Space-time co-occurrence analysis method and system for multi-source data fusion

Abstract

The invention relates to the technical field of data processing and analysis, and discloses a space-time co-occurrence analysis method and system for multi-source data fusion, and the method comprises the steps: firstly obtaining the multi-source heterogeneous data of a target area, carrying out the space-time calibration, constructing a space-time reference coordinate system, carrying out the space-time alignment, and generating a space-time weight matrix and a correlation feature vector set; and obtaining a space-time correlation characteristic spectrum through space-time sliding window segmentation and characteristic fusion, and training a co-occurrence mode mining model to extract a space-time co-occurrence mode characteristic set. And constructing an anomaly detection model and verifying to obtain an anomaly co-occurrence mode set, and constructing a parameter optimization model to dynamically adjust data acquisition parameters. And finally, updating the coordinate system according to the optimal parameters and the real-time data and generating an analysis report. According to the method and the system, multi-source heterogeneous data can be effectively processed, a space-time co-occurrence mode can be mined, abnormity can be accurately detected, data acquisition parameters can be optimized, and powerful support can be provided for decisions in related fields.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and analysis, and particularly to a spatio-temporal co-occurrence analysis method and system for multi-source data fusion. Background Art

[0002] In today's digital age, data, as a key production factor, has witnessed an explosive growth in scale and variety. In numerous fields such as urban management, environmental monitoring, and intelligent transportation, people need to deeply understand and analyze various phenomena and activities within a specific area, which makes the acquisition and processing of multi-source data an inevitable requirement. However, these multi-source data often have heterogeneity, posing many challenges to the effective utilization of data.

[0003] Geospatial data describes information such as the location, shape, and distribution of geographical entities, but its formats are diverse, such as vector data, raster data, etc., with significant differences in data structure and storage methods. Device perception data is collected by various sensor devices, and its collection frequency, accuracy, and data types vary. For example, the data formats and meanings collected by temperature sensors, humidity sensors, etc. are all different. Event record data usually exists in the form of text or structured tables, and the recording details and time annotation specifications are also inconsistent.

[0004] The spatio-temporal difference problem of multi-source heterogeneous data is prominent. Due to the different deployment locations and times of data collection devices, it is difficult to directly align and correlate data in time and space. For example, in urban traffic monitoring, the data collection times of cameras at different intersections may deviate, and the covered spatial ranges are also different, making it difficult to analyze the spatio-temporal changes of traffic flow as a whole. Traditional data processing methods are difficult to effectively process this complex multi-source heterogeneous data and cannot fully explore the potential relationships and patterns between data.

[0005] In terms of data fusion, existing technologies mainly focus on simple data splicing or fusion based on a single feature, and cannot comprehensively consider the spatio-temporal characteristics and multi-source heterogeneity of data. In spatio-temporal co-occurrence analysis, there is a lack of efficient methods to identify the co-occurrence patterns of different data sources in time and space, and it is difficult to discover valuable information hidden behind the data. For example, in urban environmental monitoring, it is impossible to quickly and accurately find the spatio-temporal co-occurrence rules between abnormal air quality and specific industrial activities and meteorological conditions.

[0006] In terms of anomaly detection, most traditional methods are based on a single data source or simple threshold judgment, with a high false alarm rate and inability to adapt to complex and changing data environments. For anomaly detection under complex spatio-temporal co-occurrence patterns, existing technologies are even more powerless. For example, in smart grid monitoring, it is difficult to accurately detect the abnormal operating state of the power system by comprehensively analyzing various power equipment data and power consumption event data.

[0007] In terms of optimizing data acquisition parameters, existing technologies usually rely on manual experience or simple rule settings, without making full use of the information in the data for dynamic adjustment. This leads to problems such as resource waste or incomplete data acquisition during the data acquisition process. For example, in the data acquisition of Internet of Things devices, due to unreasonable setting of the acquisition frequency, a large amount of redundant data may be generated, and at the same time, key information cannot be obtained in a timely manner. To sum up, there is an urgent need for a method and system that can effectively process multi-source heterogeneous data, achieve efficient spatio-temporal co-occurrence analysis, accurate anomaly detection, and intelligent optimization of data acquisition parameters. Summary of the Invention

[0008] The purpose of the present invention is to provide a spatio-temporal co-occurrence analysis method and system for multi-source data fusion to solve the problems raised in the above background technology.

[0009] To achieve the above purpose, the present invention provides the following technical solution: A spatio-temporal co-occurrence analysis method for multi-source data fusion, the method includes: Obtain multi-source heterogeneous data of the target area and perform spatio-temporal calibration, where the multi-source heterogeneous data at least includes geospatial data, device perception data, and event record data; Construct a spatio-temporal reference coordinate system and perform spatio-temporal alignment on the multi-source heterogeneous data to obtain a spatio-temporal aligned data sequence, and generate a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal aligned data sequence; Divide the spatio-temporal weight matrix through a spatio-temporal sliding window to form multiple spatio-temporal data sub-blocks, and perform feature fusion on the spatio-temporal data sub-blocks based on the associated feature vector set to obtain a spatio-temporal association feature map; Construct a co-occurrence pattern mining model and train it through the spatio-temporal association feature map to obtain a co-occurrence pattern recognition model, and then extract a spatio-temporal co-occurrence pattern feature set; Construct an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, and verify the anomaly detection model through a sliding verification mechanism to obtain an abnormal co-occurrence pattern set; According to the abnormal co-occurrence pattern set and the data acquisition parameters, construct a parameter optimization model and dynamically adjust the data acquisition parameters to obtain the optimal data acquisition parameters for the target area; Update the spatio-temporal reference coordinate system based on the optimal data acquisition parameters and real-time monitoring data, and generate a spatio-temporal co-occurrence analysis report.

[0010] Preferably, the spatio-temporal weight matrix at least includes a time dimension weight, a space dimension weight, and a data source weight; the associated feature vector set at least includes a spatial proximity vector, a time synchronization vector, and a data correlation vector; the spatio-temporal co-occurrence pattern feature set at least includes a spatial aggregation pattern, a time duration pattern, and a cross-source association pattern.

[0011] Preferably, constructing a spatio-temporal reference coordinate system and performing spatio-temporal alignment on multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence includes the following steps: Define the grid division rule and timestamp alignment rule of the spatio-temporal reference coordinate system according to the geographical boundary and time span of the target area; Perform spatial projection transformation on the multi-source heterogeneous data to make its coordinate parameters consistent with the spatio-temporal reference coordinate system; Normalize the timestamps of the multi-source heterogeneous data to generate time series data with a unified time reference; Based on the spatial proximity threshold and time synchronization threshold, perform data interpolation and conflict resolution on the normalized multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence.

[0012] Preferably, generating a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal alignment data sequence includes the following steps: Calculate the time dimension weight and the spatial dimension weight according to the acquisition frequency, accuracy, and coverage range of the data source; Calculate the data source weight based on the attribute similarity and historical correlation of the data source; Combine the time dimension weight, the spatial dimension weight, and the data source weight into a spatio-temporal weight matrix; Calculate the spatial proximity of the spatio-temporal alignment data sequence to generate a spatial proximity vector; Calculate the time synchronization of the spatio-temporal alignment data sequence to generate a time synchronization vector; Generate a data correlation vector based on the attribute correlation of the data source, and then combine it into an associated feature vector set.

[0013] Preferably, generating the spatial proximity vector includes the following steps: Calculate the spatial neighborhood coverage range of each data point in the spatio-temporal alignment data sequence; Count the number of data points within the preset neighborhood radius for each data point to obtain the spatial density value; Generate a spatial proximity vector based on the spatial density value and the neighborhood overlap degree.

