Interactive visualization analysis method and system for large-scale basin clean energy historical data based on multi-dimensional correlation

Abstract

This invention discloses an interactive visualization analysis method and system for historical clean energy data in large-scale watersheds based on multi-dimensional correlations. It completes static data aggregation and initial screening by constructing a hexagonal grid, extracts dynamic spatiotemporal features using streaming computing, and utilizes a viewport-aware data distribution mechanism based on feature pyramids. By leveraging the underlying feature data corresponding to the frame buffer image, it reconstructs the dynamic thermal field in real time in response to user interaction feedback and mines multi-dimensional correlations. Furthermore, it performs time-series evolution deduction by combining historical feature data. This not only achieves a leap from static display to dynamic in-depth analysis but also successfully generates future situation prediction results, providing a scientific and intuitive decision-making basis for the planning and scheduling of clean energy in watersheds, and comprehensively improving the visualization interaction experience and analytical efficiency in a big data environment.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to an interactive visualization analysis method and system for historical clean energy data of large-scale river basins based on multi-dimensional correlation. Background Technology

[0002] With the advancement of energy structure transformation, the scale of clean energy facilities such as wind power and photovoltaics in large river basins has expanded rapidly, generating massive amounts of historical operational data. This data is characterized by its multi-source heterogeneity (covering meteorological, equipment status, and geographic information), large spatiotemporal span, and high update frequency. To assist power grid dispatching and operation and maintenance decisions, existing visualization analysis systems typically adopt a "full backend processing + static frontend display" technical solution: all historical data is first cleaned and aggregated on the server side to generate static gridded datasets or image tiles, which are then directly transmitted to the browser for rendering and display. When users perform interactive operations such as zooming or selecting areas, the system often needs to reload the entire image, request new static tiles, or simply overlay simple layers on the front end.

[0003] However, this traditional approach has significant drawbacks when processing large-scale watershed data: due to the lack of dynamic feature extraction mechanisms and viewport-aware incremental distribution strategies, the front end cannot respond in real time to local detail queries at high resolution, resulting in massive data transmission volumes and severe redundancy. Furthermore, existing systems mostly remain at the level of static data display, making it difficult to reconstruct thermal fields in real time during interaction to uncover multidimensional relationships between facilities, and even more difficult to extrapolate future trends based on historical trajectories. This directly leads to problems such as rendering stutters, interaction delays, and limited analytical dimensions when processing massive amounts of data, failing to meet users' needs for real-time, in-depth, interactive analysis of large-scale watershed clean energy data. Summary of the Invention

[0004] This invention aims to provide an interactive visualization analysis method and system for historical clean energy data in large-scale watersheds based on multi-dimensional correlations. It achieves a leap from static display to dynamic in-depth analysis and successfully generates future situation prediction results. It provides a scientific and intuitive decision-making basis for the planning and scheduling of clean energy in watersheds and comprehensively improves the visualization and interactive experience and analysis efficiency in the big data environment.

[0005] To achieve the above objectives, the technical solution adopted by this invention is: an interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation, comprising:

[0006] Based on the geographical boundaries of the watershed, a hexagonal grid is constructed and static aggregation and initial screening of multi-source heterogeneous data are completed to generate a weighted static index matrix.

[0007] The static index matrix is ​​used to perform streaming computation, extract dynamic spatiotemporal features and reduce dimensionality, and generate a multi-level resolution feature pyramid.

[0008] Based on the feature pyramid, viewport-aware data distribution is performed, and incremental data packet streams and front-end rendering instruction sets are output.

[0009] Based on the incremental data packet stream and the front-end rendering instruction set, adaptive texture mapping and block rendering scheduling are performed to generate a frame buffer image;

[0010] In response to user interaction feedback, the dynamic thermal field is reconstructed and multi-dimensional correlations are mined based on the underlying feature data corresponding to the frame buffer image.

[0011] Based on the aforementioned multidimensional relationships and historical feature data, a time-series evolution simulation is performed to generate a future situation prediction result.

[0012] Preferably, the construction of the hexagonal grid and the completion of static aggregation and initial screening of multi-source heterogeneous data include: dividing the watershed geospatial area into hexagonal grids and introducing a terrain correction factor to calculate the effective coverage area; spatially aligning and temporally aligning the multi-source heterogeneous data according to the hexagonal grid to generate a structured dataset; constructing a multi-dimensional feature vector based on the structured dataset and removing outliers; and constructing a static index matrix based on grid identifiers using the cleaned data and preloading it into memory.

[0013] Preferably, performing streaming computation using the static index matrix includes: calculating the local spatiotemporal covariance of the data in the static index matrix based on a sliding time window; performing principal component analysis on the covariance results to generate dimensionality-reduced feature vectors; calculating the spatiotemporal gradient between adjacent grids based on the dimensionality-reduced feature vectors to identify hotspot region boundaries; and aggregating and generating multi-level resolution feature pyramids and detail level indexes based on the hotspot region boundaries.

[0014] Preferably, viewport-aware data distribution based on the feature pyramid includes: filtering a target mesh set from the feature pyramid according to the current browser viewport range and zoom level; comparing the state fingerprint of the target mesh set with the state of the previous frame to calculate the feature increment difference; packetizing and prioritizing the feature increment difference according to network bandwidth prediction; and assembling the packetization results to generate an incremental data packet stream and a front-end rendering instruction set.

[0015] Preferably, the execution of adaptive texture mapping and chunked rendering scheduling includes: dynamically constructing a texture atlas based on the level of detail in the incremental data packet stream and mapping the mesh to texture coordinates; culling invisible geometry according to the viewport projection area and compressing the vertex of the remaining mesh; decomposing the compressed geometry task into chunks according to complexity and allocating them to multi-core threads to achieve load balancing; and having each thread execute rendering in parallel and write the results to the frame buffer to complete the composite output.

[0016] Preferably, the step of responding to user interaction feedback and reconstructing the dynamic thermal field based on the underlying feature data corresponding to the frame buffer image includes: parsing user interaction gestures to obtain the spatial query range and time span; extracting the power sequence of the corresponding grid from the feature pyramid based on the spatial query range, and using the power sequence to construct an energy output correlation matrix; and combining the correlation matrix with the thermal gradient field to track the abnormal propagation path and generate a correlation analysis report.

[0017] Preferably, the construction of the energy output correlation matrix includes: extracting power data of each grid within a specified time span within the query range, calculating the Pearson correlation coefficient between the power sequences of any two grids, generating a correlation matrix that reflects synergistic or complementary relationships, and identifying potential abnormal propagation patterns based on negative correlation regions.

