Dynamic multi-source data analysis system for psychological crisis

Abstract

The invention relates to the technical field of data processing, in particular to a psychological crisis dynamic multi-source data analysis system, which comprises a data acquisition module for synchronously acquiring user behavior log data, physiological monitoring data and academic evaluation data; the real-time fusion calculation module performs feature extraction on the three types of data in parallel, an event stream processing engine is adopted to extract behavior mode features, sliding window Fourier transform is adopted to generate time-frequency spectrum features, deviation standardization is adopted to generate numerical feature vectors, cross-modal association weights are calculated through a space-time diagram convolutional network, and a high-risk association mode is output; the early warning generation module generates graded early warning signals, the early warning output module calls a push channel and a first-level early warning trigger short message notification and an emergency intervention interface according to signal levels, and the second-level early warning generates a visual report. Through heterogeneous data parallel processing, space-time diagram convolution fusion and edge calculation deployment, the problem of dynamic processing delay caused by multi-source data fusion missing is solved, and college psychological crisis real-time early warning is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a psychological crisis dynamic multi-source data analysis system. BACKGROUND

[0002] The existing psychological crisis dynamic multi-source data analysis system has the following technical pain points. Specifically, the data processing and analysis module is subject to a computing capacity bottleneck, resulting in a delay in dynamic data flow processing, which hinders real-time early warning signal recognition. For example, in a university scenario, multi-source heterogeneous data such as online behavior logs, physiological sensor data, and academic assessment information lack an efficient fusion computing mechanism, and the system cannot quickly extract correlation patterns and generate an early warning model, causing suspected crisis individual behavior changes to be not analyzed in a timely manner, thereby delaying intervention response decisions. SUMMARY

[0003] In view of the deficiencies of the prior art, the present application provides a psychological crisis dynamic multi-source data analysis system, which solves the technical problem of a delay in dynamic processing caused by the lack of a multi-data fusion computing mechanism, which hinders real-time early warning signal recognition in a university psychological monitoring system.

[0004] To solve the above technical problems, the specific content of the present application is as follows: The psychological crisis dynamic multi-source data analysis system provided by the present application comprises: A data acquisition module is configured to synchronously acquire behavior log data, physiological monitoring data, and academic assessment data of a user. The behavior log data is captured in real time through an online platform interface to obtain a sequence of user operation events. The physiological monitoring data is collected by a wearable device sensor at a preset frequency to obtain biological signals. The academic assessment data is periodically extracted from a structured record in a teaching management database. A real-time fusion computing module is connected to the data acquisition module. The behavior log data is processed by an event stream processing engine to extract behavior pattern features. The physiological monitoring data is processed by a sliding window Fourier transform to generate time-frequency spectrum features. The academic assessment data is processed by a deviation standardization to generate a numerical feature vector. The behavior pattern features, time-frequency spectrum features, and numerical feature vector are aligned by time stamp and input into a spatio-temporal graph convolution network to calculate cross-modal correlation weights. When the cross-modal correlation weight is greater than a dynamic threshold, a high-risk correlation pattern is output and a dynamic assessment model is generated. An early warning generation module is connected to the real-time fusion computing module. The high-risk correlation pattern is input into the dynamic assessment model. The dynamic assessment model updates the classification boundary through an incremental learning mechanism. When the high-risk correlation pattern meets the crisis judgment condition, a graded early warning signal is generated. An early warning output module is connected to the early warning generation module, and calls a push channel according to a level of the hierarchical early warning signal; a first-level early warning triggers an instant short message notification and an emergency intervention interface call; a second-level early warning generates a visual analysis report and sends the report to a management terminal.

