150 results about "Analysis of algorithms" patented technology

The invention discloses a security monitoring video intelligent analysis method based on multi-algorithm cooperation and unified architecture, and relates to the technical field of intelligent video monitoring, and the method comprises the steps: collecting a security monitoring video, carrying out the preprocessing, carrying out the spatial-temporal feature extraction and fusion through a CNN-LSTM spatial-temporal fusion engine, and outputting a fusion feature map; performing target detection, behavior recognition and anomaly detection on the fused feature map to obtain a multi-algorithm analysis result; a cross-module fusion mechanism based on an attention mechanism is utilized to perform weighted integration processing on the multi-algorithm analysis result to obtain a fusion event representation vector; performing event type identification and risk level evaluation on the fusion event representation vector to obtain an event classification result and an event risk level; according to the invention, through the CNN-LSTM space-time fusion engine, the front-end perception capability of abnormal behaviors in a complex scene is improved, and the event detection accuracy and the anti-interference capability are improved.

A method comprises receiving a large language model request, analyzing the large language model request using one or more machine learning algorithms, and predicting, based at least in part on the analyzing: (i) a large language model of a plurality of large language models to process and to respond to the large language model request; and (ii) at least one database from which data is to be used to generate a prompt for the large language model. The method further comprises interfacing with the large language model and the at least one database to enable the large language model to process and to respond to the large language model request.

The invention discloses a message-driven enterprise multi-service system centralized to-do processing method and system, and the method comprises the steps: S1, obtaining resource investment and task processing data, and obtaining a resource consumption and task state cluster through clustering; s2, using a regression algorithm to analyze correlation between the two, and determining a resource utilization rate change trend; s3, when the threshold value is exceeded, calculating a task completion rate statistical characteristic to obtain a resource optimization index; s4, constructing an interactive model by using a decision tree in combination with historical data, and judging an input strategy adjustment point; s5, rendering visual resource distribution; s6, abnormal tasks are screened, and an optimization area is determined; and S7, updating an instrument panel to obtain a decision view. Through data clustering, correlation analysis and dynamic adjustment, the resource and task matching degree is improved, and decision scientificity and efficiency are enhanced.

The invention relates to the technical field of nuclear power data asset automatic management, in particular to a nuclear power digital twinning data warehouse structure system, which comprises a data processing layer, and is characterized in that the data processing layer reads original data of a nuclear power industrial internet platform, performs data standardization processing on the original data to obtain standard data, and sends the standard data to the nuclear power industrial internet platform; constructing a nuclear power industry theme library according to the standard data; the data integration layer performs data coding mapping, data classification and data fusion on the data of the nuclear power industry theme library to form a thematic library; the semantic enhancement layer marks the data of the nuclear power industry theme library and the data in the thematic library formed by the data integration layer, and constructs a structured feature tag library for the data associated with feature tags; the analysis application layer reads data of the feature tag library, and constructs a digital twinborn algorithm library and a digital twinborn sample library through algorithm analysis and sample training. The system supports equipment operation and maintenance from regular maintenance to predictive maintenance based on bottom support and four-layer progressive architecture of a nuclear power industrial internet platform.

The invention discloses an electric power field operation safety monitoring method and system based on AI driving. The method comprises the following steps: synchronously acquiring voice, operation video and equipment state identification information, and de-noising to generate a multi-modal sequence; using a multi-modal cross reconstruction model to reconstruct other modals for any modal, and identifying suspicious fragments according to reconstruction errors; multi-layer consistency matching is executed on the suspicious fragments, and the consistency matching degree and mismatch attribution are calculated through time soft alignment, semantic embedding unification and semantic graph reasoning; dispersing the voice, action and equipment state elements into an event sequence, and matching the event sequence with a preset process causal graph to position an abnormal event; and finally, abnormal nodes and induced nodes are analyzed in combination with a Bayesian algorithm, and whole-process safety monitoring is realized. According to the method, password inconsistency, action out-of-order, object mismatching and equipment response abnormity in electric power operation can be intelligently identified and traced, and the real-time performance and accuracy of operation safety are improved.