[0018] Preferably, the execution of the time-series evolution simulation includes: forming system states based on historical feature data clustering and statistically analyzing state transition frequencies to construct a state transition probability matrix; starting from the current state, using the state transition probability matrix to drive a Markov chain to simulate multiple future trajectories, calculating the expectation and variance based on the future trajectories, and generating a situation simulation map.

[0019] Preferably, the generation of the situation projection map includes: calculating the statistical variance of the simulated values ​​of each grid in the future time period, marking areas with variance exceeding a threshold as high-risk nodes; determining the warning level by combining the risk probability and the scope of impact, and outputting a projection report that includes interactive parameter adjustment functions.

[0020] On the other hand, this invention proposes an interactive visualization analysis system for historical clean energy data in large-scale watersheds based on multi-dimensional correlations, comprising:

[0021] The static aggregation module is used to construct a hexagonal grid based on the geographical boundary of the watershed and to complete the static aggregation and initial screening of multi-source heterogeneous data, outputting a static index matrix;

[0022] The feature extraction module, connected to the static aggregation module, is used to perform streaming computation on the static index matrix and output a multi-level resolution feature pyramid.

[0023] The data distribution module, connected to the feature extraction module, is used to perform viewport-aware data distribution based on the feature pyramid and output incremental data packet streams and front-end rendering instruction sets.

[0024] The rendering scheduling module, connected to the data distribution module, is used to perform adaptive texture mapping and chunked rendering scheduling according to the incremental data packet stream and output a frame buffer image.

[0025] The correlation mining module, connected to the rendering scheduling module, is used to respond to user interactions, reconstruct the dynamic thermal field based on the underlying feature data, and output multi-dimensional correlation analysis results.

[0026] The situation simulation module connects the correlation mining module and the feature extraction module. It is used to perform time-series evolution simulation based on historical features and correlation relationships, and output the future situation simulation results.

[0027] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0028] This invention effectively solves the problem of low efficiency in preprocessing massive amounts of data by constructing a hexagonal grid for static aggregation and initial screening, and extracting dynamic spatiotemporal features using streaming computing. Based on this, a viewport-aware data distribution mechanism based on feature pyramids is used to accurately generate incremental data packet streams and front-end rendering instruction sets, significantly reducing network bandwidth usage and eliminating rendering overhead from invalid data. This solves the rendering stuttering and response latency problems caused by full data transmission in existing technologies. Utilizing the underlying feature data corresponding to the frame buffer image, the invention reconstructs dynamic thermal fields in real time in response to user interaction feedback and mines multi-dimensional correlations. Combined with historical feature data, it performs temporal evolution deduction, achieving a leap from static display to dynamic in-depth analysis and successfully generating future situation prediction results. This provides a scientific and intuitive decision-making basis for the planning and scheduling of clean energy in watersheds, comprehensively improving the visualization and interactive experience and analytical efficiency in a big data environment. Attached Figure Description

[0029] Figure 1 The flowchart shows the interactive visualization analysis method for historical clean energy data in large watersheds based on multi-dimensional correlation, as described in this invention.

[0030] Figure 2 This is a block diagram of the interactive visualization analysis system for historical clean energy data in large-scale watersheds based on multi-dimensional correlation, as described in this invention. Detailed Implementation

[0031] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0032] like Figure 1 As shown, this invention proposes an interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation, including:

[0033] Based on the geographical boundaries of the watershed, a hexagonal grid is constructed and static aggregation and initial screening of multi-source heterogeneous data are completed to generate a weighted static index matrix.

[0034] The process of constructing a hexagonal grid and completing the static aggregation and initial screening of multi-source heterogeneous data includes: dividing the watershed geospatial area into hexagonal grids and introducing terrain correction factors to calculate the effective coverage area; spatially aligning and temporally aligning the multi-source heterogeneous data according to the hexagonal grid to generate a structured dataset; constructing multi-dimensional feature vectors based on the structured dataset and removing outliers; and constructing a static index matrix based on grid identifiers using the cleaned data and preloading it into memory.

[0035] By introducing a terrain correction factor to calculate the effective coverage area, the spatial representation accuracy of hexagonal grids under complex watershed terrain is significantly improved. The structured dataset generated by spatial grid alignment and time alignment realizes the standardized fusion of multi-source heterogeneous data. By removing outliers based on multi-dimensional feature vectors and constructing a weighted static index matrix preloaded into memory, the accuracy and completeness of data cleaning are greatly improved. Furthermore, the storage structure is optimized from the source, providing an efficient and low-redundancy data foundation for subsequent streaming computing, thereby effectively supporting the rapid retrieval and real-time processing of large-scale data.

[0036] Streaming computation is performed using a static index matrix to extract dynamic spatiotemporal features and reduce dimensionality, generating a multi-level resolution feature pyramid.

[0037] The implementation of streaming computation using a static index matrix includes: calculating the local spatiotemporal covariance of the data in the static index matrix based on a sliding time window; performing principal component analysis on the covariance results to generate dimensionality-reduced feature vectors; calculating the spatiotemporal gradient between adjacent grids based on the dimensionality-reduced feature vectors to identify hotspot region boundaries; and aggregating and generating multi-level resolution feature pyramids and detail level indexes based on the hotspot region boundaries.

[0038] By calculating the local spatiotemporal covariance through a sliding time window and performing principal component analysis, the core dynamic trends of the data are effectively extracted and the feature dimensionality is significantly reduced, greatly reducing computational and storage overhead. Based on the identification of hotspot region boundaries using dimensionality-reduced vectors, the spatiotemporal abrupt change points of energy output within the watershed can be accurately captured. The resulting multi-level resolution feature pyramid and detail level index achieve adaptive data organization from macro-global to micro-local, providing high compression ratio and high information density data support for subsequent viewport-aware distribution, ensuring that the system can maintain smooth rendering performance at different scaling levels.

[0039] Viewport-aware data distribution is performed based on the feature pyramid, outputting incremental data packet streams and front-end rendering instruction sets. Specifically, this includes: selecting target mesh sets from the feature pyramid according to the current browser viewport range and scaling ratio; comparing the state fingerprint of the target mesh set with the state of the previous frame to calculate the feature increment difference; packetizing and prioritizing the feature increment difference based on network bandwidth prediction; and assembling the packetized results to generate incremental data packet streams and front-end rendering instruction sets.

[0040] By dynamically filtering the viewport range and zoom level, the system accurately targets the user's area of ​​interest, completely eliminating the transmission of invalid data. By using state fingerprint comparison to calculate incremental differences in features, it transmits only the changed parts instead of the full amount of data, greatly reducing network bandwidth usage and transmission latency. By combining bandwidth prediction with packet segmentation and priority sorting, it ensures that key visual information is delivered first, thus guaranteeing the smoothness of front-end rendering and real-time interactive experience even in weak network environments, and significantly improving the visualization response speed under massive data.