[0014] Preferably, constructing a co-occurrence pattern mining model and training it through a spatio-temporal association feature map to obtain a co-occurrence pattern recognition model includes the following steps: Construct a co-occurrence pattern mining model based on a convolutional neural network; Input the spatio-temporal association feature map into the co-occurrence pattern mining model for feature extraction to obtain potential co-occurrence pattern features; Perform pattern classification on the potential co-occurrence pattern features through an unsupervised clustering algorithm to obtain a spatio-temporal co-occurrence pattern feature set; Feed the classification results back into the co-occurrence pattern mining model for parameter optimization to obtain a co-occurrence pattern recognition model.

[0015] Preferably, the pattern classification of potential co-occurrence pattern features by the unsupervised clustering algorithm includes the following steps: Identify high-density regions in the potential co-occurrence pattern features based on the density clustering algorithm; Merge patterns in the high-density regions to generate a candidate co-occurrence pattern set; Evaluate the clustering effect of the candidate co-occurrence pattern set through the silhouette coefficient, and screen the optimal clustering result as the spatio-temporal co-occurrence pattern feature set.

[0016] Preferably, the construction of the anomaly detection model based on the spatio-temporal co-occurrence pattern feature set includes the following steps: Divide the spatio-temporal co-occurrence pattern feature set into a training set and a validation set; Construct an initial anomaly detection model based on the isolation forest algorithm and train the model with the training set; Use the validation set to perform sliding window verification on the initial anomaly detection model and calculate the anomaly confidence score; Adjust the model threshold parameter according to the anomaly confidence score to obtain an optimized anomaly detection model; Input the real-time monitoring data into the optimized anomaly detection model and output the anomaly co-occurrence pattern set.

[0017] Preferably, the construction of the parameter optimization model and the dynamic adjustment of the data acquisition parameters include the following steps: Define the decision variables of the parameter optimization model, and the decision variables at least include the acquisition frequency, storage capacity, and transmission interval; Construct an optimization objective function based on the historical distribution of the anomaly co-occurrence pattern set, and the objective function at least includes data coverage integrity and resource consumption cost; Set the constraint conditions, and the constraint conditions at least include the upper limit of the device workload and the data transmission delay threshold; Iteratively solve the parameter optimization model through the heuristic search algorithm to obtain the optimal data acquisition parameters.

[0018] Preferably, the present invention further includes a spatio-temporal co-occurrence analysis system for multi-source data fusion, and the shampoo includes: A data acquisition and calibration module, configured to acquire multi-source heterogeneous data of a target area at least including geospatial data, device perception data, and event record data, and perform spatio-temporal calibration; A spatio-temporal alignment processing module, which is used to construct a spatio-temporal reference coordinate system, perform spatio-temporal alignment on multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence, and generate a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal alignment data sequence; A feature fusion map construction module, which divides the spatio-temporal weight matrix through a spatio-temporal sliding window to form multiple spatio-temporal data sub-blocks, and performs feature fusion on the spatio-temporal data sub-blocks based on the associated feature vector set to obtain a spatio-temporal correlation feature map; A co-occurrence pattern mining module, which constructs a co-occurrence pattern mining model and trains it through the spatio-temporal correlation feature map to obtain a co-occurrence pattern recognition model, and then extracts a spatio-temporal co-occurrence pattern feature set; An anomaly detection module, which constructs an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, and verifies the anomaly detection model through a sliding verification mechanism to obtain an abnormal co-occurrence pattern set; A parameter optimization module, which constructs a parameter optimization model according to the abnormal co-occurrence pattern set and data acquisition parameters, and dynamically adjusts the data acquisition parameters to obtain the optimal data acquisition parameters for the target area; A report generation module, which updates the spatio-temporal reference coordinate system and generates a spatio-temporal co-occurrence analysis report based on the optimal data acquisition parameters and real-time monitoring data.

[0019] Compared with the prior art, the beneficial effects of the present invention are: In the data processing link, by obtaining multi-source heterogeneous data including geospatial data, device perception data, and event record data in the target area, and performing spatio-temporal calibration and spatio-temporal alignment, the problems of data heterogeneity and spatio-temporal differences are effectively solved. This processing method can integrate the originally scattered and differently formatted data into a unified spatio-temporal alignment data sequence, laying a solid foundation for subsequent analysis. Compared with traditional data splicing or single feature fusion methods, the present invention fully considers the spatio-temporal characteristics and multi-source heterogeneity of data, greatly improving the usability and analysis value of data.

[0020] In terms of mining the potential value of data, based on the spatio-temporal alignment data sequence, a spatio-temporal weight matrix and an associated feature vector set are generated, and then through a spatio-temporal sliding window and feature fusion, a spatio-temporal correlation feature map is obtained. A co-occurrence pattern mining model is constructed for training and pattern classification, so as to extract spatio-temporal co-occurrence pattern feature sets such as spatial aggregation patterns, time duration patterns, and cross-source association patterns. This enables in-depth mining of the co-occurrence patterns of different data sources in time and space, and discovery of valuable information hidden behind the data. Taking urban environmental monitoring as an example, the co-occurrence rules of air quality anomalies with specific industrial activities and meteorological conditions in time and space can be quickly and accurately found, providing strong data support for environmental governance.

[0021] Anomaly detection is a major highlight of the present invention. An anomaly detection model is constructed based on the spatio-temporal co-occurrence pattern feature set and verified and optimized through a sliding verification mechanism. Compared with traditional anomaly detection methods based on a single data source or simple threshold judgment, this method has a lower false alarm rate and can better adapt to complex and changing data environments. In the monitoring of smart grids, by means of the method of the present invention, various power equipment data and power consumption event data can be comprehensively analyzed to accurately detect the abnormal operating state of the power system, timely discover potential fault risks, and ensure the stable operation of the power system.

[0022] In terms of optimizing data collection parameters, a parameter optimization model is constructed according to the anomaly co-occurrence pattern set and data collection parameters, and parameters such as the collection frequency, storage capacity, and transmission interval are dynamically adjusted. This optimization process makes full use of the information in the data and avoids resource waste and incomplete data collection caused by manual experience or simple rule setting. In the data collection of Internet of Things devices, through the optimization method of the present invention, the collection frequency can be reasonably set, redundant data generation can be reduced, and at the same time, key information can be obtained in a timely manner, improving the efficiency and quality of data collection and realizing the efficient use of resources.

[0023] Finally, based on the optimal data collection parameters and real-time monitoring data, the spatio-temporal reference coordinate system is updated and a spatio-temporal co-occurrence analysis report is generated, providing users with comprehensive and accurate analysis results, helping users deeply understand the situation of the target area, and facilitating scientific and reasonable decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is the working principle diagram of the spatio-temporal co-occurrence analysis method for multi-source data fusion according to the present invention; Figure 2 is the flow chart of the construction and data alignment of the spatio-temporal reference coordinate system; Figure 3 is the flow chart of the construction and training of the co-occurrence pattern mining model; Figure 4 is the flow chart of the construction of the anomaly detection model based on the co-occurrence pattern; Figure 5 is the flow chart of the construction of the data collection parameter optimization model. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] Please refer to Figures 1-5, the present invention provides a technical solution: a spatio-temporal co-occurrence analysis method for multi-source data fusion, and the method includes: Obtain multi-source heterogeneous data of the target area and perform spatio-temporal calibration. The multi-source heterogeneous data includes at least geospatial data, device perception data, and event record data. In actual operation, special data acquisition devices and interfaces are used to obtain data from different data sources. For example, geospatial data is obtained from a geographic information system, device perception data is collected from various sensor devices, and event record data is obtained from an event management system. After obtaining, specific calibration algorithms and technologies are used to calibrate the time and space attributes of the data to ensure the consistency of the data in the spatio-temporal dimension.