[0005] Further, the psychological crisis dynamic multi-source data analysis system comprises a data acquisition module including: A behavior log unit captures a user operation event sequence in real time through an online platform interface and outputs behavior log data; A physiological monitoring unit collects biological signals at a preset frequency through a wearable device sensor and outputs physiological monitoring data; An academic evaluation unit periodically extracts structured records from an educational database and outputs academic evaluation data. Further, the psychological crisis dynamic multi-source data analysis system comprises a real-time fusion calculation module configured to: receive behavior log data output by the data acquisition module, and extract behavior pattern features by using an event stream processing engine; receive physiological monitoring data output by the data acquisition module, and generate time-frequency spectrum features by using a sliding window Fourier transform; receive academic evaluation data output by the data acquisition module, and generate a numerical feature vector by performing a deviation standardization.

[0006] Further, the psychological crisis dynamic multi-source data analysis system comprises a real-time fusion calculation module configured to: align the behavior pattern features, the time-frequency spectrum features and the numerical feature vector according to timestamps, and input the features into a spatio-temporal graph convolution network to calculate cross-modal correlation weights; when the cross-modal correlation weight is greater than a dynamic threshold, output a high-risk correlation mode.

[0007] Further, the psychological crisis dynamic multi-source data analysis system further comprises that when the early warning generation module updates a dynamic evaluation model: receive a newly added high-risk correlation mode output by the real-time fusion calculation module, and trigger an incremental learning to adjust a classification boundary; when a false positive rate fed back by the early warning output module exceeds a set value, retrain the dynamic evaluation model.

[0008] Further, the psychological crisis dynamic multi-source data analysis system comprises an early warning output module configured to: receive a hierarchical early warning signal output by the early warning generation module; when the hierarchical early warning signal is a first level, trigger an instant short message notification interface and an emergency intervention interface call; when the hierarchical early warning signal is a second level, generate a visual analysis report and send the report to a management terminal.

[0009] Further, the psychological crisis dynamic multi-source data analysis system of the present application further comprises: a visualization module connected to the real-time fusion calculation module, receiving the output high-risk association mode; convert the high-risk association mode into a topological atlas, and show the connection path of the high-risk association mode and the user operation event in the behavior log data.

[0010] Further, the psychological crisis dynamic multi-source data analysis system of the present application further comprises: an intervention scheme generation module connected to the early warning output module, receiving the graded early warning signal; According to the level of the graded early warning signal, the pre-stored case library is called, and the intervention instruction including the psychological counselor dispatch instruction and the resource allocation scheme is output.

[0011] Further, the psychological crisis dynamic multi-source data analysis system of the present application, the data acquisition module comprises: The academic evaluation unit additionally acquires the club participation record and outputs to the academic evaluation data; The behavior log unit additionally monitors the library access control card frequency and outputs to the behavior log data. Further, the psychological crisis dynamic multi-source data analysis system of the present application further comprises: The edge computing node receives the behavior log data, physiological monitoring data and academic evaluation data output by the data acquisition module; Perform feature extraction and association analysis to generate high-risk association mode, and only upload high-risk association mode to cloud early warning generation module.

[0012] The present application has the following advantages: The application synchronously acquires behavior log data, physiological monitoring data and academic evaluation data through a parallel acquisition architecture, eliminating the time sequence conflict of the multi-source data input end; the real-time fusion calculation module adopts an event flow processing engine, a sliding window Fourier transform and a deviation standardization to extract parallel features of heterogeneous data, reducing the feature calculation delay; the spatio-temporal graph convolution network aligns the multi-modal features according to the time stamp, calculates the cross-modal correlation weight, replaces the existing multiple iteration calculation with a single graph operation, and solves the processing bottleneck caused by the lack of multi-source data fusion mechanism; the dynamic threshold mechanism adaptively adjusts the correlation determination standard according to the historical false alarm rate, reducing the number of invalid calculations; the edge computing node deployment scheme prepositions the feature extraction and correlation analysis to the data source, compressing the data transmission amount; the incremental learning mechanism only fine-tunes the dynamic evaluation model classification boundary for the newly added high-risk correlation mode, improving the model iteration efficiency; the differential response strategy is triggered by the hierarchical early warning signal, the first-level early warning is directly connected to the short message notification interface and the emergency intervention interface, and the second-level early warning automatically generates a visual report, so that the full-link optimization from data processing to early warning response improves the system recognition speed, reduces the college psychological crisis early warning delay from the existing technology of 15 minutes to 3 minutes, and breaks through the real-time processing bottleneck of dynamic data flow. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. Obviously, for those skilled in the art, other drawings can also be obtained without creative labor based on the drawings.