[0041] Based on the incremental data packet stream and the front-end rendering instruction set, perform adaptive texture mapping and chunked rendering scheduling to generate framebuffer images;

[0042] The adaptive texture mapping and chunked rendering scheduling includes: dynamically constructing a texture atlas based on the level of detail in the incremental data packet stream and mapping the mesh to texture coordinates; culling invisible geometry based on the viewport projection area and compressing the vertex of the remaining mesh; decomposing the compressed geometry task into chunks according to complexity and distributing them to multi-core threads to achieve load balancing; and having each thread execute rendering in parallel and write the results to the frame buffer to complete the composite output.

[0043] By dynamically constructing texture atlases and mesh mappings, an adaptive balance between data precision and rendering efficiency is achieved. Viewport projection culling of invisible geometry and vertex compression techniques significantly reduce memory usage and GPU transfer pressure. Rendering tasks are decomposed by complexity and distributed across multiple core threads for parallel execution, fully utilizing hardware computing power bottlenecks and effectively solving rendering stuttering issues in large-scale scenes. The resulting framebuffer image, while maintaining high-fidelity visual effects, significantly improves system throughput and real-time rendering frame rate, ensuring a smooth experience under complex interactions.

[0044] In response to user interaction feedback, the system reconstructs the dynamic thermal field and mines multi-dimensional correlations based on the underlying feature data corresponding to the frame buffer image. Specifically, this includes: parsing user interaction gestures to obtain the spatial query range and time span; extracting the power sequence of the corresponding grid from the feature pyramid based on the spatial query range, and using the power sequence to construct an energy output correlation matrix; and combining the correlation matrix with the thermal gradient field to track the abnormal propagation path and generate a correlation analysis report.

[0045] The construction of the energy output correlation matrix includes: extracting power data of each grid within a specified time span from the query range, calculating the Pearson correlation coefficient between the power sequences of any two grids, generating a correlation matrix that reflects synergistic or complementary relationships, and identifying potential abnormal propagation patterns based on negative correlation regions.

[0046] By analyzing user interactions to accurately pinpoint the spatiotemporal query range, deep data penetration from macroscopic thermal fields to microscopic power sequences is achieved. The energy output correlation matrix constructed using Pearson correlation coefficients can quantify and reveal the synergistic effects and complementary relationships between grids, effectively identifying potential anomaly propagation paths. Combined with the correlation reports generated from negative correlation region analysis, static data is not only transformed into interpretable dynamic insights, but also significantly improves the accuracy of fault tracing and the timeliness of early warnings, providing strong data support for the intelligent operation and maintenance and decision optimization of the basin's energy system.

[0047] Based on multidimensional correlations and historical feature data, time-series evolution simulations are performed to generate future situation prediction results.

[0048] The execution of the time-series evolution simulation includes: clustering historical feature data to form system states and statistically analyzing state transition frequencies to construct a state transition probability matrix; starting from the current state, using the state transition probability matrix to drive a Markov chain to simulate multiple future trajectories, calculating the expectation and variance based on the future trajectories, and generating a situation simulation map.

[0049] Specifically, generating the situation projection map includes: calculating the statistical variance of the simulated values ​​of each grid in the future time period, marking areas with variance exceeding the threshold as high-risk nodes; determining the warning level by combining the risk probability and the scope of impact, and outputting a projection report that includes interactive parameter adjustment functions.

[0050] By utilizing Markov chain simulation and state transition probability matrices, quantitative prediction and multi-trajectory extrapolation of the system's future state were achieved, effectively capturing potential nonlinear evolution patterns. High-risk nodes were identified by calculating the statistical variance of the simulated values, visualizing uncertainties and risks and transforming them into specific early warning levels, significantly improving the accuracy of fault prediction. The generated interactive extrapolation report supports dynamic parameter adjustment, enabling users to intuitively evaluate the system response under different strategies, thus achieving a leap from passive monitoring to proactive defense and providing a scientific basis for emergency decision-making and optimized scheduling of watershed energy systems.

[0051] On the other hand, this invention proposes an interactive visualization analysis system for historical clean energy data in large-scale watersheds based on multi-dimensional correlations, such as... Figure 2 As shown, it includes:

[0052] The static aggregation module is used to construct a hexagonal grid based on the geographical boundary of the watershed and to complete the static aggregation and initial screening of multi-source heterogeneous data, outputting a static index matrix;

[0053] The feature extraction module, connected to the static aggregation module, is used to perform streaming computation on the static index matrix and output a multi-level resolution feature pyramid.

[0054] The data distribution module, connected to the feature extraction module, is used to perform viewport-aware data distribution based on the feature pyramid and output incremental data packet streams and front-end rendering instruction sets.

[0055] The rendering scheduling module, connected to the data distribution module, is used to perform adaptive texture mapping and chunked rendering scheduling according to the incremental data packet stream and output a frame buffer image.

[0056] The correlation mining module, connected to the rendering scheduling module, is used to respond to user interactions, reconstruct the dynamic thermal field based on the underlying feature data, and output multi-dimensional correlation analysis results.

[0057] The situation simulation module connects the correlation mining module and the feature extraction module. It is used to perform time-series evolution simulation based on historical features and correlation relationships, and output the future situation simulation results.

[0058] Specifically, the modules mentioned above are also used to implement other steps of the interactive visualization analysis method for historical clean energy data in large watersheds based on multi-dimensional correlations, as follows:

[0059] Step 1: Construct a static aggregation and initial screening process for multi-source heterogeneous data based on spatiotemporal grids;

[0060] In response to the challenges posed by the widespread distribution of clean energy facilities, the dense concentration of monitoring points, and the complex data types within large river basins, this solution first performs spatial discretization and temporal slicing of the basic data. This step aims to transform the massive, unstructured raw records into grid cell data with a unified spatiotemporal reference, laying the data foundation for subsequent high-concurrency rendering. The entire process begins with the digital reconstruction of the river basin's geographic information and culminates in the generation of a preliminary spatiotemporal index matrix, ensuring that each data point can be accurately located and assigned a unique identifier. Specifically, this includes:

[0061] Step 1.1: Hexagonal grid partitioning and dynamic weight mapping of watershed geospatial space;

[0062] First, based on the actual geographical boundaries of the watershed, a hexagonal grid (H3) system was used to fully cover the study area. Compared to the traditional rectangular grid, the hexagonal grid maintains area consistency while exhibiting more symmetrical neighborhood relationships, effectively reducing data truncation errors caused by edge effects. During this process, the initial weighting coefficient for each grid cell needs to be calculated based on the density of clean energy facilities in different areas within the watershed.

