Intelligent-driven statistical data space-time correlation analysis and trend prediction model

Abstract

The invention discloses a smart-driven statistical data space-time correlation analysis and trend prediction model, which relates to the field of computer data statistics and comprises a data acquisition module, a data preprocessing module, a space-time correlation analysis module, a smart-driven analysis module, a trend prediction module and a result output module. The data acquisition module is used for acquiring statistical data from various data sources, preliminarily storing the acquired data in an original format, classifying and packaging the data according to a preset rule, and sending the data to the data preprocessing module; the intelligent driving analysis module constructs a two-way interaction architecture of a reinforcement learning algorithm and a knowledge graph, and realizes crossing from data surface association to potential causal relationship: the knowledge graph provides domain constraints and prior knowledge, guides the reinforcement learning model to filter invalid actions, and focuses on a high-value analysis direction; the reinforcement learning mines the implicit relationship through iterative training, and feeds the newly found knowledge back in real time to update the knowledge graph.

Description

Technical Field

[0001] This invention relates to the field of computer data statistics, and in particular to a smart-driven model for spatiotemporal correlation analysis and trend prediction of statistical data. Background Technology

[0002] Spatiotemporal correlation analysis and trend prediction of statistical data are core technologies in fields such as intelligent transportation, smart healthcare, and environmental monitoring. In intelligent transportation, it is necessary to predict congestion trends by analyzing the spatiotemporal distribution of road network traffic; in smart healthcare, it is necessary to mine the spatiotemporal clustering characteristics of infectious disease cases to optimize prevention and control strategies; and in environmental monitoring, it is necessary to combine spatiotemporal data to predict pollutant diffusion paths.

[0003] Traditional spatiotemporal correlation analysis primarily relies on static algorithms, such as the global Moran index and DBSCAN clustering, to analyze data using preset parameters. This approach lacks the ability to adaptively adjust to the characteristics of the data distribution. For example, when dealing with complex nonlinear spatiotemporal data (such as the weekend cyclical fluctuations in pedestrian traffic in urban commercial districts), traditional statistical methods are still used, making it difficult to capture high-order correlation patterns. In trend prediction, traditional solutions are often based on a single model (such as the ARIMA time series model), failing to integrate domain knowledge, resulting in insufficient model interpretability and an inability to handle implicit causal relationships in the data (such as the indirect correlation between "sports events" and "surrounding road congestion"). Furthermore, the modules of traditional technologies are relatively independent, lacking a dynamic feedback mechanism across modules. For instance, spatiotemporal correlation analysis and prediction models cannot form an effective linkage, making it difficult to optimize the front-end analysis process based on prediction errors.

[0004] Traditional static algorithms cannot dynamically select analysis strategies based on data characteristics, leading to "overcomputation" in simple data scenarios and "feature omission" in complex scenarios. For example, using deep learning models for regularly distributed traffic flow data increases computational costs but provides limited improvement in accuracy; using the traditional Moran's index for non-linearly distributed medical data may miss local clustering patterns.

[0005] Traditional methods rely on a single, data-driven logic, lacking the guidance of domain knowledge, and struggle to discover multi-level indirect relationships. Furthermore, traditional technologies operate with independent modules, making it impossible to backtrack and optimize the front-end process based on analysis results. For instance, when global autocorrelation analysis fails to identify significant correlations, traditional solutions cannot automatically adjust the spatiotemporal index structure, leading to subsequent analyses continuing to rely on inefficient data organization methods.

[0006] In summary, traditional technologies have significant shortcomings in terms of the dynamic adaptability of spatiotemporal data, the depth of complex relationship mining, model interpretability, and system collaborative optimization, making it difficult to meet the needs of intelligent scenarios for high-precision, interpretable, and real-time analysis. Therefore, an intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data is proposed. Summary of the Invention

[0007] The purpose of this invention is to solve the problems in the prior art by proposing a smart-driven spatiotemporal correlation analysis and trend prediction model for statistical data.

[0008] A smart-driven model for spatiotemporal correlation analysis and trend prediction of statistical data includes a data acquisition module, a data preprocessing module, a spatiotemporal correlation analysis module, a smart-driven analysis module, a trend prediction module, and a result output module.

[0009] The data acquisition module is used to collect statistical data from multiple data sources, initially store the collected data in its original format, classify and package the data according to preset rules, and send it to the data preprocessing module.