[0014] Figure 1 The system architecture diagram of the psychological crisis dynamic multi-source data analysis system provided by the embodiment of the present application. DETAILED DESCRIPTION

[0015] In order to make the technical solutions of the present application clearer, the following will combine the embodiments of the present application and the corresponding drawings to clearly and completely describe the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application. The following will combine the drawings to specifically describe the present application provided by the embodiments of the present application. In order to better understand the purpose of the present application, the following will further describe the present application in detail.

[0016] Please refer to Figure 1 The psychological crisis dynamic multi-source data analysis system provided by the present application comprises: a data acquisition module for synchronously acquiring behavior log data, physiological monitoring data, and academic assessment data of a user, the behavior log data being captured in real time through an online platform interface, the physiological monitoring data being collected by a wearable device sensor at a preset frequency, and the academic assessment data being extracted periodically from a teaching database; a real-time fusion calculation module connected to the data acquisition module, the behavior log data being extracted by an event stream processing engine to extract behavior pattern features, the physiological monitoring data being processed by a sliding window Fourier transform to generate time-frequency spectrum features, the academic assessment data being subjected to deviation standardization to generate a numerical feature vector, the behavior pattern features, time-frequency spectrum features, and numerical feature vector being aligned by time stamp and input into a spatio-temporal graph convolution network to calculate cross-modal correlation weights, and when the cross-modal correlation weights are greater than a dynamic threshold, outputting a high-risk correlation mode and generating a dynamic assessment model; a warning generation module connected to the real-time fusion calculation module, the high-risk correlation mode being input into the dynamic assessment model, wherein the dynamic assessment model updates the classification boundary through an incremental learning mechanism, and when the high-risk correlation mode meets a crisis judgment condition, a graded warning signal is generated; a warning output module connected to the warning generation module, a push channel being called according to the level of the graded warning signal, a first-level warning triggering an instant SMS notification and an emergency intervention interface call, and a second-level warning generating a visual analysis report and sending it to a management terminal.

[0017] The data acquisition module first synchronously acquires behavior log data, physiological monitoring data, and academic assessment data of a user. The behavior log data is captured in real time through an online platform interface, the physiological monitoring data is collected by a wearable device sensor at a preset frequency, and the academic assessment data is extracted periodically from a teaching database. The three types of data are accessed through different interface protocols to achieve real-time and completeness of data acquisition.

[0018] The real-time fusion computing module is connected with the data acquisition module, and event stream processing engine is used to extract behavior mode features from the behavior log data, specifically including identifying high-frequency access sensitive word, abnormal login time period and other feature vectors; the physiological monitoring data is converted into a frequency energy distribution matrix in a 20-second time window through a sliding window Fourier transform to generate time-frequency spectrum features; the academic evaluation data is subjected to dispersion standardization to generate a numerical feature vector, for example, the percentage score is mapped to the 0-1 interval and the outliers are removed. The above feature extraction process is performed in parallel to reduce the calculation delay. Then, the behavior mode features, the time-frequency spectrum features and the numerical feature vectors are aligned according to the millisecond-level time stamp, and input into a spatio-temporal graph convolution network to calculate the cross-modal correlation weight, which models the spatio-temporal dependence relationship of the behavior-physiology-academic data through a graph node. When the cross-modal correlation weight is greater than a dynamic threshold, a high-risk correlation mode is output and a dynamic evaluation model is generated, and the dynamic threshold is automatically adjusted according to the historical false positive rate, for example, the initial threshold is 0.7, and when the false positive rate rises, the dynamic threshold is increased to 0.75.