The invention relates to the crossing field of engineering cost management and building informatization (BIM) technology, and discloses a BIM-based multi-specialty collaborative engineering cost allocation and settlement auditing system, which comprises an engineering cost allocation and settlement auditing system, and the engineering cost allocation and settlement auditing system comprises a sensing layer, a platform layer, a core layer, an application layer and a guarantee layer. According to the BIM-based multi-specialty collaborative project cost allocation and settlement auditing system, a five-layer system architecture (a sensing layer, a platform layer, a core layer, an application layer and a guarantee layer) is constructed, so that the whole process from data acquisition, standardized processing and intelligent algorithm analysis to multi-role collaborative application is realized; according to the method, cost allocation and settlement auditing services under multiple specialties and multiple scenes are supported, and multi-dimensional cost intelligent allocation and automatic anomaly detection based on BIM components are realized by integrating a BIM semantization engine, a dynamic allocation algorithm engine and an AI auditing engine.

The invention discloses a main beam formwork erection elevation adjustment method based on machine learning, and belongs to the technical field of elevation adjustment. The method comprises the steps that a space-time database is constructed by collecting data of the construction period of a cable-free section; processing the time sequence data by adopting an LSTM-Transform hybrid model, extracting long-period characteristics, and outputting an initial prediction value of the elevation; analyzing nonlinear variables such as a cable force change rate, a sunlight gradient and a material age by using an XG-Boost algorithm, and outputting a compensation factor; fusing and generating a joint prediction value; constructing a reinforcement learning agent, and outputting an adjustment instruction by taking a construction stage as a state space, taking an elevation adjustment amount as an action space, minimizing deviation between a predicted value and a measured value and taking construction stability as a reward function; the formwork erecting elevation is adjusted according to the driving hydraulic system, and the database is updated in real time. According to the invention, through multi-dimensional data fusion and intelligent optimization, the elevation adjustment precision and stability are improved.

The invention discloses a state diagnosis and early warning system and method for power switch equipment, and belongs to the technical field of power equipment monitoring. According to the system, multi-dimensional sensing data of temperature, vibration, partial discharge and the like of the switch cabinet are uniformly collected and locally processed through an edge calculation module of a built-in multi-protocol Internet of Things gateway. On one hand, the edge calculation module carries out real-time state judgment based on a multi-dimensional threshold value; and on the other hand, the trend diagnosis module analyzes historical data by using an XGBoost algorithm to realize trend prediction and early anomaly recognition. And the comprehensive diagnosis module fuses the two types of results and dynamically adjusts the state level so as to realize accurate early warning. According to the method, comprehensive perception and on-site intelligent analysis are realized, the problems of one-sided perception, response delay, insufficient intelligence and data island existing in a traditional scheme are effectively solved, and the real-time performance, diagnosis accuracy and operation and maintenance efficiency of equipment state monitoring are remarkably improved.

The invention relates to an algorithm and a system for evaluating and detecting agile strength movement postures of special children. The system comprises a user information initialization module, a detection task guide module, a posture recognition module, an index extraction and algorithm analysis module, an auxiliary equipment module and a detection result output module. The system recognizes skeleton key points by collecting video images, extracts multi-dimensional motion indexes such as rhythm synchronization, attitude stability, reaction time delay and motion symmetry, and performs comprehensive scoring by using normalization processing and weight superposition algorithms. In the detection process, a guiding interface is matched with voice prompt to achieve standardized operation, and a structured image-text evaluation report and personalized training suggestions are output. The system is suitable for individualized action ability detection in education, rehabilitation and family scenes, has the characteristics of convenient deployment, accurate recognition, quantifiable result, strong adaptability and the like, and effectively solves the problems of scattered evaluation process, subjective scoring, functional dimension deficiency, poor individual adaptability and the like in the prior art.

The invention relates to the technical field of data processing, and discloses an AMR transmission path data comprehensive processing method and system. The method comprises the following steps: processing environment sensing data through a data fusion algorithm to obtain an environment state matrix for recording information of an obstacle, a dynamic target and a passable area; robot state information and the environment state matrix are combined to obtain a joint state space integrating global and local information; training the joint state space through a multi-agent reinforcement learning algorithm to obtain a collaborative decision model composed of an independent strategy network and a shared evaluation network; processing the control instruction through a trajectory prediction model to obtain a collaborative path set storing robot path information; and analyzing the collaborative path set through a conflict detection algorithm, and triggering a scheduling strategy to obtain a path optimization result when a conflict is detected. According to the invention, the cooperative efficiency of the multi-robot system and the intelligent level of path planning are obviously improved.