[0010] The data preprocessing module is used to receive data sent by the data acquisition module, preprocess the data, and send the data to the spatiotemporal correlation analysis module after processing.

[0011] The spatiotemporal correlation analysis module is used to receive the data processed by the data preprocessing module, organize and manage the data based on spatiotemporal coordinate information using spatiotemporal indexing technology, and mine the correlation between the data in the time and space dimensions through spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis methods, and send the spatiotemporal correlation data obtained from the analysis to the intelligent driving analysis module.

[0012] The intelligent driving analysis module is used to receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module. It has a built-in reinforcement learning algorithm model based on deep learning, and constructs a knowledge graph by combining domain knowledge and expert experience. Through the interaction between the reinforcement learning algorithm and the knowledge graph, it performs in-depth analysis on the spatiotemporal correlation data, mines the potential complex relationships and patterns in the data, generates intelligent analysis results, and sends them to the trend prediction module.

[0013] The trend prediction module is used to receive the intelligent analysis results sent by the intelligent driving analysis module, and use time series analysis algorithm, regression analysis algorithm or machine learning prediction algorithm, combined with spatiotemporal correlation information and intelligent analysis results, to predict the future trend of statistical data, generate prediction results, and send the prediction results to the result output module.

[0014] The result output module is used to receive the prediction results sent by the trend prediction module, display the prediction results in the form of visual charts and reports, and also provides data export function.

[0015] Preferably, the spatiotemporal correlation analysis module is implemented using the following steps:

[0016] S1. Receive the data processed by the data preprocessing module, extract the spatiotemporal coordinate information from the data, and record the statistical characteristics of the data;

[0017] S2. Use spatiotemporal indexing technology to organize and manage data, divide the data into corresponding index nodes according to spatiotemporal coordinates, and establish a mapping relationship between data and index nodes;

[0018] S3. Based on the statistical characteristics of the data recorded in S1, dynamically select a spatiotemporal autocorrelation analysis algorithm to perform spatiotemporal autocorrelation analysis;

[0019] S4. Based on the statistical characteristics of the data, dynamically select a local autocorrelation analysis method to conduct further spatiotemporal autocorrelation analysis;

[0020] S5. Based on the data distribution density and data volume characteristics, dynamically select clustering algorithms for spatiotemporal clustering analysis;

[0021] S6. Integrate the spatiotemporal correlation data obtained from spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis, and package the algorithm types and parameter settings used in the analysis process together, and send them to the intelligent driving analysis module.

[0022] Preferably, in step S1, the spatiotemporal coordinate information includes timestamps and geographic location coordinates, and the statistical characteristics of the data include data type, data volume, and data distribution pattern.

[0023] Preferably, in step S3, dynamically selecting a spatiotemporal autocorrelation analysis algorithm for spatiotemporal autocorrelation analysis includes: if the data has a simple and regular spatiotemporal distribution, the traditional global Moran index calculation method is used to initially determine the degree of autocorrelation of the data within the overall spatiotemporal range; if the data exhibits complex and nonlinear spatiotemporal distribution characteristics, a spatiotemporal graph convolutional network algorithm based on deep learning is used for autocorrelation analysis; when using the global Moran index calculation, if the global Moran index does not reach the set threshold, return to step S2 and readjust the spatiotemporal index structure; if the threshold is reached, proceed to step S4.

[0024] Preferably, the step S4 of dynamically selecting a local autocorrelation analysis method for further spatiotemporal autocorrelation analysis includes: for discrete data, using the local Geary's C coefficient to analyze the autocorrelation characteristics within the local spatiotemporal region; for continuous data, based on the calculation of the local Moran index, combining the kernel density estimation method to more precisely determine the local correlation patterns of the data, such as high-high clustering and low-low clustering.