[0027] Construct a spatio-temporal reference coordinate system and perform spatio-temporal alignment on the multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence. Based on the spatio-temporal alignment data sequence, generate a spatio-temporal weight matrix and a set of associated feature vectors. According to the actual geographical boundary and time span of the target area, determine the grid division rule and timestamp alignment rule of the spatio-temporal reference coordinate system. Through spatial projection conversion technology, convert the coordinate parameters of the multi-source heterogeneous data into a form consistent with the spatio-temporal reference coordinate system. Normalize the timestamps of the data to form time series data with a unified time reference. Then, according to the set spatial proximity threshold and time synchronization threshold, perform data interpolation and conflict resolution operations on the normalized data to obtain a spatio-temporal alignment data sequence. After that, calculate the time dimension weight and space dimension weight according to factors such as the acquisition frequency, accuracy, and coverage range of the data source; calculate the data source weight based on the attribute similarity and historical correlation of the data source, and combine these weights into a spatio-temporal weight matrix. At the same time, calculate the spatial proximity vector by calculating the spatial proximity of the spatio-temporal alignment data sequence, calculate the time synchronization vector by calculating the time synchronization degree, and generate a data correlation vector based on the attribute correlation of the data source, and then combine them into a set of associated feature vectors.

[0028] Divide the spatio-temporal weight matrix through a spatio-temporal sliding window to form multiple spatio-temporal data sub-blocks, and perform feature fusion on the spatio-temporal data sub-blocks based on the set of associated feature vectors to obtain a spatio-temporal association feature map. Use a preset spatio-temporal sliding window to slide on the spatio-temporal weight matrix at a certain step length to divide it into multiple spatio-temporal data sub-blocks. For each spatio-temporal data sub-block, combine each vector in the set of associated feature vectors and use a specific feature fusion algorithm to fuse features of different dimensions to finally generate a spatio-temporal association feature map.

[0029] Construct a co-occurrence pattern mining model and train it through a spatio-temporal correlation feature map to obtain a co-occurrence pattern recognition model, and then extract the spatio-temporal co-occurrence pattern feature set. Build a co-occurrence pattern mining model based on a convolutional neural network, input the generated spatio-temporal correlation feature map into this model for feature extraction to obtain potential co-occurrence pattern features. Use an unsupervised clustering algorithm to classify the potential co-occurrence pattern features to obtain the spatio-temporal co-occurrence pattern feature set. Feed the classification result back into the co-occurrence pattern mining model to optimize the parameters of the model, thereby obtaining the co-occurrence pattern recognition model.

[0030] Construct an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, and verify the anomaly detection model through a sliding verification mechanism to obtain an abnormal co-occurrence pattern set. Divide the spatio-temporal co-occurrence pattern feature set into a training set and a verification set, construct an initial anomaly detection model based on the Isolation Forest algorithm, and use the training set to train the model. Use the verification set to perform sliding window verification on the initial anomaly detection model and calculate the anomaly confidence score. Adjust the threshold parameter of the model according to the anomaly confidence score to obtain an optimized anomaly detection model. Input the real-time monitoring data into the optimized anomaly detection model to output the abnormal co-occurrence pattern set.

[0031] According to the abnormal co-occurrence pattern set and the data collection parameters, construct a parameter optimization model and dynamically adjust the data collection parameters to obtain the optimal data collection parameters for the target area. Define the decision variables of the parameter optimization model, and the decision variables at least include the collection frequency, storage capacity, and transmission interval. Construct an optimization objective function based on the historical distribution of the abnormal co-occurrence pattern set, and the objective function at least includes data coverage integrity and resource consumption cost. Set constraint conditions, and the constraint conditions at least include the upper limit of device workload and the data transmission delay threshold. Iteratively solve the parameter optimization model through a heuristic search algorithm to obtain the optimal data collection parameters.

[0032] Based on the optimal data collection parameters and the real-time monitoring data, update the spatio-temporal reference coordinate system and generate a spatio-temporal co-occurrence analysis report. Use the obtained optimal data collection parameters and real-time monitoring data to update the spatio-temporal reference coordinate system so that it can better reflect the current data situation. According to the updated coordinate system and relevant data, use a special report generation algorithm and template to generate a spatio-temporal co-occurrence analysis report.

[0033] The present invention will be further described below in conjunction with Embodiments 1 to 5: Embodiment

[0034] After spatio-temporal calibration of multi-source heterogeneous data, in the process of constructing a spatio-temporal reference coordinate system and performing spatio-temporal alignment, first determine the grid division rule and timestamp alignment rule of the spatio-temporal reference coordinate system according to the geographical boundary and time span of the target area. Suppose the target area is an urban area, the geographical boundary covers each administrative division of the city, and the time span is one month. For the grid division rule, the city can be divided into several appropriately sized grids according to the functional areas and geographical features of the city, and each grid represents a spatial unit. The timestamp alignment rule can be set to align in hours, and the timestamps of all data are unified to this time precision.

[0035] Perform spatial projection transformation on multi-source heterogeneous data. If the geospatial data uses a specific projection coordinate system while the spatio-temporal reference coordinate system uses a general geographical coordinate system, professional geospatial information processing software or algorithms are needed to convert the coordinate parameters of the geospatial data into coordinates in the spatio-temporal reference coordinate system. For device perception data and event record data, if they contain spatial information, they are also converted in the same way.

[0036] Normalize the timestamps of multi-source heterogeneous data. If the timestamp format of geospatial data is "year-month-day hour:minute:second", the timestamp format of device perception data is "seconds (counted from a certain starting point in time)", and the timestamp format of event record data is "YYYYMMDDHHMMSS", then a unified time conversion method needs to be designed to convert them all into time series data in hours. For example, extract the hour number from the data in the "year-month-day hour:minute:second" format, convert the "seconds" into the corresponding hour number, and parse the hour number from the data in the "YYYYMMDDHHMMSS" format.

[0037] Based on the spatial proximity threshold and time synchronization threshold, perform data interpolation and conflict resolution on the normalized multi-source heterogeneous data. Suppose the spatial proximity threshold is set to 500 meters and the time synchronization threshold is set to 1 hour. If at a certain time point, there are other data points within 500 meters around a device perception data point, but the time difference exceeds 1 hour, data interpolation needs to be performed according to the situation of the surrounding data points, such as using linear interpolation or other appropriate algorithms to supplement the missing data. If there are data conflicts, such as two data points having different data values at the same spatio-temporal position, conflict resolution needs to be performed according to the credibility of the data or other rules, and a more reliable data value is selected. Through these steps, a spatio-temporally aligned data sequence is finally obtained.

[0038] Taking a city traffic management scenario as an example, in the city traffic management, there are various multi-source heterogeneous data involved, such as geospatial data (city road map data), device perception data (data collected by various sensors set on the roads, including traffic flow sensors, vehicle speed sensors, etc.), and event record data (traffic accident records, road construction records, etc.).