The invention discloses a hydrological telemetering rainfall data cleaning method based on machine learning, and the method comprises the following steps: S1, association site screening: selecting a telemetering rainfall target site, employing an XGBoost algorithm to analyze the space-time correlation between sites, and selecting an auxiliary rainfall site with strong correlation; s2, identification of missing measurement data: based on the auxiliary rainfall station data selected in the step S1, identifying a rainfall missing measurement value of a target station through a space-time correlation comparison and discrimination algorithm; s3, abnormal data detection: carrying out secondary identification on non-missing detection data by adopting a difference value discrimination algorithm, and positioning a rainfall abnormal value; s4, model construction and training; S5, hyper-parameter optimization; and S6, data correction and storage. According to the method, accurate screening of data exceptions is realized by adopting a two-stage recognition architecture, data correction is more scientific and reliable, and the data quality of the whole process is greatly improved. The system is convenient and fast to use, and provides scientific basis and technical support for regional water resource simulation, scheduling and emergency decision making.

The invention provides an abnormal transaction early warning method, device and equipment, a medium and a program product, which can be applied to the field of big data and the field of financial science and technology. The method comprises the steps of calculating an average path length of transaction data to be detected based on a transaction feature isolation forest model, and determining a first class score according to the average path length; wherein the transaction feature isolation forest model is constructed by analyzing a first type of historical transaction data through an isolation forest algorithm; on the basis of transaction discrimination features, a second class score of the transaction data to be tested is calculated through a random forest algorithm, and the transaction discrimination features are obtained by analyzing second class historical transaction data through the random forest algorithm; analyzing the first class score and the second class score, and determining an abnormal level of the to-be-tested transaction data; and based on the abnormal level, early warning the abnormal risk of the to-be-tested transaction data in real time.

The invention discloses an English reading ability training system and method based on artificial intelligence, and aims to solve the problems of insufficient personalization, poor interactivity, feedback lagging, imperfect evaluation system and low resource matching degree in an existing English learning platform. The system comprises a data acquisition module, a user analysis and personalized path planning module, a dynamic content generation and practice design module, a real-time monitoring and feedback adjustment module and a comprehensive evaluation and report generation module. Basic information and real-time feedback of the user are collected, learning characteristics, interests, preferences and ability levels of the user are accurately identified by using a machine learning algorithm, and a personalized learning path is provided for the user; meanwhile, natural language processing, machine learning and text generation technologies are integrated, and reading materials meeting user requirements are dynamically generated; analyzing the comprehensive difficulty of the generated material by utilizing a machine learning algorithm, wherein the comprehensive difficulty is quantified by adopting a text comprehensive difficulty score TCS; reading understanding exercises including vocabularies, grammar, reasoning and critical thinking are provided, and the subsequent exercise difficulty is intelligently adjusted based on user performance data so as to optimize the learning effect. The system also continuously tracks user performance data, analyzes progress and problems through a machine learning algorithm, improves the regular evaluation capability, automatically adjusts a learning path and a material strategy, and periodically generates a multi-dimensional comprehensive evaluation report.

The invention discloses a digital human virtual human setting generation method, and relates to the technical field of data processing, and the method comprises the steps: responding to an input task theme, a target region identifier and an industry classification identifier, and generating a human setting genotype structure according to the target region identifier and the industry classification identifier; constructing and generating an environment control file based on the human-set genotype structure; loading the generation environment control file into a pre-trained generation model, and generating differentiated contents according to the task theme; the cognitive homogenization risk of the differentiated content is analyzed through a graph isomorphism detection algorithm; and if the graph similarity of the cognitive homogenization risk exceeds a preset threshold value, updating the human genotype structure. According to the method, the homogenization degree of the output content of the large model is reduced.