[0063] Let the total area of ​​the watershed be If the total number of grids is N, then the area of ​​the i-th grid is... It should be approximately equal to However, considering the impact of river course and topographic relief on data collection, a topographic correction factor is introduced. Its value ranges from 0.8 to 1.2, representing the complexity of the area relative to the average terrain. Finally, the effective coverage area of ​​the i-th grid... Given by the formula:

[0064] ;

[0065] in, This is a dimensionless parameter for terrain sensitivity adjustment, with a value of 0.5. It controls the correction magnitude of terrain factors on the effective area of ​​the grid. This formula quantifies the actual data carrying capacity of the grid under different terrain conditions, avoiding weight imbalances between flat and mountainous areas during data processing. The output of this step is a set of hexagonal grids with initial weight attributes, providing a spatial framework for subsequent data placement.

[0066] Step 1.2: Spatiotemporal alignment and standardization normalization of multi-source heterogeneous data;

[0067] Based on the hexagonal mesh skeleton generated in step 1.1, historical data from different monitoring devices and communication protocols are spatially and temporally aligned. Clean energy data within the watershed mainly includes wind power generation, photovoltaic irradiance, water flow, water level changes, and meteorological elements. The temporal resolution of these data often varies, ranging from seconds to minutes or hours. To address this issue, a time window sliding matching strategy is adopted to uniformly map all data to a time window. On a standard time axis with intervals.

[0068] Let the raw observation value of the j-th type of data source at time t be... Its corresponding sampling frequency is To map it to a standard time step. We need to calculate the weighted average value within that time step. Define the time-aligned data values. (Where k is the standard time step index), the calculation formula is as follows:

[0069] ;

[0070] In the formula, This is the time weighting coefficient; when the data sampling frequency is higher than the standard step size, Pick When the step size is below the standard step size, Set the value to 1. This formula ensures that the physical meaning of the data (such as total energy) is not destroyed during time compression, but rather its statistical characteristics are preserved through weighted averaging. After this step, all multi-source data are uniformly converted to have the same time granularity. Structured datasets with the same spatial affiliation (i.e., the hexagonal grid ID to which they belong) eliminate data gaps caused by differences in sampling frequency.

[0071] Step 1.3: Outlier removal and integrity verification based on multidimensional feature vectors;

[0072] Using the structured dataset generated in step 1.2, further data quality cleaning is performed. Due to sensor malfunctions, packet loss, or environmental interference, outliers or missing values ​​are often mixed into the original data. This step constructs a multidimensional feature vector space, representing the data of each grid at each time step as a vector. This vector contains multiple physical quantities such as wind speed, light intensity, power output, and temperature. By calculating the Euclidean distance between vectors and comparing it with a preset threshold, data points that deviate from the normal trajectory are identified.

[0073] Let the eigenvector of the i-th grid at time k be... Its theoretical expected vector is The deviation modulus between the two is :

[0074] ;

[0075] Where M is the number of feature dimensions. For the actual observed value of the m-th dimension feature, This is the historical mean of this dimension. This represents the historical standard deviation for that dimension. The formula is essentially a standardized variant of Mahalanobis distance, used to measure the deviation of the current data point from its historical distribution. If... Exceeding the preset threshold (For example If the value is less than or equal to 3, the data point is considered an outlier and is either removed or marked as missing. After removing outliers, the continuity of the grid data needs to be checked. For grids that are missing for more than L consecutive time steps, their participation in subsequent calculations is temporarily frozen to prevent erroneous data from contaminating the overall trend. This step outputs a cleaned, completed, and highly complete spatiotemporal data matrix, providing a clean data source for subsequent aggregation analysis.

[0076] Step 1.4: Construct a static index matrix based on grid ID and preload cache;

[0077] Finally, the cleaned data processed in step 1.3 is used to construct a static index matrix based on the grid ID and time step index, and then written into the memory preload area. This index matrix not only contains the data itself but also records the physical address offset of each data point in the storage space, enabling rapid location in subsequent steps. Let the index matrix be... Its dimensions are (N, K), where N is the total number of grid cells and K is the total number of time steps. The elements in the matrix... The system stores the data pointer and metadata (such as data confidence and source type) of the i-th grid at time k. To support subsequent high-frequency access, a block storage strategy is adopted, dividing the matrix into blocks of size B×B, with each block stored contiguously in memory. The block indices of the i-th grid and the k-th time step are calculated. The formula is:

[0078] ;

[0079] This step establishes efficient addressing logic, enabling the system to locate the memory block simply by performing a single division and rounding operation when reading data from any grid at any time, significantly reducing the overhead of random reads and writes. This completes the transformation from raw, messy data to a structured, indexed, and quickly accessible static data matrix, laying a solid foundation for subsequent dynamic aggregation and feature extraction.

[0080] Step 2: Implement dynamic spatiotemporal feature extraction and dimensionality reduction based on streaming computing;

[0081] After establishing a high-quality static data foundation in Step One, this step focuses on extracting representative spatiotemporal features from massive datasets in real time and reducing the dimensionality of high-dimensional data to a low-dimensional space suitable for front-end rendering. With increasing browser-side interaction demands, directly transmitting raw high-dimensional data can lead to network bandwidth bottlenecks; therefore, deep data extraction and compression must be performed before front-end requests are triggered. The core of this step lies in utilizing streaming computing to perform rolling window analysis on the static index matrix generated in Step One, extracting key feature vectors that reflect the energy distribution trends in the watershed, specifically including:

[0082] Step 2.1: Calculation of local spatiotemporal correlation based on sliding window;

[0083] First, for the static index matrix constructed in step one, a sliding time window of length W is set, and local statistical analysis is performed on each grid cell. The aim is to capture the correlation between short-term fluctuations and long-term trends in the data. Let the window start time be... The end time is For the i-th grid, calculate its time series covariance matrix within the window. This matrix describes the degree of linear correlation between different physical quantities (such as wind speed and power) within the grid. Elements of the covariance matrix. Defined as:

[0084] ;

[0085] Where u and v represent different dimensions of physical quantities. It is the u-th dimension data of the i-th grid at time t. This is the mean of that dimension within the window. By calculating this matrix, the coupling relationship between different energy types in the same area can be quantified, such as whether photovoltaic output is affected by cloud cover during strong winds.