[0025] Preferably, in step S5, dynamically selecting a clustering algorithm for spatiotemporal clustering analysis includes: if the data distribution is relatively dense and there is noise, an improved DBSCAN algorithm is selected, and its density threshold and neighborhood radius are optimized by a genetic algorithm; if the data presents a hierarchical clustering structure, a hierarchical partitioning-based clustering algorithm is adopted; based on spatiotemporal coordinates, the selected clustering algorithm is used to perform clustering operations on the data to identify clustering regions in the spatiotemporal data;

[0026] When using the improved DBSCAN algorithm for spatiotemporal clustering analysis, the density threshold ∈ and neighborhood radius μ are optimized using a genetic algorithm. The specific optimization process is as follows:

[0027] The density threshold ∈ and the neighborhood radius μ are encoded into a chromosome X = [ò, μ] for the genetic algorithm, and the quality of the chromosome is evaluated by the following fitness function F(X):

[0028]

[0029] Where J(X) is the silhouette coefficient of the clustering results obtained by the DBSCAN algorithm with parameter X, which is used to measure the compactness and separation of the clusters. max N represents the maximum contour coefficient obtained during the historical optimization process. noise The number of noise points in the clustering results is α, and β are weighting coefficients, with α+β=1. The ratio of the two is adjusted according to the characteristics of the actual data to balance the clustering quality and the noise processing effect.

[0030] During the iteration of the genetic algorithm, the chromosome population is updated through selection, crossover, and mutation operations. When the rate of change of the fitness function is less than the set threshold δ for n consecutive generations, the iteration stops. The chromosome parameters at this time are used as the optimal density threshold and neighborhood radius of the improved DBSCAN algorithm for spatiotemporal data clustering.

[0031] Preferably, the intelligent drive analysis module is implemented using the following steps:

[0032] S1. Receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module, parse the data and convert its format to meet the input requirements of the built-in algorithm model;

[0033] S2. Utilize domain knowledge and expert experience to extract key knowledge from relevant documents and databases through knowledge extraction technology, and integrate the extracted knowledge using knowledge fusion technology to construct a knowledge graph;

[0034] S3. Construct a reinforcement learning algorithm model based on deep learning, initialize model parameters, and set a reward mechanism;

[0035] S4. Input the spatiotemporal correlation data into the reinforcement learning algorithm model. The model selects an action based on the current state, obtains relevant knowledge and constraints through interaction with the knowledge graph, and evaluates and adjusts the action.

[0036] S5. Based on the results of the action execution, calculate the reward value. Then, using the reward value and optimization methods such as the policy gradient algorithm, adjust the parameters of the reinforcement learning algorithm model to maximize long-term rewards.

[0037] S6. Repeat steps S4 and S5 to iteratively train the model, conduct in-depth analysis of spatiotemporal correlation data, and uncover potential complex relationships and patterns in the data.

[0038] S7. When the model reaches the set number of training iterations or meets the convergence condition, generate intelligent analysis results and send them to the trend prediction module.

[0039] Compared with existing technologies, the advantages of this invention are:

[0040] 1. In this invention, the spatiotemporal correlation analysis module uses spatiotemporal indexing technology to manage data in a structured manner and combines a dynamic algorithm selection mechanism to achieve multi-dimensional analysis: for simple regular distribution data, the global Moran index is used to quickly locate the overall correlation trend; for complex nonlinear data, a spatiotemporal graph convolutional network is introduced to mine deep patterns; and at the same time, spatiotemporal clustering regions and correlation rules are accurately identified through local autocorrelation analysis and intelligent clustering algorithms.

[0041] 2. The intelligent-driven analysis module in this invention constructs a two-way interactive architecture of "reinforcement learning algorithm + knowledge graph", realizing the leap from surface data correlation to potential causal relationship: the knowledge graph provides domain constraints and prior knowledge, guiding the reinforcement learning model to filter invalid actions and focus on high-value analysis directions; reinforcement learning mines implicit relationships through iterative training and feeds back the newly discovered knowledge to update the knowledge graph in real time.

[0042] 3. In this invention, the spatiotemporal correlation analysis module and the intelligent driving analysis module form a dynamic optimization closed loop through two-way feedback of data and knowledge: the former provides spatiotemporal features to the latter, and the latter triggers the former to adjust the analysis strategy through the knowledge reasoning results; when the global Moran index does not reach the threshold, the system automatically backtracks to optimize the spatiotemporal index structure. This collaborative mechanism significantly enhances the system's fault tolerance for abnormal data and reduces the misjudgment rate. At the same time, through cross-module parameter linkage, it realizes the dynamic matching of analysis strategy and data characteristics, and improves the accuracy and efficiency of the whole process analysis. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the trend prediction structure in this invention.

[0044] Figure 2This is a flowchart of the dynamic algorithm selection process for spatiotemporal correlation analysis in this invention.

[0045] Figure 3 This is a timing diagram of the workflow of the intelligent drive analysis module in this invention.