A computer implemented system useful for determining turns in a user's trajectory, including one or more processors; one or more memories; and one or more programs stored in the one or more memories, wherein the one or more programs executed by the one or more processors (1) receive input data, comprising orientation data and acceleration from one or more sensors carried by a user taking steps along a trajectory; (2) analyze the input data using a first trained machine learning algorithm so as to detect a plurality of n straight sections when the user is walking along an approximately straight path; (3) comprise an orientation tracker tracking an orientation of the user in each of the n straight sections, wherein the orientation comprises an estimated orientation taking into account drift of the input data outputted from the one or more sensors; and (4) comprise a turn detector detecting each of the turns, wherein each of the turns is a change in the estimated orientation of the user in the nth straight section as compared to the estimated orientation in the (n−1)th straight section.

The application relates to an artificial intelligence-based multi-water-source dynamic allocation system and method. The system comprises a data acquisition module, which collects equipment operation data and environmental factor data of multiple water source points through a sensor network and aggregates the data into water source characteristic data; a data classification module, which classifies the water source characteristic data using a support vector machine algorithm to obtain a parameter set; a trend analysis module, which analyzes the change trend of the water source characteristic data using a random forest algorithm; a weight adjustment module, which dynamically adjusts the parameters according to the trend to generate a weighted parameter set; a scheme generation module, which generates a water resource allocation scheme by simulating a multi-water-source coordination scenario based on the weighted parameter set through a pre-trained neural network model; an effect verification module, which verifies the effect of the scheme by using a simulation method; and a scheme optimization module, which optimizes the dispatching parameters according to the verification result to obtain an optimized water resource allocation scheme. The method can realize efficient dynamic allocation of water resources and optimization of overall operation efficiency.

The invention belongs to the field of educational training, and mainly relates to an AI digital human modeling method based on a user portrait, which comprises the following steps: setting model constraint conditions by combining course details, and generating a digital human appearance conforming to course features by using a conditional generative adversarial network; in the teaching process, real-time video data of a teaching site is acquired by using image acquisition equipment, identity and emotion information is extracted through a facial recognition and detection algorithm, and behavior logic of an AI digital human assistant is triggered according to an analysis result to assist a teacher in regulating and controlling the classroom rhythm; in combination with the learning record and the answering behavior data, constructing a knowledge point mastering graph, evaluating the learning condition of the student, and analyzing individual characteristic data through a machine learning algorithm; along with accumulation and updating of the data, the user portrait is dynamically improved, so that targeted teaching suggestions are provided for teachers, behavior logic and teaching paths of the AI digital human assistant are optimized, and personalized and differentiated teaching support is realized.

The invention provides a multistage unit inter-stage fault diagnosis method based on flow resistance characteristic identification, and belongs to the technical field of hydraulic machinery fault diagnosis. The problems that in the prior art, the internal structure of a multi-stage pump turbine is closed, and inter-stage faults are difficult to locate and determine can be at least partially solved. The method comprises the following steps: acquiring operating parameters of a unit and opening degrees of guide vanes at all levels in real time; calling a preset flow resistance characteristic analysis model based on a fluid dynamic mechanism, and calculating a theoretically shared water head of each stage of runner under the current working condition; meanwhile, the actual shared water head of each stage of runner is obtained; calculating a flow resistance deviation coefficient between the actual sharing water head and the theoretical sharing water head; and according to the numerical value interval of the flow resistance deviation coefficient, determining a fault type of a specific level. According to the method, hardware equipment does not need to be added, online accurate diagnosis of faults between sealing ring abrasion leakage and flow channel foreign matter blocking grades can be achieved only through algorithm analysis, and the operation and maintenance cost is remarkably reduced.

Technical field: The invention belongs to the field of online educational platforms and networks, specifically systems and methods for supporting students through online mentoring and tutoring. Abstract: The invention relates to an online mentoring and tutoring network for supporting students in achieving their learning goals. The network integrates experts from various fields who offer individually tailored guidance. A central, cloud-based system connects mentors and students, enabling scalable and secure data processing and storage. An algorithm using machine learning and AI analyzes student and mentor profiles to create optimal pairings. The platform supports real-time interactions through WebRTC technology, enabling video calls, chat, and screen sharing. Mentors create customized learning plans based on data analysis of learning progress and the student's individual needs. The platform utilizes big data and analytics tools to monitor learning progress and generate reports.After each session, students and mentors can provide ratings and feedback to continuously improve the support. Inventive achievement: The invention integrates modern technologies such as cloud computing, machine learning, AI, WebRTC, big data and sentiment analysis to create a personalized, efficient and flexible learning environment, offering a significant technological advancement over existing platforms.