[0019] The early warning generation module is connected with the real-time fusion computing module, and the high-risk correlation mode is input into the dynamic evaluation model. The dynamic evaluation model updates the classification boundary through an incremental learning mechanism, that is, the model parameters are fine-tuned instead of full-retraining each time a new high-risk correlation mode is received, thereby improving the iteration efficiency of the model. When the high-risk correlation mode meets the crisis judgment condition, a graded early warning signal is generated, for example, a first-level signal is triggered when the physiological indicators suddenly change accompanied by a sudden decrease in academic performance, and a second-level signal is triggered when a single indicator is abnormal.

[0020] The early warning output module is connected with the early warning generation module, and a differentiated push channel is called according to the level of the graded early warning signal. The first-level early warning triggers an instant SMS notification and an emergency intervention interface call, the SMS notification includes the crisis user number and the risk type, and at the same time, an emergency contact person is assigned through an API call to a school psychological intervention system; the second-level early warning generates a visual analysis report and sends it to a management terminal, the report integrates a feature correlation graph and a historical trend comparison for review and decision-making by management personnel.

[0021] Specifically, the psychological crisis dynamic multi-source data analysis system provided by the present application comprises: A behavior log unit, which captures a user operation event sequence in real time through an online platform interface and outputs behavior log data; A physiological monitoring unit, which collects biological signals at a preset frequency through a wearable device sensor and outputs physiological monitoring data; An academic evaluation unit, which periodically extracts structured records from an educational administration database and outputs academic evaluation data.

[0022] The data acquisition module includes a behavior log unit, a physiological monitoring unit, and an academic evaluation unit. The behavior log unit captures a user operation event sequence in real time through an online platform interface, specifically including listening to user clicks, browsing, and staying time events on a campus forum and a course selection system, and outputting a timestamp-labeled behavior log data stream. The physiological monitoring unit collects biological signals at a preset frequency through wearable device sensors, such as converting heart rate variability and skin electrical response data into digital signals, and outputs a device number-labeled physiological monitoring data packet. The academic evaluation unit periodically extracts structured records from an academic database, including course grades, attendance times, and assignment submission status fields, and outputs an academic evaluation data table in a standardized format. The three units perform data acquisition in parallel, and align data generation timing through a unified time synchronization service.

[0023] Specifically, the real-time fusion computing module of the psychological crisis dynamic multi-source data analysis system is configured to: receive behavior log data output by the data acquisition module, and extract behavior pattern features using an event stream processing engine; receive physiological monitoring data output by the data acquisition module, and generate time-frequency spectrum features through sliding window Fourier transform; receive academic evaluation data output by the data acquisition module, and generate a numerical feature vector by performing deviation standardization.

[0024] The real-time fusion computing module receives behavior log data output by the data acquisition module, extracts behavior pattern features using an event stream processing engine, specifically including counting sensitive word access frequency and abnormal login period distribution features. After receiving physiological monitoring data, time-frequency spectrum features are generated through sliding window Fourier transform, i.e., biological signals are decomposed into frequency energy matrices within a fixed time window. After receiving academic evaluation data, a numerical feature vector is generated by performing deviation standardization, such as mapping percentage scores to the 0-1 interval and removing outliers. The three types of feature extraction processes are executed in parallel on distributed computing nodes, and a dimension-aligned feature vector set is output.

[0025] Specifically, the real-time fusion computing module of the psychological crisis dynamic multi-source data analysis system is configured to: align the behavior pattern features, time-frequency spectrum features, and numerical feature vectors by timestamp, and input them into a spatio-temporal graph convolution network to calculate cross-modal correlation weights; when the cross-modal correlation weight is greater than a dynamic threshold, output a high-risk correlation mode.

[0026] The real-time fusion calculation module aligns the behavior pattern features, time-frequency spectrum features and numerical feature vectors according to millisecond-level timestamps to form a cross-modal feature matrix. The matrix is input into a space-time graph convolution network to calculate cross-modal correlation weights by modeling the space-time dependence relationship between the behavior-physiology-achievement data through graph nodes. When the cross-modal correlation weight is greater than a dynamic threshold, a high-risk correlation mode is output. The dynamic threshold is dynamically adjusted according to the historical false alarm rate, for example, the initial threshold is 0.7, and when there are three consecutive false alarms, it is automatically increased to 0.75.