[0086] Step 2.2: Principal Component Analysis-Driven Adaptive Feature Dimensionality Reduction and Projection;

[0087] Based on the covariance matrix calculated in step 2.1, principal component analysis (PCA) is performed on the local data features of each grid to achieve data dimensionality reduction. Since the original data has high dimensionality and multicollinearity, direct transmission is inefficient. This step extracts the principal components that can explain the maximum variance by solving for the eigenvalues ​​and eigenvectors of the covariance matrix. Let the covariance matrix of the i-th grid be... eigenvalues The corresponding feature vector is The first R principal components are selected to form the projection matrix. Where R is the optimal number of dimensions determined based on the cumulative contribution rate. Cumulative contribution rate The calculation formula is:

[0088] ;

[0089] Set threshold (e.g., 0.95), when Stop increasing the number of principal components at that point. Then, convert the original high-dimensional data vector... Projecting onto a lower-dimensional space yields the dimensionality-reduced feature vectors. :

[0090] ;

[0091] in, This is the mean vector of the grid. This process not only significantly reduces the amount of data but also retains the most crucial variation information in the data, allowing subsequent heatmap rendering to focus on the primary issues and ignore secondary noise. The dimensionality-reduced feature vectors will serve as the basic input for subsequent heatmap generation, significantly reducing the computational burden of data transmission and rendering.

[0092] Step 2.3: Hotspot region identification and boundary delineation based on spatiotemporal gradient;

[0093] Using the low-dimensional feature vectors generated in step 2.2, energy hotspot regions and their boundaries within the watershed are further identified. Hotspot regions are defined as sets of grids with drastic changes in feature values ​​or significantly higher values ​​than the background level. The degree of spatial variation can be quantified by calculating the gradient of the feature vectors between adjacent grids. Let the feature difference between the i-th grid and its neighbor j be . Its gradient magnitude The calculation is as follows:

[0094] ;

[0095] in, Denotes the Euclidean norm. The geographical distance (in kilometers) between the center points of the two grids. This formula combines feature differences with geographical distance, avoiding misjudgments caused by simple numerical differences and ensuring that gradient calculation conforms to geospatial logic. Set the gradient threshold. ,like If i and j are considered to have a clear regional boundary, then by traversing all adjacent grid pairs, the entire watershed can be divided into several relatively uniform "feature clusters," with smooth data within each cluster and significant differences between clusters. These feature clusters are potential hotspot regions, and their boundary information will be passed to the next step to guide the allocation of rendering accuracy for the heatmap.

[0096] Step 2.4: Generate a multi-level resolution feature pyramid and LOD level index;

[0097] Finally, based on the hotspot regions and feature clusters identified in step 2.3, a multi-resolution feature pyramid is constructed, and a corresponding Level of Detail (LOD) index is established. To accommodate rendering requirements at different scaling levels, the feature data is divided into L levels according to resolution from high to low. The highest level (Level 0) retains the finest mesh data, while the lowest level (Level L-1) retains only the average features of the region. For each resolution level, the mesh center coordinates and feature values ​​of that level are recalculated. Let the mesh side length of the L-th resolution be... Then the eigenvalues ​​of this level of grid It is obtained by weighted averaging of the eigenvalues ​​of its subordinate subgrids:

[0098] ;

[0099] Where Children(i) represents the set of all child grids belonging to the i-th parent grid. The area of ​​the sub-mesh is represented by this recursive aggregation formula. This formula ensures that the acquired feature data maintains statistical accuracy regardless of how the user zooms in on the view, preventing the loss of crucial details due to over-aggregation. The generated feature pyramid and LOD index enable the system to dynamically select the appropriate data level for transmission and rendering based on the user's viewport size, completely resolving the stuttering issue during high-resolution interactions.

[0100] Step 3: Implement a viewport-aware dynamic data distribution and incremental update strategy;

[0101] Following the multi-level feature pyramid and LOD index generated in step two, this step aims to address the data distribution efficiency issue in high-concurrency browser scenarios. Traditional static data loading methods cannot handle the fluctuations in data requests caused by frequent dragging and zooming operations by users. Therefore, a viewport-aware mechanism is needed to achieve on-demand data distribution and incremental updates. The core logic of this step is to accurately extract the required data from the pyramid structure in step two based on the current browser viewport range and zoom level, and calculate the incremental differences by combining historical states, transmitting only the changed parts to minimize network load and rendering latency. Specifically, this includes:

[0102] Step 3.1: Real-time capture and mesh filtering of viewport geometry;

[0103] First, capture the current viewport geometry of the browser in real time, including the coordinates of the top-left corner. lower right corner coordinates And the current zoom level. Based on the LOD index built in step two, quickly filter out the set of grids that fall within the viewport area. Let the projection range of the viewport in the map coordinate system be... The filtering criterion is the grid center point. satisfy:

[0104] ;

[0105] At the same time, Zoom determines the required LOD level based on the scaling ratio. If the zoom level is greater than the threshold If the resolution is between [the required resolution], then the highest resolution level is selected; if it is between [the required resolution], then [the required resolution] is selected. and If it is between, then select the middle level; if it is less than If the lowest level is selected, the selection process utilizes the index structure established in step two. Through binary search or hash mapping, the target grid list can be located within milliseconds. This step ensures that only data within the user's truly visible area is included in subsequent processing, avoiding the transmission of invalid data.

[0106] Step 3.2: Calculate incremental data differences based on state fingerprints;

[0107] After determining the target mesh list, it's necessary to determine whether this data has changed relative to the previous rendering state. To avoid repeatedly transmitting unchanged data, this step employs state fingerprinting technology. A unique state fingerprint is generated for each mesh at a specific LOD level. The fingerprint consists of the hash value of the current feature vector of the grid and a timestamp. Let the state fingerprint of grid i at the time of the previous frame rendering be... The fingerprint currently calculated is Then the difference flag bit Defined as:

[0108] ;

[0109] like This indicates that the grid data has not undergone substantial changes and does not require retransmission; if If a grid is marked as needing updating, then it is further calculated that the change in its eigenvector is required. This incremental data is much smaller than the full dataset, which significantly reduces packet size, especially in scenarios with long-term operation and slow environmental changes. This step strictly relies on the LOD-level feature data provided in step two, ensuring the accuracy and consistency of the difference calculation.