The application relates to the technical field of civil aviation information processing, and provides a feature screening method and system for civil aviation price anomaly detection and an electronic device, the method comprising the following steps: performing data cleaning on an original price calculation log file; screening all effective information which has a significant influence on a civil aviation price anomaly detection task from the cleaned price calculation log file, to form an effective information set, and extracting a feature vector; constructing the feature vector based on a construction rule provided by a business expert and a business feature which influences the civil aviation price in a civil aviation business scenario; and performing effectiveness verification on the extracted feature vector and the constructed feature vector. The application can accurately and efficiently screen out key features which influence the civil aviation price, provide a data feature vector with high-value semantic information for a civil aviation anomaly detection task, and improves the accuracy and reliability of subsequent algorithm analysis.

A computer program for managing and analyzing a user's personal wagering data to generate real-time personalized wagering statistics for the user's review. When executed by a processor, the computer program causes the processor to receive the user's personal wagering data and track the outcomes of games. The processor executes an algorithm to analyze the user's wagering data and generate personalized wagering statistics. The processor generates an output illustrating the personalized wagering statistics and the output is delivered to a user interface for the user's review.

The invention particularly relates to an AI cloud desktop operation strategy intelligent issuing method and system. The AI cloud desktop operation strategy intelligent issuing method comprises the following steps: dividing a client cluster into a plurality of resource pools, and assigning an initial main client for each resource pool; each client acquires request platform state data, request response time and response content in real time, encrypts the acquired data and uploads the encrypted data to the algorithm analysis module; the algorithm analysis module analyzes the collected data and generates an adaptive request strategy; and the strategy broadcasting module broadcasts the generated request strategy to all the slave clients in the same resource pool from the master client, and each client executes request operation according to the request strategy and feeds back an execution result to the algorithm analysis module for dynamic adjustment. According to the AI cloud desktop operation strategy intelligent issuing method and system, the peak pressure of the server is relieved, the strategy issuing efficiency is improved, the operation cost is reduced, and the use satisfaction degree of the user is improved.

The application discloses a kind of multi-modal intelligent question and chart generation system and method based on large model, it is related to big data analysis technical field, the application is analyzed by calling historical intelligent question service log, parse and extract complete structured database query statement, after eliminating invalid log, the corresponding relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of key word and SQL core element is analyzed by parsing correlation, and the correspondence relationship of

The invention relates to the field of public safety. The invention discloses a suspected target avoidance behavior pattern recognition method, which comprises the following steps: abnormity study and judgment: S1, quantifying a target trajectory space-time entropy, and distinguishing abnormal and non-abnormal behaviors; and S2, analyzing a track missing area by using a mining algorithm, identifying information black holes, and generating an abnormal target list. Abnormality early warning: A1, according to a black hole surrounding track and a background element, a target position path in a black hole is assumed; a2, establishing a recurrence period model to identify abnormal access; and A3, determining a suspicion schedule in combination with movement continuity and a recurrence period, and assisting interception monitoring. The method is divided into two stages of anomaly study and judgment and early warning, breaks through the limitation of a traditional technology, identifies a suspected target by taking an information black hole as a core and by means of space-time entropy, space-time prisms and the like, and fills the blank of no observation area behavior mining; the two-stage framework converts data into intelligence, constructs a new analysis system, expands location big data application, and improves public security dynamic management and control and interception efficiency.

The invention relates to a self-adaptive management method and system for a database plan cache, and the method comprises the steps: firstly judging whether the plan cache is hit or not after a database receives an SQL statement, and triggering a hard analysis generation plan and caching the plan when the plan cache is not hit; if the cache is hit, further judging whether a corresponding multi-execution plan exists or not; and when only a single execution plan exists, analyzing the relationship between the number distribution of execution tuples and the number of execution times by adopting an execution efficiency algorithm to dynamically judge whether the data is inclined. And when a plurality of execution plans exist, entering a multi-plan cache management stage, selecting the execution plans according to a matching result of the current binding variable predicate selection rate and the cache plan historical selection rate, and generating and caching a new plan when the matching fails. According to the method, the data skew is accurately identified by executing the efficiency algorithm, intelligent matching and multiplexing of plans are realized by means of a multi-plan cache management mechanism, and the overall performance of the database in a data skew scene is improved.