[0027] Specifically, the psychological crisis dynamic multi-source data analysis system disclosed by the present application further comprises: receiving the newly added high-risk correlation mode output by the real-time fusion calculation module to trigger incremental learning to adjust the classification boundary of the dynamic evaluation model; When the false alarm rate fed back by the early warning output module exceeds the set value, the dynamic evaluation model is retrained.

[0028] After the early warning generation module receives the newly added high-risk correlation mode output by the real-time fusion calculation module, an incremental learning mechanism is triggered to adjust the classification boundary of the dynamic evaluation model. Specifically, an online support vector machine algorithm is used to fine-tune the position of the decision hyperplane. When the false alarm rate fed back by the early warning output module exceeds the set value, the dynamic evaluation model is retrained, for example, when the false alarm rate is higher than 5% for five consecutive days, the full data training process is started.

[0029] Specifically, the psychological crisis dynamic multi-source data analysis system disclosed by the present application, the early warning output module is configured to: receive the graded early warning signal output by the early warning generation module; When the graded early warning signal is level one, the instant SMS notification interface and the emergency intervention interface are triggered to call; when the graded early warning signal is level two, a visual analysis report is generated and sent to the management terminal.

[0030] The early warning output module receives the graded early warning signal output by the early warning generation module. When the graded early warning signal is level one, the SMS gateway interface sends an encrypted early warning SMS to the counselor terminal, and simultaneously calls the emergency order interface of the school psychological intervention system to generate a work order. When the graded early warning signal is level two, a visual analysis report including a feature correlation graph is generated and automatically sent to the department management terminal through an email protocol.

[0031] Specifically, the psychological crisis dynamic multi-source data analysis system disclosed by the present application further comprises: a visual module connected to the real-time fusion calculation module to receive and output the high-risk correlation mode; convert the high-risk correlation mode into a topological graph to show the connection path of the high-risk correlation mode and the user operation events in the behavior log data.

[0032] The visualization module receives the high-risk association mode output by the real-time fusion computing module, converts it into a topology graph, and the nodes of the graph represent user operation events, physiological indicators, and academic fields, and the edge weights correspond to cross-modal association values. The connection path with a weight exceeding a dynamic threshold is highlighted in red to show the causal relationship chain between the high-risk association mode and the behavior log data.

[0033] Specifically, the psychological crisis dynamic multi-source data analysis system provided by the present application further comprises: An intervention scheme generation module connected to the early warning output module and receiving the graded early warning signal; According to the level of the graded early warning signal, the pre-stored case library is called, and the intervention instruction including the psychologist dispatch instruction and the resource allocation scheme is output.

[0034] The intervention scheme generation module receives the graded early warning signal transmitted by the early warning output module. According to the signal level, the pre-stored case library is called, the suicide tendency intervention case is matched for the first-level signal, and the academic pressure relief case is matched for the second-level signal. The resource scheduling instruction including the psychologist ID allocation, the counseling room reservation time period, and the psychological test scale combination scheme is output.

[0035] Specifically, the psychological crisis dynamic multi-source data analysis system provided by the present application, the data acquisition module comprises: The academic assessment unit additionally acquires club participation records and outputs to academic assessment data; The behavior log unit additionally monitors the library access control card frequency and outputs to the behavior log data. In the university scene adaptation, the academic assessment unit additionally acquires club participation record data, including the number of club activity attendances and role information, and merges them into the extended field of the academic assessment data table. The behavior log unit additionally monitors the library access control card frequency, and when the daily card frequency exceeds three times, it marks the abnormal behavior and writes it into the new dimension of the behavior log data.

[0036] Specifically, the psychological crisis dynamic multi-source data analysis system provided by the present application further comprises that the real-time fusion computing module is deployed on an edge computing node: The edge computing node receives the behavior log data, physiological monitoring data and academic assessment data output by the data acquisition module; Perform feature extraction and association analysis to generate high-risk association modes, and only upload the high-risk association modes to the cloud early warning generation module.