[0039] After obtaining these multi-source heterogeneous data, start the spatio-temporal calibration. For geospatial data, it is obtained from the city geographic information database, and there may be different map projection methods; device perception data is collected in real time from sensors distributed on each road, and its timestamps may be inconsistent due to the clock differences of the sensors themselves; event record data is obtained from the event registration system of the traffic management department, and the time formats and recording standards are also different. Through professional geographic information processing tools and time synchronization algorithms, calibrate the time and space attributes of these data to ensure that the data is initially consistent in the spatio-temporal dimension.

[0040] Construct a spatio-temporal reference coordinate system and perform spatio-temporal alignment. According to the geographical boundaries of the city, covering various urban areas, suburban areas, etc. of the city, as well as the time span, assuming that data within a week is selected for analysis, determine the grid division rules and timestamp alignment rules of the spatio-temporal reference coordinate system. Spatially, divide the city into square grids with a side length of 1 kilometer as spatial units according to the road layout and regional functions of the city; temporally, align with a 15-minute time interval.

[0041] Perform spatial projection conversion on the multi-source heterogeneous data. If the city road map data originally uses the Gauss-Krüger projection, and the set spatio-temporal reference coordinate system uses the Web Mercator projection, it is necessary to use the projection conversion tool in the geographic information system (GIS) to convert the coordinate parameters of the geospatial data from the Gauss-Krüger projection to the Web Mercator projection to ensure consistency with the spatio-temporal reference coordinate system. For the spatial information contained in the device perception data and event record data, it is also processed according to the same projection conversion method.

[0042] Normalize the timestamps of the multi-source heterogeneous data. The timestamps recorded by traffic flow sensors may be relative times starting from the device startup, the timestamps of vehicle speed sensors may be accurate to seconds, the time format of traffic accident records is "year / month / day hour:minute:second", and the time format of road construction records is "YYYY-MM-DD HH:MM". For these timestamps in different formats, develop a special time conversion program to convert them all into unified time series data at 15-minute intervals. For example, convert the relative time to the absolute time corresponding to the system time, and then round it at 15-minute intervals.

[0043] Based on the spatial proximity threshold and the time synchronization threshold, data imputation and conflict resolution are performed on the normalized multi-source heterogeneous data. Suppose the spatial proximity threshold is set to 500 meters and the time synchronization threshold is set to 30 minutes. If within a 15-minute time interval, the data of another vehicle speed sensor within 500 meters of a traffic flow sensor is missing, but the data of other surrounding sensors is normal, the missing vehicle speed data can be supplemented according to the change trend of the data of the surrounding sensors by using linear interpolation or other appropriate algorithms. If the traffic flow data recorded by two sensors at the same spatio-temporal position (the same grid and the same 15-minute interval) varies greatly, by comparing the historical data accuracy, device status and other information of the two sensors, a more reliable data value is selected to complete the conflict resolution operation. Through these steps, a spatio-temporal alignment data sequence suitable for the urban traffic management analysis is finally obtained, providing an accurate data basis for subsequent traffic condition analysis, congestion prediction and other work. Embodiment

[0044] When generating the spatio-temporal weight matrix and the set of associated eigenvectors based on the spatio-temporal alignment data sequence, the time dimension weight and the spatial dimension weight are calculated. For the time dimension weight, the acquisition frequency of the data source is considered. If the acquisition frequency of one sensor is once every 10 minutes and that of another sensor is once per hour, the time dimension weight corresponding to the sensor with a higher acquisition frequency is relatively higher. Because the data with a higher acquisition frequency can reflect the data change situation more timely and is more important in the time dimension. At the same time, the accuracy of the data source is also considered, and the data with higher accuracy will also have a higher weight in the calculation of the time dimension weight. For example, for a sensor measuring temperature, the sensor with an accuracy of 0.1°C will have a higher time dimension weight than the sensor with an accuracy of 1°C. For the spatial dimension weight, it is determined according to the coverage range of the data source. If the coverage range of a geospatial data is the entire city, while the coverage range of another device sensing data is only a small area in the city, then the geospatial data with a larger coverage range will have a higher spatial dimension weight.

[0045] When calculating the data source weight, it is based on the attribute similarity and historical correlation of the data source. Suppose the data sources include weather station data, traffic flow data and environmental monitoring data. If the temperature in the weather station data is similar to the air temperature attribute in the environmental monitoring data, and in the historical data, when the temperature changes, some relevant indicators in the environmental monitoring data also change accordingly, indicating a high historical correlation between these two data sources, then their data source weights will be relatively higher.

[0046] Combine the time - dimension weight, space - dimension weight, and data - source weight into a spatio - temporal weight matrix. According to certain rules, for example, taking the time - dimension weight, space - dimension weight, and data - source weight as different column or row elements of the matrix respectively, a multi - dimensional matrix is formed, and this matrix is the spatio - temporal weight matrix.

[0047] When generating the spatial proximity vector, calculate the spatial neighborhood coverage range of each data point in the spatio - temporal aligned data sequence. A specific radius can be set with each data point as the center, and the area covered by this radius is the spatial neighborhood coverage range of the data point. Count the number of data points within the preset neighborhood radius for each data point to obtain the spatial density value. For example, if the preset neighborhood radius is 300 meters, count the number of data points within this range. The larger the number, the larger the spatial density value. Based on the spatial density value and the neighborhood overlap degree, generate the spatial proximity vector. If the spatial density values of two data points are both large and their neighborhood overlap degree is also high, it indicates that the spatial proximity of these two data points is high, and the corresponding element value in the spatial proximity vector will be large.

[0048] Calculate the time synchronization degree of the spatio - temporal aligned data sequence to generate a time synchronization degree vector. If the timestamps of two data points differ by a small amount within the set time synchronization degree threshold range, then their time synchronization degree is high, and the corresponding element value in the time synchronization degree vector is large. Generate a data correlation degree vector based on the attribute correlation of the data sources. For example, the temperature in the meteorological station data and the air temperature attribute in the environmental monitoring data are related, and the corresponding element values in their data correlation degree vectors are large. Combine the spatial proximity vector, time synchronization degree vector, and data correlation degree vector into an association feature vector set.

[0049] Taking the monitoring scenario of an intelligent agricultural park as an example, the implementation manner of Embodiment 2 is described in detail. In this intelligent agricultural park, there are multiple data sources, including meteorological data provided by meteorological stations (a type of geospatial data), soil data collected by various sensors distributed in the fields (device - sensed data), and agricultural activity records (event - record data).

[0050] When generating the spatio-temporal weight matrix and the associated feature vector set based on the spatio-temporal aligned data sequence, the temporal dimension weight and the spatial dimension weight are calculated first. For the temporal dimension weight, the collection frequency of the weather station is once every 15 minutes, and that of the soil moisture sensor is once per hour. Since the meteorological data has a higher collection frequency and can reflect the impact of meteorological changes on crops more timely, the data of the weather station has a relatively higher weight in the temporal dimension. At the same time, considering the accuracy factor, if the temperature measurement accuracy of the weather station is 0.1 °C and that of the soil temperature sensor is 0.5 °C, the temperature data of the weather station with higher accuracy will further increase the weight in the calculation of the temporal dimension weight. For the spatial dimension weight, the monitoring range of the weather station covers the entire agricultural park, while the soil sensor only monitors the small farmland area where it is located. Obviously, the data of the weather station has a larger coverage range and thus a higher spatial dimension weight.

[0051] When calculating the data source weight, it is based on the attribute similarity and historical correlation of the data sources. There is a certain correlation between the precipitation in meteorological data and the soil moisture data. When the precipitation increases, the soil moisture usually also rises, indicating a relatively high historical correlation between these two data sources. Moreover, there is a connection between the temperature in meteorological data and the crop growth stage in the agricultural activity records. At specific crop growth stages, there are certain requirements for temperature, and the attribute similarity between the two is relatively high. Based on these factors, the meteorological data, the soil moisture data, and the agricultural activity record data will obtain relatively higher weights in the calculation of the data source weight.