[0110] Step 3.3: Dynamic packet splitting and priority ranking based on network bandwidth prediction;

[0111] Based on the amount of data to be updated selected in step 3.2, combined with the current network bandwidth prediction value... A dynamic packet splitting strategy should be developed. Network bandwidth predictions can be estimated using the exponential moving average method of historical transmission rates, as shown in the following formula:

[0112]

[0113] Where Rate(t) is the actual transmission rate at the current moment. The smoothing coefficient is set to 0.3. Based on the predicted bandwidth, the data to be updated is divided into several data packets, each with a size not exceeding [a certain value]. At the same time, priority is assigned based on the importance of the grid. Importance score. Determined by both the energy output intensity within the grid and the viewport coverage:

[0114] ;

[0115] The numerator represents the sum of the absolute values ​​of the characteristic changes, and the denominator represents the magnitude of the characteristic vector. The larger the ratio, the more drastic the change. The total area is the projected area of ​​the grid within the viewport, where TotalViewArea is the total area of ​​the viewport. This formula comprehensively considers the drastic changes in data and their visual proportion, ensuring that high-value data is transmitted first. Through this step, the system can intelligently allocate limited network resources, guaranteeing smooth data refresh in core areas while data in peripheral areas can be appropriately delayed.

[0116] Step 3.4: Construct the incremental data packet stream and front-end rendering instruction set;

[0117] Finally, the segmented and sorted data is assembled into an incremental data packet stream, along with a front-end rendering instruction set. The data packet stream contains the grid ID, LOD level, incremental feature data, and state fingerprint. The rendering instruction set instructs the front-end on how to process this data, such as "replace," "overlay," or "interpolate." For incremental data, the front-end must perform a backprojection operation to restore the original feature values.

[0118] ;

[0119] Subsequently, the corresponding rendering engine interface is called according to the instruction set to map the new data to the Canvas or WebGL context. The instruction set also includes transition animation parameters, such as the duration of transparency fading. This ensures visual smoothness during data updates. The data packet stream and instruction set output by this step directly drive the visualization rendering on the browser side, achieving a seamless connection from backend data calculation to frontend visual presentation.

[0120] Step 4: Perform adaptive texture mapping and tiled rendering scheduling based on viewport projection;

[0121] Following the incremental data packet stream and front-end rendering instruction set generated in step three, this step aims to address the real-time rendering performance bottleneck of large-scale data on high-resolution screens in browsers. Traditional point-by-point rendering or full-image redrawing methods easily lead to excessive GPU load and frame rate drops when dealing with massive mesh data. Therefore, based on the target mesh set and its LOD level information selected in step three, this step constructs a dynamic texture mapping mechanism to transform discrete mesh data into continuous texture maps and employs a block-parallel rendering strategy to ensure a smooth visual experience even in complex interactive scenarios. Specifically, this includes:

[0122] Step 4.1: Dynamic construction and coordinate mapping of texture atlas based on LOD level;

[0123] First, using the multi-level feature pyramid data constructed in step two, combined with the current LOD level determined in step three... This involves dynamically constructing a texture atlas adapted to the current viewport. Because of the significant differences in mesh density across different levels, directly allocating independent texture units to each mesh would waste video memory. This step employs a dynamic atlas algorithm to merge spatially adjacent meshes within the same level into one or more texture blocks. Let the l-th level contain... There are i valid meshes, which are mapped to a 2D texture space of size W×H. For the i-th mesh, its normalized coordinates in the texture space are... Its relative position in the hexagonal grid index is determined by the following formula:

[0124] ;

[0125] in, The coordinates of the grid center in the local coordinate system. and These are the minimum and maximum boundary coordinates of all valid meshes at this level. This formula linearizes the irregular hexagonal distribution onto a regular square texture region, facilitating batch sampling by the GPU. Simultaneously, based on the incremental data transmitted in step three, only the pixel values ​​of the corresponding regions in the texture map are updated, while unchanged regions remain unchanged. This step achieves the transformation from discrete logical data to continuous graphics data, laying the texture foundation for subsequent efficient rendering.

[0126] Step 4.2: Geometry simplification and vertex compression based on viewport occlusion culling;

[0127] After constructing the texture atlas, further processing of geometry rendering efficiency is needed. Although the meshes selected in step three are within the viewport, some edge meshes may be difficult to distinguish with the naked eye due to perspective distortion or excessive distance. Continuing to render these fragmented geometries would consume a large amount of vertex processing power. This step implements viewport occlusion culling and geometry simplification strategies. The projected area of ​​each mesh in the current camera view is calculated. Screen pixel ratio :

[0128] ;

[0129] Set minimum visibility threshold (For example ),like If a mesh is deemed invisible or a low-priority object, it is classified as such. For meshes classified as low-priority, independent geometric patches are not generated; instead, their vertex information is merged and simplified with adjacent high-priority meshes. Let the number of vertices in the simplified mesh be... The original number of vertices is Then vertex compression ratio Defined as:

[0130] ;

[0131] By removing redundant vertices and merging coplanar triangles, the number of vertices that the GPU needs to process is significantly reduced. This step relies heavily on the texture coordinates generated in step four and the viewport range data in step three, ensuring that the simplified geometry is visually consistent with the original data. Only minor geometric precision is sacrificed in non-critical areas in exchange for a huge performance improvement.

[0132] Step 4.3: Multi-core parallel-based chunked rendering task decomposition and load balancing;

[0133] To fully utilize the multi-threading capabilities of modern browsers and avoid UI lag caused by main thread blocking, this step divides the geometry to be rendered into several independent sub-tasks (tiles) and distributes them to different WebWorker threads for parallel processing. Assume the total rendering area is divided into M×N rectangular blocks, each block containing a set of meshes. Calculate the rendering complexity of each block. Its relationship with the number of grids within that block and average texture resolution Proportional:

[0134] ;

[0135] in, This is a geometric complexity coefficient, dependent on the number of mesh normals and the computational cost of lighting. The system is based on the number of available CPU cores. The blocks are sorted from highest to lowest complexity, and a greedy algorithm is used to distribute them to various worker threads, so that the total computational load of each thread is minimized. As balanced as possible. The load balancing degree (Balance) is defined as:

[0136] ;

[0137] The goal is to bring the balance close to zero. This dynamic chunking and load balancing mechanism ensures a uniform distribution of rendering tasks across a multi-core environment, avoiding rendering latency caused by single-thread overload. The output of this step is a queue of decomposed rendering tasks assigned to specific threads, ready for final image compositing.

[0138] Step 4.4: Execute the parallel rendering pipeline and composite the framebuffer output;

[0139] Finally, after receiving the task queue, each worker thread immediately starts the parallel rendering pipeline. Each thread calls the GPU API (such as WebGL or WebGPU), loads the corresponding texture atlas and simplified geometry data, and performs rasterization and fragment shading operations. During rendering, the texture coordinates generated in step 4.1 are used for precise sampling, and vertex transformations are performed according to the simplified geometry in step 4.2. After rendering is complete, each thread writes the generated image data to a designated area of ​​the shared framebuffer. Due to the use of a block-based strategy, the threads do not interfere with each other, only performing simple memory copying in the final compositing stage. The composited complete image data is submitted to the display driver through a double-buffering mechanism to complete the rendering output of one frame. This process releases hardware performance through parallel computing, ensuring smooth rendering of over 60fps even with large amounts of data, meeting the stringent real-time requirements of interactive visualization.