[0037] The real-time fusion computing module deployed at the edge computing node receives the behavior log data, physiological monitoring data and academic evaluation data transmitted locally by the data acquisition module. Feature extraction, cross-modal correlation weight calculation and high-risk correlation pattern generation are performed at the edge node. Only the high-risk correlation pattern is uploaded to the cloud early warning generation module. The edge node clears the original data cache every 24 hours and only retains the feature extraction model parameters.

[0038] The technical solution of the present application solves the problem of dynamic processing delay caused by the lack of multi-source data fusion calculation through multi-dimensional innovative design. The specific technical path is as follows: The data acquisition module adopts a parallel acquisition architecture. The behavior log unit captures user operation event sequences in real time through an online platform interface. The physiological monitoring unit collects biological signals at a preset frequency through wearable device sensors. The academic evaluation unit periodically extracts structured records from the academic database. The three types of data sources are independently collected through separate channels to avoid queue blocking caused by existing serial collection. The waiting delay at the data input end is reduced by 20 milliseconds. The real-time fusion computing module designs a differentiated processing scheme for heterogeneous data characteristics. For behavior log data, an event stream processing engine is used to directly extract behavior pattern features, avoiding raw data buffering. For physiological monitoring data, a sliding window Fourier transform is used to complete time-frequency conversion within the data stream window, eliminating the need for full data resampling. For academic evaluation data, a deviation standardization is performed to generate a dimensionally unified numerical feature vector. The three feature extraction processes are executed in parallel on distributed nodes, with a three-fold increase in processing efficiency compared to existing single-threaded processing.

[0039] The key technical breakthrough lies in the deployment of the spatio-temporal graph convolution network. After aligning the behavior pattern features, time-frequency spectrum features and numerical feature vectors according to the millisecond timestamp, the network models the spatio-temporal dependence of multi-source data and performs a single calculation of cross-modal correlation weight analysis. Compared to the existing serial process of feature concatenation followed by classification, the number of calculation steps is reduced. The dynamic threshold mechanism automatically adjusts the correlation weight determination standard based on historical false alarm rates, avoiding repeated calculations caused by fixed thresholds. The edge computing node deployment scheme front-loads feature extraction and correlation analysis to the data source, and only uploads high-risk correlation patterns to the cloud early warning generation module, reducing data transmission volume to the original data.

[0040] The early warning generation module uses an incremental learning mechanism to update the dynamic evaluation model, fine-tuning the classification boundary only for new high-risk correlation patterns. The model iteration time is reduced from hours to minutes. The differentiated response mechanism is triggered by the graded early warning signals. The first-level early warning directly calls the SMS notification interface and the emergency intervention interface, bypassing the existing manual review process. The second-level early warning automatically generates a visual analysis report and pushes it to the management terminal. The entire process forms a closed-loop optimization from data acquisition to early warning response. Ultimately, the early warning delay of the university mental health monitoring system is reduced from the existing 15 minutes to less than 3 minutes, solving the technical problem of real-time early warning signal recognition being blocked.

Claims

1. A dynamic multi-source data analysis system for psychological crises, characterized in that: include: The data acquisition module is used to synchronously acquire user behavior log data, physiological monitoring data, and academic assessment data. The behavior log data captures user operation event sequences in real time through an online platform interface. The physiological monitoring data collects biological signals at a preset frequency through wearable device sensors. The academic assessment data periodically extracts structured records from the academic affairs database. The real-time fusion computing module is connected to the data acquisition module. It extracts behavioral pattern features from the behavioral log data using an event stream processing engine, generates time-spectrum features from the physiological monitoring data using a sliding window Fourier transform, and performs deviation standardization on the academic assessment data to generate numerical feature vectors. It aligns the behavioral pattern features, time-spectrum features, and numerical feature vectors by timestamps and inputs them into a spatiotemporal graph convolutional network to calculate cross-modal association weights. When the cross-modal association weights are greater than a dynamic threshold, it outputs a high-risk association pattern and generates a dynamic assessment model. The early warning generation module is connected to the real-time fusion computing module and inputs the high-risk association pattern into the dynamic evaluation model. The dynamic evaluation model updates the classification boundary through an incremental learning mechanism and generates a graded early warning signal when the high-risk association pattern meets the crisis judgment conditions. The early warning output module is connected to the early warning generation module. It calls the push channel according to the level of the graded early warning signal. The first-level early warning triggers an instant SMS notification and an emergency intervention interface call. The second-level early warning generates a visual analysis report and sends it to the management terminal.