[0052] The temporal dimension weight, the spatial dimension weight, and the data source weight are combined into the spatio-temporal weight matrix. For example, the temporal dimension weight is set as the elements of the first row of the matrix, the spatial dimension weight is set as the elements of the second row, and the data source weight is set as the elements of the third row, forming a matrix with 3 rows and multiple columns (the number of columns is determined according to the number of data points). This matrix is the spatio-temporal weight matrix.

[0053] When generating the spatial proximity vector, calculate the spatial neighborhood coverage range of each data point in the spatio-temporal aligned data sequence. Taking the data point of the soil sensor as an example, a circular area with a radius of 100 meters centered on its location is set as the spatial neighborhood coverage range. Count the number of data points within the preset neighborhood radius for each data point to obtain the spatial density value. Suppose there are 3 other soil sensors within the 100-meter neighborhood of a certain soil sensor, and its spatial density value is 3. Based on the spatial density value and the neighborhood overlap degree, generate the spatial proximity vector. If the spatial density values of two soil sensors are both relatively high and their neighborhood overlap degree is also relatively high, it indicates a relatively high spatial proximity between these two sensors, and the corresponding element value in the spatial proximity vector will be larger.

[0054] Calculate the time synchronization degree of the spatio-temporal aligned data sequence to generate a time synchronization degree vector. If the data collected by a weather station and a certain soil sensor at the same moment have exactly the same timestamps and are within the set time synchronization degree threshold range (assuming the threshold is 5 minutes), their time synchronization degree is relatively high, and the corresponding element value in the time synchronization degree vector is relatively large. Generate a data correlation degree vector based on the attribute correlation of data sources. For example, the light intensity in meteorological data is related to the light requirements of crops in agricultural activity records, and the corresponding element value in their data correlation degree vector is relatively large. Finally, combine the spatial proximity vector, time synchronization degree vector, and data correlation degree vector into an association feature vector set to provide rich feature information for subsequent data analysis. Embodiment

[0055] When constructing a co-occurrence pattern mining model and training it through a spatio-temporal association feature map, construct a co-occurrence pattern mining model based on a convolutional neural network. First, select a suitable convolutional neural network architecture. For example, you can choose classic architectures such as LeNet and AlexNet, or improve these architectures according to actual needs. Adjust the parameters of the convolutional neural network according to the characteristics and size of the spatio-temporal association feature map, such as the size and number of convolutional kernels, and the parameters of the pooling layer.

[0056] Input the spatio-temporal association feature map into the co-occurrence pattern mining model for feature extraction. The convolutional layer in the convolutional neural network will perform a convolutional operation on the spatio-temporal association feature map to extract local features. Through the combination of multiple convolutional layers and pooling layers, gradually extract more advanced and abstract potential co-occurrence pattern features. For example, the convolutional layer can extract local aggregation features of data in space, change features in time, etc.

[0057] Classify the potential co-occurrence pattern features through an unsupervised clustering algorithm. Identify high-density regions in the potential co-occurrence pattern features based on the density clustering algorithm. The density clustering algorithm will divide the regions with higher density into different clusters according to the density of data points. In the potential co-occurrence pattern feature dataset, find those regions with a relatively large density of surrounding data points, and these regions may represent different co-occurrence patterns. Merge the patterns in the high-density regions to generate a candidate co-occurrence pattern set. If two high-density regions are similar in features or their distance is relatively close, they can be merged into one candidate co-occurrence pattern. Evaluate the clustering effect of the candidate co-occurrence pattern set through the silhouette coefficient, and screen the optimal clustering result as the spatio-temporal co-occurrence pattern feature set. The silhouette coefficient is an index used to evaluate the clustering effect, which comprehensively considers the intra-class compactness and inter-class separation. Calculate the silhouette coefficient of each candidate co-occurrence pattern set, and select the candidate co-occurrence pattern set with the largest silhouette coefficient as the final spatio-temporal co-occurrence pattern feature set.

[0058] The classification results are fed back into the co-occurrence pattern mining model for parameter optimization to obtain the co-occurrence pattern recognition model. According to the spatio-temporal co-occurrence pattern feature set obtained by classification, analyze the deficiencies in the feature extraction process of the model, such as some features not being fully extracted, or the extracted features having noise problems. Optimize the model by adjusting the parameters of the convolutional neural network, such as the learning rate, weights, etc., so that the model can more accurately identify the co-occurrence pattern, and finally obtain the co-occurrence pattern recognition model.

[0059] Taking a city public safety monitoring scenario as an example, in this scenario, the data sources include urban surveillance cameras (providing video image data, belonging to device perception data), public security incident records (incident record data), and urban geographic information data (geospatial data). After preprocessing such as spatio-temporal calibration and spatio-temporal alignment, a spatio-temporal correlation feature map is obtained, and then it enters the stage of constructing and training the co-occurrence pattern mining model.

[0060] Construct a co-occurrence pattern mining model based on the convolutional neural network. Select the improved AlexNet network architecture because it performs well in image feature extraction and can better handle the spatial and temporal features in the surveillance video image data. According to the size of the spatio-temporal correlation feature map (assuming the map is 224×224 pixels in the spatial dimension and contains data of 10 time steps in the temporal dimension), adjust the parameters of the convolutional neural network. For example, set the convolutional kernel size of the first convolutional layer to 11×11, the stride to 4, and the number of convolutional kernels to 96. Such settings can effectively extract the local features in the map.

[0061] Input the spatio-temporal correlation feature map into the co-occurrence pattern mining model for feature extraction. Assume that the spatio-temporal correlation feature map contains surveillance video image data at different times and locations and the corresponding public security incident record information. The convolutional layer of the convolutional neural network will perform a convolution operation on the map. For example, the first convolutional layer slides the convolutional kernel on the map and performs a weighted sum calculation on the local area of the image. The formula is:

[0062] Among them, represents the convolution output value of the th convolutional kernel at the position in the map; is the weight of the th convolutional kernel at the position; is the pixel value at the position in the map; and are the height and width of the convolutional kernel respectively; is the The bias value corresponding to each convolutional kernel. Through such convolutional operations, potential co-occurrence pattern features such as areas of people gathering and abnormal behavior actions are gradually extracted.

[0063] Perform pattern classification on the potential co-occurrence pattern features through unsupervised clustering algorithms. Identify high-density regions in the potential co-occurrence pattern features based on the density clustering algorithm (DBSCAN). Assume that in the dataset of potential co-occurrence pattern features extracted by the convolutional neural network, each data point represents a feature vector of the monitoring scenario at a specific time and location. The DBSCAN algorithm is based on the set neighborhood radius (assume ) and the minimum number of points (assume ), and divide the density-connected data points into different clusters. For example, in a certain area, when the number of data points within the radius around a data point is greater than or equal to , this area may be identified as a high-density area.

[0064] Merge the patterns in the high-density areas to generate a set of candidate co-occurrence patterns. If two high-density areas are similar in features, such as both containing features of people gathering and abnormal behavior, and the distance between them is relatively close (within a certain distance threshold), they are merged into one candidate co-occurrence pattern.