[0046] Figure 4 This is a simplified training cycle diagram of the application of the intelligent driving analysis module in intelligent medical management in this invention. Detailed Implementation

[0047] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0048] Reference Figure 1-3 As shown, a smart-driven statistical data spatiotemporal correlation analysis and trend prediction model includes a data acquisition module, a data preprocessing module, a spatiotemporal correlation analysis module, a smart-driven analysis module, a trend prediction module, and a result output module.

[0049] The data acquisition module is used to collect statistical data from multiple data sources, including but not limited to historical data and real-time data. The data sources include databases, sensor networks, web page data obtained by web crawlers, etc. The collected data is initially stored in its original format, and then classified and packaged according to preset rules before being sent to the data preprocessing module.

[0050] The data preprocessing module is used to receive data sent by the data acquisition module and preprocess the data. After processing, the data is sent to the spatiotemporal correlation analysis module. The preprocessing includes: first, handling missing values ​​in the data by filling in missing data through interpolation, multiple imputation, and other methods; then, detecting and correcting outliers by using statistical methods or machine learning-based anomaly detection algorithms to identify outlier data and correcting or removing it in a reasonable way; and finally, standardizing the data to convert data of different dimensions to the same scale.

[0051] The spatiotemporal correlation analysis module is used to receive the data processed by the data preprocessing module, organize and manage the data based on spatiotemporal coordinate information using spatiotemporal indexing technology, and mine the correlation between the data in the time and space dimensions through spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis methods, and send the spatiotemporal correlation data obtained from the analysis to the intelligent driving analysis module.

[0052] The intelligent driving analysis module is used to receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module. It has a built-in reinforcement learning algorithm model based on deep learning, and constructs a knowledge graph by combining domain knowledge and expert experience. Through the interaction between the reinforcement learning algorithm and the knowledge graph, it performs in-depth analysis on the spatiotemporal correlation data, mines the potential complex relationships and patterns in the data, generates intelligent analysis results, and sends them to the trend prediction module.

[0053] The trend prediction module is used to receive the intelligent analysis results sent by the intelligent driving analysis module, and use time series analysis algorithm, regression analysis algorithm or machine learning prediction algorithm, combined with spatiotemporal correlation information and intelligent analysis results, to predict the future trend of statistical data, generate prediction results, and send the prediction results to the result output module.

[0054] The result output module is used to receive the prediction results sent by the trend prediction module, and display the prediction results in the form of visual charts and reports. It also provides a data export function to facilitate users to further process and analyze the prediction results.

[0055] The data acquisition module, data preprocessing module, spatiotemporal correlation analysis module, intelligent driving analysis module, trend prediction module, and result output module are connected in sequence, with the data processed by the previous module serving as the input data for the next module.

[0056] The spatiotemporal correlation analysis module is implemented in the following steps:

[0057] S1. Receive the data processed by the data preprocessing module, extract the spatiotemporal coordinate information from the data, and record the statistical characteristics of the data. The spatiotemporal coordinate information includes timestamps and geographic location coordinates. The statistical characteristics of the data include the data type (such as discrete, continuous, time-series, etc.), the data volume, and the data distribution pattern (such as normal distribution, skewed distribution, etc.).

[0058] S2. Use spatiotemporal indexing technology (such as R-tree, quadtree, etc.) to organize and manage the data, divide the data into corresponding index nodes according to spatiotemporal coordinates, and establish a mapping relationship between the data and the index nodes. In this process, if the data volume is too large, the spatiotemporal indexing optimization strategy based on distributed storage will be automatically enabled, such as storing the data in blocks on multiple nodes and building a distributed index structure.

[0059] S3. Based on the statistical characteristics of the data recorded in S1, dynamically select a spatiotemporal autocorrelation analysis algorithm to perform spatiotemporal autocorrelation analysis. If the data has a simple and regular spatiotemporal distribution, use the traditional global Moran's index calculation method to preliminarily determine the degree of autocorrelation of the data within the overall spatiotemporal range. If the data exhibits complex and nonlinear spatiotemporal distribution characteristics, use a spatiotemporal graph convolutional network algorithm based on deep learning for autocorrelation analysis. When using the global Moran's index calculation, if the global Moran's index does not reach the set threshold, return to step S2 and readjust the spatiotemporal index structure. If the threshold is reached, proceed to step S4.