[0140] Step 5: Implement dynamic thermal field reconstruction and multi-dimensional correlation mining based on user interaction feedback;

[0141] After achieving efficient basic rendering in step four, this step focuses on responding to the viewer's deeper interactive behaviors, transforming static visualizations into dynamic analysis tools. When users perform actions such as clicking, selecting, zooming, or dragging on the interface, the system no longer simply re-renders the view, but triggers underlying spatiotemporal correlation analysis, reconstructing the heat field in real time and uncovering multidimensional relationships between data, specifically including:

[0142] Step 5.1: Capture the interaction intent and parse the spatial query range and time span;

[0143] First, capture the user's interactive gestures in real time and parse the current spatial query range. With time span Whether clicking on a specific facility or selecting an area, the system needs to convert it into a mathematically described spatial polygon and time interval. Let the coordinates of the top-left corner of the user's selection be... The coordinates of the lower right corner are The start time is The end time is The query range can then be represented as:

[0144] ;

[0145] The system combines the viewport status from step three to determine whether the query is a global exploration or a local in-depth analysis. If the query scope is small and the time span is short, it is considered a local detail analysis; otherwise, it is considered a macro trend judgment. This step transforms ambiguous user actions into precise numerical parameters, providing clear input boundaries for subsequent aggregation calculations. This process directly calls the viewport data structure from step four's rendering, ensuring the continuity and responsiveness of interactive operations.

[0146] Step 5.2: Constructing an energy output correlation matrix based on spatial neighborhood relationships;

[0147] Based on the spatial query range parsed in step 5.1, all grid data within the range are extracted to construct a local energy output correlation matrix. The aim is to reveal the synergistic or complementary relationships between different clean energy facilities at the same spatiotemporal scale. Assume there are K grids within the range, with their power sequences as follows: Calculate the Pearson correlation coefficient between any two grids i and j. :

[0148] ;

[0149] in, Let be the average power of grid i over the time span. This coefficient ranges from -1 to 1, with positive values ​​indicating positive correlation (both rising and falling together) and negative values ​​indicating negative correlation (complementary effect). A K×K correlation matrix is ​​generated by iterating through all grid pairs. This matrix not only reflects the physical laws governing energy flow, but also implies the influence patterns of meteorological conditions on different microenvironments within the watershed.

[0150] Step 5.3: Dynamic reconstruction and anomaly propagation path tracing based on thermodynamic gradient field;

[0151] Using the correlation matrix constructed in step 5.2, the current thermal field distribution is dynamically reconstructed, and the propagation path of anomalies is traced. Traditional heatmaps only display numerical magnitudes, while this step introduces the concept of a gradient field to show the direction and intensity of data changes. The gradient vector field of the heatmap is calculated. Its components are approximated by the space derivative:

[0152]

[0153] Where H(x,y) is the normalized energy density of the grid (x,y) at the current time. This represents the spatial step size. By visualizing the gradient vector, users can intuitively see the center of energy accumulation and the direction of diffusion. Simultaneously, combined with the negative correlation region from step 5.2, potential abnormal propagation paths are identified. If a grid experiences a sudden power drop and its downstream adjacent grids show a strong positive correlation, it is determined to be fault propagation; if a strong negative correlation is shown, it may be load shifting. Tracking path length. It can be calculated using a breadth-first search algorithm, and the simplified formula is as follows:

[0154] ;

[0155] in, Here, d represents the nodes along the path, and d represents the geographical distance. This step transforms static data into a dynamic physical process simulation, helping users quickly pinpoint the source and scope of the problem.

[0156] Step 5.4: Generate a multidimensional correlation analysis report and interactive recommendation suggestions

[0157] Finally, by combining the correlation matrix from step 5.2 and the gradient field analysis results from step 5.3, a multidimensional correlation analysis report is automatically generated, providing interactive recommendations. The report covers major correlation hotspots, the topological structure of abnormal propagation paths, and potential risk warnings. Based on the analysis results, the system pushes next operational suggestions to the user, such as "It is recommended to zoom in to view the coupling relationship between region A and region B" or "Abnormal fluctuations have been detected in region C; it is recommended to investigate upstream meteorological data." The recommended confidence level (Conf) is determined by the absolute value of the correlation coefficient and the magnitude of the gradient.

[0158] ;

[0159] in, For the weighting coefficients, satisfying This step transforms complex mathematical calculations into intuitive decision support information, completing the leap from data visualization to intelligent analysis.

[0160] Step Six: Perform time-series evolution projection and future situation prediction based on historical trajectory backtracking;

[0161] Following the multidimensional correlation analysis and thermodynamic field reconstruction established in step five, this step aims to extend the perspective to the future along the timeline, utilizing the evolutionary patterns of historical data for situational projection. By simulating data change trends under different scenarios, it predicts future energy distribution patterns and potential risks for users, specifically including:

[0162] Step 6.1: Extract historical time-series evolution features and construct the state transition probability matrix;

[0163] First, we review the historical features extracted in steps one through five, particularly the correlation structure between the dimensionality-reduced feature vector generated in step two and the analysis obtained in step five. Based on long-term historical data, we construct a state transition probability matrix to describe the evolution of the system across different energy states. Let the system state at time t be... This state is determined by the dimension-reduced feature vectors defined earlier. Clustered. Statistics are derived from states. Transition to state Calculate the transition probability based on the frequency. :

[0164] ;

[0165] in, This indicates the state in historical data. The status then appeared The matrix characterizes the intrinsic dynamics of the watershed energy system, reflecting the combined impact of natural and anthropogenic factors such as weather changes and equipment operation. Through this step, the system transforms nonlinear time-series data into a computable probabilistic model, providing a mathematical foundation for subsequent extrapolations.

[0166] Step 6.2: Future scenario simulation based on Markov chain Monte Carlo method;

[0167] Using the state transition probability matrix constructed in step 6.1, the Markov Chain Monte Carlo (MCMC) method is employed to perform multiple stochastic simulations of the system state over a future period. The simulation duration is set to [value missing]. The initial state is the currently observed state. Through iterative sampling, M possible future evolution trajectories are generated. For the m-th trajectory, its state at time t is... From the state of the previous moment Determined by the transition probability matrix:

[0168] ;

[0169] Repeat this process until it is achieved. Each trajectory represents a possible future scenario, encompassing a range of possibilities from stable operation to extreme fluctuations. By collecting these M trajectories, a probability density function for future energy distribution can be constructed, thereby quantifying future uncertainty. This step transforms the abstract probabilistic model into a concrete numerical sequence, providing rich data support for visualization and simulation.