[0065] Evaluate the clustering effect of the set of candidate co-occurrence patterns through the silhouette coefficient, and select the optimal clustering result as the set of spatio-temporal co-occurrence pattern features. The calculation formula of the silhouette coefficient is:

[0066] Among them, represents the silhouette coefficient of the th data point; is the average distance from the data point to other data points in the same category, measuring the intra-class compactness; is the minimum value of the average distance from the data point to data points in other different categories, measuring the inter-class separation. Calculate the silhouette coefficient of each set of candidate co-occurrence patterns, and select the set of candidate co-occurrence patterns with the largest silhouette coefficient as the final set of spatio-temporal co-occurrence pattern features, because the closer the silhouette coefficient is to 1, the better the clustering effect, that is, the data points within the class are closely aggregated and the separation degree between classes is large.

[0067] Feed the classification results back into the co-occurrence pattern mining model for parameter optimization to obtain a co-occurrence pattern recognition model. Analyze the deficiencies of the model during the feature extraction process based on the spatio-temporal co-occurrence pattern feature set obtained from classification. For example, it is found that some abnormal behavior features are not fully extracted, or there are problems such as noisy features extracted. Optimize the model by adjusting the parameters of the convolutional neural network, such as the learning rate (assuming the initial learning rate is 0.001 and adjusted to 0.0005 according to the feedback), weights, etc., so that the model can more accurately identify co-occurrence patterns, and finally obtain a co-occurrence pattern recognition model for subsequent analysis and prediction of the urban public safety situation.

[0068] Example 4: When constructing an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, divide the spatio-temporal co-occurrence pattern feature set into a training set and a validation set. It can be divided according to a certain ratio, for example, 70% of the data as the training set and 30% of the data as the validation set. Ensure that the data distributions of the training set and the validation set are representative during the division, and can reflect the overall characteristics of the spatio-temporal co-occurrence pattern feature set.

[0069] Construct an initial anomaly detection model based on the Isolation Forest algorithm and train the model through the training set. The Isolation Forest algorithm is a tree-structured anomaly detection algorithm that divides data by constructing multiple isolation trees. During the training process, input the spatio-temporal co-occurrence pattern feature data in the training set into the Isolation Forest model, and the model will learn the distribution pattern of normal data. Each isolation tree will randomly divide the data, and evaluate the anomaly degree of data points by calculating indicators such as the path length of data points to the root node.

[0070] Use the validation set to perform sliding window validation on the initial anomaly detection model and calculate the anomaly confidence score. Set the size and step length of a sliding window, and slide the window on the validation set. For the data within each window, input it into the initial anomaly detection model, and the model will output the anomaly score of each data point. Calculate the anomaly confidence score based on these anomaly scores. For example, statistical methods can be used to calculate the proportion of data points whose anomaly scores exceed a certain threshold within the window, and this proportion is the anomaly confidence score.

[0071] Adjust the model threshold parameter according to the anomaly confidence score to obtain an optimized anomaly detection model. If the anomaly confidence score is too high, it means that the model may misclassify too many normal data as abnormal, and the threshold of the model needs to be appropriately increased; if the anomaly confidence score is too low, it means that the model may not detect enough abnormal data, and the threshold of the model needs to be appropriately decreased. By continuously adjusting the threshold parameter, make the detection effect of the model on the validation set the best, and obtain an optimized anomaly detection model.

[0072] Input the real-time monitoring data into the optimized anomaly detection model, and output the set of anomaly co-occurrence patterns. Process the real-time obtained monitoring data according to the same preprocessing method as the training data, and then input it into the optimized anomaly detection model. The model will analyze these data to determine whether each data point belongs to the anomaly co-occurrence pattern. Collect all the data points determined to be anomaly co-occurrence patterns to form the set of anomaly co-occurrence patterns.

[0073] Taking the goods flow monitoring scenario of a large e-commerce warehouse as an example, in this e-commerce warehouse, a variety of data acquisition devices are deployed to obtain goods flow-related data, and at the same time, data on various warehouse operation events are also recorded. These data are processed to obtain the spatio-temporal co-occurrence pattern feature set, and an anomaly detection model is constructed based on this.

[0074] When dividing the spatio-temporal co-occurrence pattern feature set into a training set and a validation set, assume that the data on the hourly inbound and outbound quantities of goods in the past month, as well as the changes in the storage locations of goods in different areas of the warehouse, etc., are obtained, and the spatio-temporal co-occurrence pattern feature set is generated after preliminary processing. Divide it according to a ratio of 7:3, using the data of the first 21 days as the training set and the data of the last 9 days as the validation set. During the division process, ensure that the training set and the validation set are continuous in chronological order and cover various scenarios of the warehouse's daily operations, such as during promotional activities and daily sales periods, to ensure that the data distribution is representative.

[0075] Construct an initial anomaly detection model based on the Isolation Forest algorithm and train the model with the training set. In this warehouse scenario, the Isolation Forest algorithm will process the spatio-temporal co-occurrence pattern feature data in the training set. Each isolation tree will randomly divide the data. Taking the inbound quantity of goods as an example, assume that the range of the inbound quantity in the training set is 0 - 1000 pieces per hour. The isolation tree will randomly select a split point, such as 500 pieces per hour, divide the data into two parts: the inbound quantity is less than 500 pieces per hour and greater than or equal to 500 pieces per hour, and then continue to perform similar random divisions in the subsets until each subset has only one data point or meets the stopping condition. Evaluate the anomaly degree of each data point by calculating the path length from the data point to the root node. If a data point is quickly isolated during the construction of the tree, that is, its path length is short, then it is considered more likely to be an anomaly point.

[0076] Use the validation set to perform sliding window validation on the initial anomaly detection model and calculate the anomaly confidence score. Set the sliding window size to 3 hours and the step size to 1 hour. On the validation set, starting from the 1st hour of the first day, input the data containing the 1st - 3rd hours into the initial anomaly detection model. The model will output an anomaly score for each data point within these 3 hours (such as the number of goods out of the warehouse per hour, the number of times goods move on a specific shelf, etc.). Assume the anomaly score ranges from 0 - 1, and the closer the score is to 1, the more likely it is an anomaly point. Calculate the proportion of data points within the window whose anomaly scores exceed 0.5 (which can be adjusted according to the actual situation) as the anomaly confidence score for this window. For example, within a certain 3 - hour window, there are 10 data points, and 3 of them have anomaly scores exceeding 0.5, then the anomaly confidence score for this window is 30%.

[0077] Adjust the model threshold parameter according to the anomaly confidence score to obtain an optimized anomaly detection model. If the anomaly confidence score is too high, it means the model may misjudge too many normal data as anomalies. For example, if the anomaly confidence scores of multiple windows are all above 50%, while there should not be such a high proportion of anomalies in the actual warehouse operation, then the threshold of the model needs to be appropriately increased, such as increasing the anomaly score judgment threshold from 0.5 to 0.7. On the contrary, if the anomaly confidence score is too low, it indicates that the model may not detect enough abnormal data, and the threshold needs to be decreased. After multiple adjustments and validations, make the detection effect of the model on the validation set reach the best, and obtain an optimized anomaly detection model.

[0078] Input the real - time monitoring data into the optimized anomaly detection model and output the anomaly co - occurrence pattern set. During the daily operation of the warehouse, obtain the hourly goods flow data in real - time, such as the quantity of goods received in the current hour, the number of times of picking goods from a certain shelf, etc. Process these real - time data according to the same pre - processing method as the training data, and then input them into the optimized anomaly detection model. The model will analyze and judge these data. For example, if the model detects that the number of times of picking goods from a certain shelf is abnormally frequent within a certain hour, and at the same time the inventory quantity of goods in the area where the shelf is located drops sharply in a short period of time, and the co - occurrence pattern of these two situations in space and time is quite different from the normal pattern in the training set, the model will determine it as an anomaly co - occurrence pattern. Collect all the data points determined as anomaly co - occurrence patterns to form an anomaly co - occurrence pattern set. Warehouse managers can timely discover potential problems in warehouse operation, such as goods loss, operation errors, etc., based on this anomaly co - occurrence pattern set and take corresponding measures to handle them.