[0060] S4. Based on the statistical characteristics of the data, dynamically select local autocorrelation analysis methods for further spatiotemporal autocorrelation analysis; for discrete data, use local Geary's C coefficient to analyze the autocorrelation characteristics in local spatiotemporal regions; for continuous data, based on the calculation of local Moran's index, combine kernel density estimation methods to more precisely determine local correlation patterns such as high-high clustering and low-low clustering of the data.

[0061] S5. Based on the data distribution density and data volume characteristics, dynamically select clustering algorithms for spatiotemporal clustering analysis;

[0062] If the data distribution is relatively dense and there is noise, the improved DBSCAN algorithm is selected, and its density threshold and neighborhood radius are optimized by genetic algorithm; if the data presents a hierarchical clustering structure, a hierarchical partitioning-based clustering algorithm is adopted; based on spatiotemporal coordinates, the selected clustering algorithm is used to perform clustering operations on the data, such as the BIRCH algorithm, to identify the clustering regions in the spatiotemporal data.

[0063] When using the improved DBSCAN algorithm for spatiotemporal clustering analysis, the density threshold ∈ and neighborhood radius μ are optimized using a genetic algorithm. The specific optimization process is as follows:

[0064] The density threshold ∈ and the neighborhood radius μ are encoded into a chromosome X = [ò, μ] for the genetic algorithm, and the quality of the chromosome is evaluated by the following fitness function F(X):

[0065]

[0066] Where J(X) is the silhouette coefficient of the clustering results obtained by the DBSCAN algorithm with parameter X, which is used to measure the compactness and separation of the clusters. max N represents the maximum contour coefficient obtained during the historical optimization process. noiseThe fitness function represents the number of noise points in the clustering results. α and β are weighting coefficients, and α+β=1. The ratio of the two is adjusted according to the actual data characteristics to balance the clustering quality and noise processing effect. This fitness function aims to reduce noise interference while improving the clustering quality, so that the clustering results can better reflect the true distribution of spatiotemporal data.

[0067] During the iteration of the genetic algorithm, the chromosome population is updated through selection, crossover, and mutation operations. When the rate of change of the fitness function is less than the set threshold δ for n consecutive generations, the iteration stops. The chromosome parameters at this time are used as the optimal density threshold and neighborhood radius of the improved DBSCAN algorithm for spatiotemporal data clustering. Through the above optimization process, compared with the traditional DBSCAN algorithm, this method can improve the clustering accuracy by more than 20% when processing noisy dense spatiotemporal data.

[0068] S6. Integrate the spatiotemporal correlation data obtained from spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis, and package the algorithm types and parameter settings used in the analysis process together, and send them to the intelligent driving analysis module for subsequent traceability and algorithm optimization.

[0069] The intelligent drive analysis module is implemented in the following steps:

[0070] S1. Receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module, parse the data and convert its format to meet the input requirements of the built-in algorithm model;

[0071] S2. Utilize domain knowledge and expert experience to extract key knowledge from relevant documents and databases through knowledge extraction technology, and integrate the extracted knowledge using knowledge fusion technology to construct a knowledge graph;

[0072] S3. Construct a reinforcement learning algorithm model based on deep learning, initialize the model parameters, and set a reward mechanism (such as giving positive rewards for improved prediction accuracy and negative rewards for incorrect predictions).

[0073] S4. Input the spatiotemporal correlation data into the reinforcement learning algorithm model. The model selects an action (such as performing a certain feature extraction or analysis operation on the data) based on the current state (the characteristics of the input data). It obtains relevant knowledge and constraints by interacting with the knowledge graph and evaluates and adjusts the action.

[0074] S5. Based on the results after the action is executed (such as a data relationship or pattern obtained from analysis), calculate the reward value. Then, based on the reward value and optimization methods such as the policy gradient algorithm, adjust the parameters of the reinforcement learning algorithm model to optimize the analysis process with the goal of maximizing long-term rewards.

[0075] S6. Repeat steps S4 and S5 to iteratively train the model, conduct in-depth analysis of spatiotemporal correlation data, and uncover potential complex relationships and patterns in the data.

[0076] S7. When the model reaches the set number of training iterations or meets the convergence condition (such as the reward value change tending to stabilize), generate the intelligent analysis results and send them to the trend prediction module.

