characteristic index, obtaining an adaptive truncation parameter to carry out the next round of iterative decomposition until a decomposition result of any sub-sequence is obtained, and obtaining the decomposition result of each sub-sequence to carry out data analysis, thereby improving the analysis accuracy.

The present application relates to the technical field of engineering blasting optimization, in particular to a dynamic optimization design method based on engineering blasting parameters, comprising: collecting stratum geological parameters such as rock hardness profile and rock mass fracture distribution of blasting site and blasting vibration three-direction velocity time history monitoring data, carrying out time-frequency joint analysis on time history record to extract dispersion curve and energy attenuation envelope, and constructing a digital rock mass structure model capable of calculating blasting stress wave propagation path. The vibration characteristic data is used as boundary condition to carry out numerical simulation to obtain simulated vibration velocity field, according to the difference between simulated data and measured data, combining with rock mass wave impedance distribution, the intelligent optimization algorithm is used to iteratively optimize the charge structure parameters and initiation time sequence scheme, and the optimized blast hole charge quantity, section and initiation delay are output. The method can improve the accuracy of blasting vibration propagation simulation, reduce the deviation between simulated and measured vibration velocity fields, and make the blasting parameter design more suitable for the actual geological environment of the engineering site.

Provided are a computer program product, a system, and a computer-implemented method for performing data discrepancy analysis. Data is received from a plurality of source data systems. The data is analyzed to identify a data discrepancy of a data field. A recommendation is generated that indicates whether the data discrepancy is fixable. In response to the recommendation indicating that the data discrepancy is fixable, the data discrepancy is fixed by modifying a value of the data field, and the data is routed to a first downstream system service that does not rely on the data field and to a second downstream system service that does rely on the data field. In response to the recommendation indicating that the data discrepancy is not fixable, the data is routed to the first downstream system service, while the second downstream system service is disabled.

The invention discloses a fire-fighting inspection prediction method and system based on the Internet of Things, and relates to the technical field of fire fighting, and the method comprises the steps: collecting multi-source data, such as temperature, smoke concentration and electrical parameters, in an environment through multiple types of sensors, and transmitting the multi-source data to a data fusion terminal; and in the data fusion terminal, dynamic weighted fusion is carried out on the multi-source data through an adaptive data fusion algorithm, a basic data set for fire risk prediction is generated, and a prediction model is initialized. The problem of fusion precision caused by data differences of different sensors is solved; nonlinear anomaly detection is carried out by adopting a Gaussian mixture model, so that wrong data are accurately eliminated, and the data quality is improved; and the risk factor weight and association relationship are dynamically adjusted in combination with an optimization prediction algorithm, so that the real-time performance and accuracy of fire risk prediction are enhanced.