[0079] Example 5: When constructing a parameter optimization model and dynamically adjusting data acquisition parameters, define the decision variables of the parameter optimization model. The decision variables at least include the acquisition frequency, storage capacity, and transmission interval. For the acquisition frequency, a value range can be set. For example, the minimum acquisition frequency is once per hour, and the maximum acquisition frequency is once per minute. The storage capacity can be determined according to the actual storage capacity of the device and the size of the data volume. For example, the minimum storage capacity of the device is 1GB, and the maximum storage capacity is 10GB. A range can also be set for the transmission interval. For example, the shortest transmission interval is 5 minutes, and the longest transmission interval is 1 hour.

[0080] Construct an optimization objective function based on the historical distribution of the abnormal co-occurrence pattern set. The objective function at least includes data coverage integrity and resource consumption cost. Data coverage integrity can be measured by calculating the coverage ratio of the abnormal co-occurrence pattern in different time periods and spatial regions. If the abnormal co-occurrence pattern can be accurately detected in a certain time period or spatial region, it indicates a high data coverage integrity. The resource consumption cost includes the energy consumption of the device, storage cost, transmission cost, etc. For example, the higher the acquisition frequency, the greater the energy consumption of the device; the larger the storage capacity, the higher the storage cost; the shorter the transmission interval, the higher the transmission cost. Considering these factors comprehensively, construct an objective function that can balance data coverage integrity and resource consumption cost.

[0081] Set constraint conditions, which at least include the upper limit of the device workload and the data transmission delay threshold. The upper limit of the device workload refers to the maximum working pressure that the device can withstand within a certain period of time. If the acquisition frequency is too high, the storage capacity is too large, or the transmission interval is too short, it may cause the device workload to exceed the upper limit and affect the normal operation of the device. The data transmission delay threshold refers to the maximum time delay allowed for data to be transmitted from acquisition to the target location. If the transmission interval is too long, it may cause the data transmission delay to exceed the threshold and affect the real-time nature of the data. Reasonably set these constraint conditions according to the actual performance and application requirements of the device.

[0082] Iteratively solve the parameter optimization model through a heuristic search algorithm to obtain the optimal data acquisition parameters. The heuristic search algorithm can choose genetic algorithms, particle swarm optimization algorithms, etc. Taking the genetic algorithm as an example, first randomly generate a set of initial decision variable values, which represent different combinations of acquisition frequency, storage capacity, and transmission interval, called individuals. Calculate the objective function value of each individual and evaluate their pros and cons. Then, through genetic operations such as selection, crossover, and mutation, generate new individuals. Continuously iterate this process so that the individuals in the population gradually approach the optimal solution. During the iteration process, ensure that each individual meets the set constraint conditions. When certain stop conditions are met, such as the objective function value no longer changes or changes very little, the optimal data acquisition parameters are obtained.

[0083] Taking the cargo transportation management of a large logistics park as an example, in this logistics park, there are a large number of cargo transportation vehicles, warehouses, and various data collection devices. In order to optimize data collection and resource utilization, it is necessary to build a parameter optimization model and dynamically adjust the data collection parameters.

[0084] Define the decision variables of the parameter optimization model. In the context of the logistics park scenario, the collection frequency is related to the frequency of obtaining information such as vehicle location and cargo status. For example, the current collection frequency of vehicle location information is set to be collected once every 5 minutes to every 30 minutes. The storage capacity is related to storing data such as vehicle driving trajectories and cargo transportation records. The minimum storage capacity of the data storage devices in the logistics park is 500GB, and the maximum can be expanded to 5TB. The transmission interval determines the time interval for data to be transmitted from the collection device to the management system, and its value range is set to 10 minutes to 60 minutes. These collection frequencies, storage capacities, and transmission intervals are the decision variables of the parameter optimization model.

[0085] Construct an optimization objective function based on the historical distribution of the abnormal co-occurrence pattern set. In the logistics park, abnormal co-occurrence patterns may include situations where a vehicle stays in a non-designated area for a long time and the cargo status is not updated for a long time, or a large number of vehicles gather at the entrance of a certain warehouse at the same time, causing congestion. In terms of data coverage integrity, it is measured by calculating the proportion of abnormal co-occurrence patterns accurately monitored in different time periods and regions. For example, during the busy transportation period in the logistics park (such as 10 am - 2 pm), if more than 90% of the abnormal co-occurrence patterns can be monitored, it indicates that the data coverage integrity in this period is relatively high. The resource consumption cost includes equipment energy consumption, storage cost, and transmission cost. The higher the collection frequency, the greater the energy consumption of the positioning devices and cargo status monitoring devices on the vehicle; the larger the storage capacity, the higher the purchase and maintenance costs of the storage devices; the shorter the transmission interval, the higher the network traffic cost consumed by data transmission. Incorporating data coverage integrity and resource consumption cost into the objective function, the goal is to minimize the resource consumption cost on the premise of ensuring a certain data coverage integrity. Assume that the data coverage integrity is represented by (the value range , represents complete coverage), and the resource consumption cost is represented by . The objective function can be expressed as , where and are weight coefficients set according to actual needs, used to balance the importance of data coverage integrity and resource consumption cost.

[0086] Set constraint conditions. The upper limit of the device workload is a key constraint condition. Both the data collection devices and transmission devices in the logistics park have certain workload limits. For example, when the positioning device on a vehicle has too high a collection frequency, problems such as overheating and performance degradation may occur. Suppose its maximum tolerable collection frequency is once every 5 minutes, which is the upper limit constraint of the collection frequency. The data transmission delay threshold is also very important. The goods transportation data needs to be transmitted to the management system in a timely manner for scheduling decisions. If the transmission delay is too long, it may lead to incorrect decisions. Set the data transmission delay threshold to 30 minutes, that is, the time from data collection to transmission to the management system cannot exceed 30 minutes, otherwise it is considered to exceed the threshold.

[0087] Iteratively solve the parameter optimization model through a heuristic search algorithm to obtain the optimal data collection parameters. Here, the particle swarm optimization algorithm is used. First, a group of initial decision variable values are randomly generated, that is, different combinations of collection frequencies, storage capacities, and transmission intervals. Each combination is regarded as a particle. For example, the parameter combination of particle 1 is a collection frequency of once every 15 minutes, a storage capacity of 1 TB, and a transmission interval of 30 minutes; the parameter combination of particle 2 is a collection frequency of once every 20 minutes, a storage capacity of 2 TB, and a transmission interval of 40 minutes, etc. Calculate the objective function value of each particle to evaluate its quality. The particle swarm optimization algorithm enables the particles to move in a more optimal direction through information sharing and cooperation among the particles. In each iteration, the particle adjusts its speed and position according to its own historical optimal position and the global optimal position of the group, that is, adjusts the values of the collection frequency, storage capacity, and transmission interval. After multiple iterations, when the objective function value no longer changes significantly or meets specific convergence conditions, the optimal data collection parameters are obtained. For example, the final optimal parameter combination is a collection frequency of once every 20 minutes, a storage capacity of 1.5 TB, and a transmission interval of 40 minutes. The logistics park can adjust the settings of the data collection devices according to these parameters to achieve more efficient data collection and resource utilization, while better monitoring abnormal situations during the goods transportation process and improving the overall operation and management level.