[0077] As a preliminary step, the spatiotemporal correlation analysis module mines explicit correlation patterns (such as clustering areas and diffusion trends) in the time and space dimensions of data based on spatiotemporal coordinate information. The spatiotemporal correlation data output is the direct input of the intelligent driving analysis module. After receiving the spatiotemporal correlation data, the intelligent driving analysis module does not stop at the surface correlation relationship, but uses the interaction of reinforcement learning and knowledge graph to mine the complex causal relationships and domain knowledge hidden behind the data, realizing the process from data correlation to knowledge reasoning.

[0078] In smart healthcare management scenarios, taking hospital patient condition monitoring and treatment prediction as an example, the various steps of the smart-driven analysis module are integrated, clearly demonstrating how data processing enables in-depth analysis and effective application of medical data. Figure 4 As shown, in a smart healthcare management scenario, the goal is to achieve accurate prediction of disease progression and optimization of treatment plans by analyzing patients' medical data. The specific steps are as follows:

[0079] Step 101: Data Input and Status Representation

[0080] Spatiotemporal correlation data: Output from the spatiotemporal correlation analysis module, including:

[0081] Spatial characteristics: patient's department (e.g., cardiology, respiratory medicine), ward location, and hospital area distribution. Temporal characteristics: date of visit, length of hospital stay, medication interval, and vital sign monitoring timestamps.

[0082] Association patterns: clustering of infections among patients in the same ward; increase in patients with a certain type of disease within a specific time period.

[0083] State encoding: converting the above features into a vector representation that the model can process.

[0084] state = [Patient ID, Department Code, Length of Hospital Stay, Temperature Sequence Code, Blood Pressure Fluctuation Index, Medication Record Code, Association Pattern Type]

[0085] Step 102: Action Selection and Execution

[0086] Action space: The set of operations that the model can perform, for example:

[0087] action_1: Analyze the correlation between specific drugs and changes in patients' vital signs.

[0088] Action 2: Investigating the transmission route of infection among patients in the same ward.

[0089] Action 3: Explore potential factors that exacerbate the condition of patients with a certain type of disease.

[0090] Policy network decision-making: Based on the current state, the policy network outputs an action probability distribution:

[0091] P(action_1)=0.4, P(action_2)=0.3, P(action_3)=0.3

[0092] ε-greedy selection: Explore randomly with a probability of ε = 0.1, and finally choose action_1.

[0093] Step 103: Knowledge Graph Interaction and Constraint Verification

[0094] Knowledge retrieval: The model queries the knowledge graph for domain knowledge related to action_1.

[0095] Search path: [Drug - Impact on Vital Signs] → [Patient Allergy History Constraints] → [Drug Contraindications]

[0096] Constraints:

[0097] Knowledge graph returns: "Drugs containing a β-lactam structure are contraindicated in patients with penicillin allergy."

[0098] The patient currently has a history of penicillin allergy. If action_1 involves related drugs, the analysis direction needs to be adjusted, and a suitable drug should be selected for correlation analysis.

[0099] Step 104: Action Execution and Result Evaluation

[0100] Execution analysis: Regression analysis was performed on the selected drugs and patient vital signs, and the results showed that:

[0101] After using a certain antihypertensive drug, the patient's blood pressure dropped by an average of 15 mmHg within 2 hours of administration.

[0102] The drug's effectiveness was weakly negatively correlated with the patient's age (correlation coefficient -0.2).

[0103] Reward Calculation:

[0104] Improved prediction accuracy: +6 points (based on clinical data validation)

[0105] Discover new knowledge: +4 points (The knowledge graph does not contain records of the drug's effects on a specific age group)

[0106] Analytical value: +2 points (The results provide guidance for adjusting subsequent medication regimens)

[0107] Total reward points: 12

[0108] Step 105: Parameter Update and Experience Replay

[0109] Gradient calculation: Parameter update direction is calculated based on the policy gradient algorithm.

[0110]

[0111] Where: πθ(a|s) is the policy network, and Q(s,a) is the action value function, which is approximated by the reward value;

[0112] Experience storage: Store the quadruple (s, a, r, s′) into the experience replay buffer:

[0113] (Current state, action_1, 12, next state)

[0114] Batch Update: Sample multiple experiences from the buffer and update the policy network parameters using the Adam optimizer.

[0115] Step 106: Knowledge Graph Update

[0116] New knowledge added: "The difference in the antihypertensive effect of a certain antihypertensive drug on patients of different age groups" has been added as a new fact to the knowledge graph.