Systems and methods provide for integrating legacy systems into a unified platform. Embodiments include a system configured to ingest, normalize, and store data from multiple sources using a data layer, analyze and transform data using AI algorithms by a processing layer, and provide a dynamic user interface for data visualization and interaction by a presentation layer. Methods are provided for receiving, by a computing device, data such as email orders, vendor catalogs, CRM data, etc., extracting data from the content, processing the data using natural language processing techniques, analyzing the data using machine learning models, mapping the data to a unified platform's schema, managing the integration workflow using a workflow orchestration engine, providing real-time data insights, and resolving data errors or anomalies using an exception management system.

The application relates to the technical field of architectural design, in particular to a project work hour statistical analysis system for architectural design application behavior monitoring and algorithm analysis, which comprises a page behavior record module, a task trajectory generation module, a project identification collection module, a personnel time consumption verification module and an error offset tracking module. In the application, a task window and an operation path sequence in design software are extracted, irrelevant operation contents are removed, a task behavior chain is reconstructed, a clear task execution structure is formed, a continuous operation time period is aggregated by combining task attribution information, a precise mapping relationship between behavior data and project content is established, effective work hours and responsible personnel are matched and compared after collection, a time offset section between behavior records and declaration data is identified, an offset rate is quantified, and work hour declaration rationality is determined, so that dynamic checking and precision improvement of the work hour distribution and the task corresponding relationship are realized.

The invention discloses a multi-point jet flow self-adaptive grading pressure relief control system and method, and relates to the technical field of pressure relief control, and the system comprises a multi-sensor array module, a data processing module, a three-dimensional dynamic pressure field modeling module, a control decision module and a multi-point jet flow execution module. The multi-sensor array module is arranged in a key area of a pressure relief object and used for collecting pressure, temperature and flow velocity data of a jet point in real time. According to the invention, multi-dimensional data are collected through a multi-sensor array, a three-dimensional dynamic pressure field model is constructed, a pressure propagation rule is analyzed in combination with an intelligent algorithm, and precise decoupling control of jet point pressure is realized; the problems that in the prior art, triggering depends on a static pressure threshold value, a complex time-varying pressure field cannot be adapted, and pressure relief lag or excessive adjustment is prone to occurring are solved, precise adaption of a pressure relief strategy and dynamic changes of the pressure field is achieved, the accuracy and reliability of pressure relief control are improved, stable and safe operation of the system is guaranteed, and meanwhile the energy utilization efficiency is optimized.

The invention relates to the technical field of traffic management and deep learning, and discloses a key vehicle illegal behavior identification and disposal management method and system based on deep learning. Historical track data, vehicle-mounted sensor data and a driver behavior video sequence of key vehicles are acquired, spatial-temporal features are extracted by using a three-dimensional convolutional neural network and a graph convolutional neural network, and a behavior node embedding vector set is generated; analyzing the time sequence dependency relationship by using a Granger causality test algorithm to generate a causality adjacency matrix; constructing a causal chain for explicitly modeling driving behavior evolution by a time sequence causal map; and after the real-time behavior characteristics are mapped to map nodes, multi-hop reasoning is carried out by using a map attention propagation algorithm, activation probability distribution of each illegal behavior node in a future time window is calculated, and high-probability illegal behavior nodes are extracted to generate a pre-judgment result set and output to a disposal management system. According to the invention, active pre-judgment before illegal behaviors occur is realized, and the initiative and the prevention effect of disposal management are improved.

The present application relates to dynamic time warping and graph neural network technology, in particular to a load data space-time correlation analysis method based on space-time graph neural network, which solves the technical problem that the accuracy improvement space of the traditional wind-solar-power prediction method is limited due to single consideration factor, and discloses the content of using space-time graph neural network to analyze the space-time correlation of load data, first using the FastDTW algorithm to analyze the space-time correlation of the obtained load data, constructing a correlation matrix, and finally using the obtained data set to train multiple independent parallel space-time graph neural networks to obtain the space-time characteristics of the load data and provide high-quality basic data for prediction.