[0088] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device.

[0089] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A spatio-temporal co-occurrence analysis method for multi-source data fusion, characterized in that It includes the following steps: Obtain multi-source heterogeneous data of the target area and perform spatio-temporal calibration, where the multi-source heterogeneous data at least includes geospatial data, device perception data, and event record data; Construct a spatio-temporal reference coordinate system and perform spatio-temporal alignment on the multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence, and generate a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal alignment data sequence; Divide the spatio-temporal weight matrix through a spatio-temporal sliding window to form multiple spatio-temporal data sub-blocks, and perform feature fusion on the spatio-temporal data sub-blocks based on the associated feature vector set to obtain a spatio-temporal association feature map; Construct a co-occurrence pattern mining model and train it through the spatio-temporal association feature map to obtain a co-occurrence pattern recognition model, and then extract a spatio-temporal co-occurrence pattern feature set; Construct an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, and verify the anomaly detection model through a sliding verification mechanism to obtain an anomaly co-occurrence pattern set; Construct a parameter optimization model according to the anomaly co-occurrence pattern set and data acquisition parameters, and dynamically adjust the data acquisition parameters to obtain the optimal data acquisition parameters of the target area; Based on the optimal data acquisition parameters and real-time monitoring data, update the spatio-temporal reference coordinate system and generate a spatio-temporal co-occurrence analysis report.

2. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, wherein The spatio-temporal weight matrix at least includes a time dimension weight, a space dimension weight, and a data source weight; the associated feature vector set at least contains a spatial proximity vector, a time synchronization vector, and a data association vector; the spatio-temporal co-occurrence pattern feature set at least includes a spatial aggregation pattern, a time duration pattern, and a cross-source association pattern.

3. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, characterized in that, The step of constructing a spatio-temporal reference coordinate system and performing spatio-temporal alignment on the multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence includes the following steps: Define the grid division rule and timestamp alignment rule of the spatio-temporal reference coordinate system according to the geographical boundary and time span of the target area; Perform spatial projection transformation on the multi-source heterogeneous data to make its coordinate parameters consistent with the spatio-temporal reference coordinate system; Normalize the timestamps of the multi-source heterogeneous data to generate time series data with a unified time reference; Based on the spatial proximity threshold and time synchronization threshold, perform data interpolation and conflict resolution on the normalized multi-source heterogeneous data to obtain a spatio-temporal alignment data sequence.

4. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, wherein The step of generating a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal alignment data sequence includes the following steps: Calculate the time dimension weight and space dimension weight according to the acquisition frequency, accuracy, and coverage range of the data source; Calculate the data source weight based on the attribute similarity and historical correlation of the data source; Combine the time dimension weight, space dimension weight, and data source weight into a spatio-temporal weight matrix; Calculate the spatial proximity of the spatio-temporal alignment data sequence to generate a spatial proximity vector; Calculate the time synchronization of the spatio-temporal alignment data sequence to generate a time synchronization vector; Generate a data association vector based on the attribute correlation of the data source, and then combine it into an associated feature vector set.

5. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 4, wherein The step of generating the spatial proximity vector includes the following steps: Calculate the spatial neighborhood coverage range of each data point in the spatio-temporal alignment data sequence; Count the number of data points within the preset neighborhood radius for each data point to obtain a spatial density value; Generate a spatial proximity vector based on spatial density values and neighborhood overlap degrees.

6. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, characterized in that The steps of constructing a co-occurrence pattern mining model and training it through a spatio-temporal correlation feature map to obtain a co-occurrence pattern recognition model include the following: Construct a co-occurrence pattern mining model based on a convolutional neural network; Input the spatio-temporal correlation feature map into the co-occurrence pattern mining model for feature extraction to obtain potential co-occurrence pattern features; Perform pattern classification on the potential co-occurrence pattern features through an unsupervised clustering algorithm to obtain a spatio-temporal co-occurrence pattern feature set; Feed the classification result back into the co-occurrence pattern mining model for parameter optimization to obtain a co-occurrence pattern recognition model.

7. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 6, characterized in that, The steps of performing pattern classification on the potential co-occurrence pattern features through an unsupervised clustering algorithm include the following: Identify high-density regions in the potential co-occurrence pattern features based on a density clustering algorithm; Merge patterns in the high-density regions to generate a candidate co-occurrence pattern set; Evaluate the clustering effect of the candidate co-occurrence pattern set through the silhouette coefficient, and select the optimal clustering result as the spatio-temporal co-occurrence pattern feature set.

8. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, wherein, The steps of constructing an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set include the following: Divide the spatio-temporal co-occurrence pattern feature set into a training set and a validation set; Construct an initial anomaly detection model based on the isolation forest algorithm and train the model through the training set; Use the validation set to perform sliding window validation on the initial anomaly detection model and calculate the anomaly confidence score; Adjust the model threshold parameter according to the anomaly confidence score to obtain an optimized anomaly detection model; Input real-time monitoring data into the optimized anomaly detection model and output an anomaly co-occurrence pattern set.

9. The spatio-temporal co-occurrence analysis method for multi-source data fusion according to claim 1, wherein The steps of constructing a parameter optimization model and dynamically adjusting data acquisition parameters include the following: Define the decision variables of the parameter optimization model, where the decision variables at least include the acquisition frequency, storage capacity, and transmission interval; Construct an optimization objective function based on the historical distribution of the anomaly co-occurrence pattern set, where the objective function at least includes data coverage integrity and resource consumption cost; Set constraint conditions, where the constraint conditions at least include the upper limit of device workload and the data transmission delay threshold; Iteratively solve the parameter optimization model through a heuristic search algorithm to obtain the optimal data acquisition parameters.

10. A spatio-temporal co-occurrence analysis system for multi-source data fusion, characterized in that, Include: A data acquisition and calibration module for acquiring multi-source heterogeneous data of a target area at least including geospatial data, device perception data, and event record data, and performing spatio-temporal calibration; A spatio-temporal alignment processing module for constructing a spatio-temporal reference coordinate system, performing spatio-temporal alignment on the multi-source heterogeneous data to obtain a spatio-temporal aligned data sequence, and generating a spatio-temporal weight matrix and an associated feature vector set based on the spatio-temporal aligned data sequence; A feature fusion map construction module that divides the spatio-temporal weight matrix through a spatio-temporal sliding window to form multiple spatio-temporal data sub-blocks, and performs feature fusion on the spatio-temporal data sub-blocks based on the associated feature vector set to obtain a spatio-temporal correlation feature map; A co-occurrence pattern mining module that constructs a co-occurrence pattern mining model and trains it through the spatio-temporal correlation feature map to obtain a co-occurrence pattern recognition model, and further extracts a spatio-temporal co-occurrence pattern feature set; Anomaly detection module, which constructs an anomaly detection model based on the spatio-temporal co-occurrence pattern feature set, and verifies the anomaly detection model through a sliding verification mechanism to obtain an anomaly co-occurrence pattern set; Parameter optimization module, which constructs a parameter optimization model and dynamically adjusts the data acquisition parameters according to the anomaly co-occurrence pattern set and the data acquisition parameters to obtain the optimal data acquisition parameters for the target area; Report generation module, which updates the spatio-temporal reference coordinate system and generates a spatio-temporal co-occurrence analysis report based on the optimal data acquisition parameters and real-time monitoring data.

Images