[0117] Relationship weight adjustment:

[0118] Weighting of the relationship between "specific drugs" and "changes in vital signs in patients of different ages"

[0119] Update the probability distribution parameters of factors influencing drug efficacy

[0120] Step 107: Termination Condition Determination

[0121] Convergence check:

[0122] The average reward fluctuation over 10 consecutive episodes is <0.5%.

[0123] Policy network parameter update magnitude < 1e-4

[0124] Maximum number of iterations reached: 800 training iterations completed.

[0125] This solution utilizes multi-threaded parallel acquisition technology in the data acquisition module to efficiently acquire historical and real-time data, and dynamically adjusts the acquisition frequency based on the characteristics of the data source. After intelligent missing value imputation (e.g., automatically selecting interpolation or multiple imputation methods based on data type), outlier joint detection using multiple algorithms (combining statistical methods and machine learning algorithms), and standardization processing in the data preprocessing module, the data enters the spatiotemporal correlation analysis module. This module uses spatiotemporal indexing techniques (such as R-trees and quadtrees) to structure the data and designs a dynamic algorithm selection mechanism based on the data distribution characteristics: for simple regular distribution data, a global Moran's index is used for spatiotemporal autocorrelation analysis; for complex nonlinear data, a spatiotemporal graph convolutional network (GCN) is introduced. In the local analysis stage, discrete data uses Geary's C coefficient, and continuous data is optimized by combining kernel density estimation with the Moran's index. Finally, spatiotemporal clustering is achieved through DBSCAN (including genetic algorithm parameter optimization) or the BIRCH algorithm. This process, through the combination of "global-local" and "linear-nonlinear" algorithms, accurately captures the explicit correlation patterns of data in the spatiotemporal dimension.

[0126] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A smart-driven model for spatiotemporal correlation analysis and trend prediction of statistical data, characterized in that: It includes a data acquisition module, a data preprocessing module, a spatiotemporal correlation analysis module, a smart-driven analysis module, a trend prediction module, and a results output module; The data acquisition module is used to collect statistical data from multiple data sources, initially store the collected data in its original format, classify and package the data according to preset rules, and send it to the data preprocessing module. The data preprocessing module is used to receive data sent by the data acquisition module, preprocess the data, and send the data to the spatiotemporal correlation analysis module after processing. The spatiotemporal correlation analysis module is used to receive the data processed by the data preprocessing module, organize and manage the data based on spatiotemporal coordinate information using spatiotemporal indexing technology, and mine the correlation between the data in the time and space dimensions through spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis methods, and send the spatiotemporal correlation data obtained from the analysis to the intelligent driving analysis module. The intelligent driving analysis module is used to receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module. It has a built-in reinforcement learning algorithm model based on deep learning, and constructs a knowledge graph by combining domain knowledge and expert experience. Through the interaction between the reinforcement learning algorithm and the knowledge graph, it performs in-depth analysis on the spatiotemporal correlation data, mines the potential complex relationships and patterns in the data, generates intelligent analysis results, and sends them to the trend prediction module. The trend prediction module is used to receive the intelligent analysis results sent by the intelligent driving analysis module, and use time series analysis algorithm, regression analysis algorithm or machine learning prediction algorithm, combined with spatiotemporal correlation information and intelligent analysis results, to predict the future trend of statistical data, generate prediction results, and send the prediction results to the result output module. The result output module is used to receive the prediction results sent by the trend prediction module, display the prediction results in the form of visual charts and reports, and also provides data export function.

2. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 1, characterized in that: The spatiotemporal correlation analysis module is implemented in the following steps: S1. Receive the data processed by the data preprocessing module, extract the spatiotemporal coordinate information from the data, and record the statistical characteristics of the data; S2. Use spatiotemporal indexing technology to organize and manage data, divide the data into corresponding index nodes according to spatiotemporal coordinates, and establish a mapping relationship between data and index nodes; S3. Based on the statistical characteristics of the data recorded in S1, dynamically select a spatiotemporal autocorrelation analysis algorithm to perform spatiotemporal autocorrelation analysis; S4. Based on the statistical characteristics of the data, dynamically select a local autocorrelation analysis method to conduct further spatiotemporal autocorrelation analysis; S5. Based on the data distribution density and data volume characteristics, dynamically select clustering algorithms for spatiotemporal clustering analysis; S6. Integrate the spatiotemporal correlation data obtained from spatiotemporal autocorrelation analysis and spatiotemporal clustering analysis, and package the algorithm types and parameter settings used in the analysis process together, and send them to the intelligent driving analysis module.

3. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 2, characterized in that: In step S1, the spatiotemporal coordinate information includes timestamps and geographic location coordinates, and the statistical characteristics of the data include the data type, data volume, and data distribution pattern.

4. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 2, characterized in that: In step S3, dynamically selecting a spatiotemporal autocorrelation analysis algorithm for spatiotemporal autocorrelation analysis includes: if the data has a simple and regular spatiotemporal distribution, the traditional global Moran index calculation method is used to initially determine the degree of autocorrelation of the data within the overall spatiotemporal range; if the data exhibits complex and nonlinear spatiotemporal distribution characteristics, a spatiotemporal graph convolutional network algorithm based on deep learning is used for autocorrelation analysis; when using the global Moran index calculation, if the global Moran index does not reach the set threshold, return to step S2 and readjust the spatiotemporal index structure; if the threshold is reached, proceed to step S4.

5. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 2, characterized in that: The step S4, which involves dynamically selecting a local autocorrelation analysis method for further spatiotemporal autocorrelation analysis, includes: for discrete data, using the local Geary's C coefficient to analyze the autocorrelation characteristics within the local spatiotemporal region; for continuous data, based on the calculation of the local Moran index, combining the kernel density estimation method to more precisely determine local correlation patterns such as high-high clustering and low-low clustering of the data.

6. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 2, characterized in that: In step S5, dynamically selecting a clustering algorithm for spatiotemporal clustering analysis includes: if the data distribution is relatively dense and there is noise, an improved DBSCAN algorithm is selected, and its density threshold and neighborhood radius are optimized by a genetic algorithm; if the data presents a hierarchical clustering structure, a hierarchical partitioning-based clustering algorithm is used; based on spatiotemporal coordinates, the selected clustering algorithm is used to perform clustering operations on the data to identify clustering regions in the spatiotemporal data; When using the improved DBSCAN algorithm for spatiotemporal clustering analysis, the density threshold ∈ and neighborhood radius μ are optimized using a genetic algorithm. The specific optimization process is as follows: Encode the density threshold ∈ and the neighborhood radius μ into chromosomes for the genetic algorithm. The quality of chromosomes is evaluated using the following fitness function F(X): Where J(X) is the silhouette coefficient of the clustering results obtained by the DBSCAN algorithm with parameter X, which is used to measure the compactness and separation of the clusters. max N represents the maximum contour coefficient obtained during the historical optimization process. noise The number of noise points in the clustering results is α, and β are weighting coefficients, with α+β=1. The ratio of the two is adjusted according to the characteristics of the actual data to balance the clustering quality and the noise processing effect. During the iteration of the genetic algorithm, the chromosome population is updated through selection, crossover, and mutation operations. When the rate of change of the fitness function is less than the set threshold δ for n consecutive generations, the iteration stops. The chromosome parameters at this time are used as the optimal density threshold and neighborhood radius of the improved DBSCAN algorithm for spatiotemporal data clustering.

7. The intelligent-driven spatiotemporal correlation analysis and trend prediction model for statistical data according to claim 1, characterized in that: The intelligent drive analysis module is implemented in the following steps: S1. Receive spatiotemporal correlation data sent by the spatiotemporal correlation analysis module, parse the data and convert its format to meet the input requirements of the built-in algorithm model; S2. Utilize domain knowledge and expert experience to extract key knowledge from relevant documents and databases through knowledge extraction technology, and integrate the extracted knowledge using knowledge fusion technology to construct a knowledge graph; S3. Construct a reinforcement learning algorithm model based on deep learning, initialize model parameters, and set a reward mechanism; S4. Input the spatiotemporal correlation data into the reinforcement learning algorithm model. The model selects an action based on the current state, obtains relevant knowledge and constraints through interaction with the knowledge graph, and evaluates and adjusts the action. S5. Based on the results of the action execution, calculate the reward value. Then, using the reward value and optimization methods such as the policy gradient algorithm, adjust the parameters of the reinforcement learning algorithm model to maximize long-term rewards. S6. Repeat steps S4 and S5 to iteratively train the model, conduct in-depth analysis of spatiotemporal correlation data, and uncover potential complex relationships and patterns in the data. S7. When the model reaches the set number of training iterations or meets the convergence condition, generate intelligent analysis results and send them to the trend prediction module.

Images