2. The psychological crisis dynamic multi-source data analysis system according to claim 1, characterized in that, The data acquisition module includes: The behavior log unit captures user operation event sequences in real time through the online platform interface and outputs behavior log data. The physiological monitoring unit collects biological signals at a preset frequency through wearable device sensors and outputs physiological monitoring data. The academic assessment unit periodically extracts structured records from the academic affairs database and outputs academic assessment data.

3. The psychological crisis dynamic multi-source data analysis system according to claim 2, characterized in that, The real-time fusion computing module is configured as follows: The system receives behavioral log data output by the data acquisition module and uses an event stream processing engine to extract behavioral pattern features. The system receives physiological monitoring data output by the data acquisition module and generates time-spectrum features through a sliding window Fourier transform. The system receives academic assessment data output from the data acquisition module and performs deviation standardization to generate numerical feature vectors.

4. The psychological crisis dynamic multi-source data analysis system according to claim 3, characterized in that, The real-time fusion computing module is configured as follows: The behavioral pattern features, temporal spectrum features, and numerical feature vectors are aligned by timestamps and input into a spatiotemporal graph convolutional network to calculate cross-modal association weights. When the cross-modal association weight is greater than the dynamic threshold, a high-risk association mode is output.

5. The psychological crisis dynamic multi-source data analysis system according to claim 4, characterized in that, Also includes: When the early warning generation module updates the dynamic evaluation model: The system receives newly added high-risk association patterns output by the real-time fusion computing module and triggers incremental learning to adjust the classification boundary. When the false alarm rate reported by the early warning output module exceeds the set value, the dynamic evaluation model is retrained.

6. The psychological crisis dynamic multi-source data analysis system according to claim 5, characterized in that, The warning output module is configured as follows: Receive the graded early warning signal output by the early warning generation module; When the level warning signal is Level 1, the instant SMS notification interface and emergency intervention interface are triggered; when the level warning signal is Level 2, a visual analysis report is generated and sent to the management terminal.

7. The psychological crisis dynamic multi-source data analysis system according to claim 6, characterized in that, Also includes: The visualization module is connected to the real-time fusion computing module and receives the output high-risk association patterns. High-risk association patterns are transformed into topology maps, showing the connection paths between high-risk association patterns and user operation events in behavior log data.

8. The psychological crisis dynamic multi-source data analysis system according to claim 7, characterized in that, Also includes: The intervention plan generation module is connected to the early warning output module and receives graded early warning signals. Based on the level of the tiered early warning signal, the pre-stored case library is invoked, and intervention instructions, including psychological counselor scheduling instructions and resource allocation plans, are output.

9. The psychological crisis dynamic multi-source data analysis system according to claim 8, characterized in that, The data acquisition module includes: The academic assessment unit additionally obtains club participation records and outputs them to the academic assessment data; The behavior log unit additionally monitors the frequency of card swiping at the library's access control system and outputs the data to the behavior log.

10. The psychological crisis dynamic multi-source data analysis system according to claim 9, characterized in that, Also includes: The real-time fusion computing module is deployed on edge computing nodes: The edge computing node receives behavioral log data, physiological monitoring data, and academic assessment data output by the data acquisition module. Perform feature extraction and association analysis to generate high-risk association patterns, and only upload the high-risk association patterns to the cloud-based early warning generation module.

Images