[0170] Step 6.3: Generate multi-scenario situational simulation maps and risk warnings for key nodes;

[0171] Based on the multiple evolution trajectories generated in step 6.2, the expected value and variance of each grid at future times are calculated to generate a multi-scenario situation projection map. The map not only displays the expected energy distribution but also uses color depth to indicate the degree of uncertainty in the prediction (i.e., the magnitude of the variance). For variances exceeding a preset threshold... Areas that meet the risk level are marked as high-risk areas, triggering an early warning mechanism. The warning level is determined by the risk probability. and scope of influence Joint decision:

[0172] ;

[0173] in, This represents the probability that an outlier will occur in a given grid within a future time period. This represents the area of ​​the high-risk cluster to which the grid belongs. This step transforms complex statistical extrapolation results into an intuitive graphical language, enabling users to easily identify potential future risk points.

[0174] Step 6.4: Output an interactive simulation report and decision support solution;

[0175] Finally, the situational simulation map and risk warning generated in step 6.3 are integrated into an interactive simulation report, providing decision support solutions. The report allows users to adjust key parameters (such as weather conditions and equipment maintenance plans) and observe changes in the simulation results in real time, enabling "if...then..." scenario analysis. Based on the optimal simulation path, the system automatically recommends scheduling strategies, such as "increasing photovoltaic output before the expected wind speed decreases" or "allocating energy storage resources in advance to cope with peak loads." Recommended strategy scores are also provided. Evaluated jointly by expected returns and risk control:

[0176] ;

[0177] Where E[Profit] represents the expected economic benefit. For risk exposure, This represents the preference coefficient. This step ultimately integrates historical data, real-time analysis, and future projections to form a complete decision support system, achieving the goal of interactive and visual analysis of historical clean energy data in large-scale river basins.

[0178] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. An interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation, characterized in that, include: Based on the geographical boundaries of the watershed, a hexagonal grid is constructed and static aggregation and initial screening of multi-source heterogeneous data are completed to generate a weighted static index matrix. The static index matrix is ​​used to perform streaming computation, extract dynamic spatiotemporal features and reduce dimensionality, and generate a multi-level resolution feature pyramid. Based on the feature pyramid, viewport-aware data distribution is performed, and incremental data packet streams and front-end rendering instruction sets are output. Based on the incremental data packet stream and the front-end rendering instruction set, adaptive texture mapping and block rendering scheduling are performed to generate a frame buffer image; In response to user interaction feedback, the dynamic thermal field is reconstructed and multi-dimensional correlations are mined based on the underlying feature data corresponding to the frame buffer image. Based on the aforementioned multidimensional relationships and historical feature data, a time-series evolution simulation is performed to generate a future situation prediction result.

2. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, The process of constructing a hexagonal grid and completing the static aggregation and initial screening of multi-source heterogeneous data includes: The watershed geospatial area is divided into hexagonal grids, and a topographic correction factor is introduced to calculate the effective coverage area; Multi-source heterogeneous data are spatially and temporally aligned using a hexagonal grid to generate a structured dataset; A multidimensional feature vector is constructed based on the structured dataset, and outliers are removed; A static index matrix based on grid identifiers is constructed using the cleaned data and preloaded into memory.

3. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, Performing streaming computation using the static index matrix includes: Calculate the local spatiotemporal covariance of the data in the static index matrix based on a sliding time window; Principal component analysis is performed on the covariance results to generate dimensionality-reduced eigenvectors. Calculate the spatiotemporal gradient between adjacent grids based on the dimensionality-reduced feature vectors to identify hotspot region boundaries; Based on the boundaries of the hotspot regions, a multi-level resolution feature pyramid and detail level index are generated.

4. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, Performing viewport-aware data distribution based on the feature pyramid includes: Based on the current browser viewport size and zoom level, filter the target mesh set from the feature pyramid; Compare the state fingerprint of the target grid set with the state of the previous frame, and calculate the difference in feature increment; Based on network bandwidth prediction, the incremental differences in the features are divided into packets and prioritized. The packetized results are assembled to generate an incremental data packet stream and a front-end rendering instruction set.

5. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, The execution of adaptive texture mapping and chunked rendering scheduling includes: Based on the level of detail in the incremental data packet stream, a texture atlas is dynamically constructed and the mesh is mapped to texture coordinates; Invisible geometry is removed based on the viewport projection area, and vertex compression is performed on the remaining mesh; The compressed geometry task is decomposed into blocks according to its complexity and distributed to multi-core threads to achieve load balancing. Each thread performs rendering in parallel and writes the results to the frame buffer to complete the composite output.

6. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, The process of reconstructing the dynamic thermal field based on the underlying feature data corresponding to the frame buffer image in response to user interaction feedback includes: Analyze user interaction gestures to obtain the spatial query range and time span; Based on the spatial query range, the power sequence of the corresponding grid is extracted from the feature pyramid, and the power sequence is used to construct an energy output correlation matrix; The correlation matrix and thermal gradient field are combined to trace the abnormal propagation path and generate a correlation analysis report.

7. The interactive visualization analysis method for historical clean energy data in large watersheds based on multi-dimensional correlation as described in claim 6, characterized in that, The construction of the energy output correlation matrix includes: Extract power data for each grid within a specified time span from the query range, calculate the Pearson correlation coefficient between power sequences of any two grids, generate a correlation matrix reflecting synergistic or complementary relationships, and identify potential anomalous propagation patterns based on negatively correlated regions.

8. The interactive visualization analysis method for historical clean energy data in large-scale watersheds based on multi-dimensional correlation as described in claim 1, characterized in that, The execution timing evolution deduction includes: Based on historical feature data, the system state is formed by clustering and the state transition frequency is statistically analyzed to construct a state transition probability matrix; Starting from the current state, the state transition probability matrix is ​​used to drive the Markov chain to simulate multiple future trajectories. Based on the future trajectories, the expectation and variance are calculated to generate a situation projection map.

9. The interactive visualization analysis method for historical clean energy data in large watersheds based on multi-dimensional correlation as described in claim 8, characterized in that, The generated situation projection map includes: Calculate the statistical variance of the simulated values ​​for each grid in future time periods, and mark the regions with variances exceeding a threshold as high-risk nodes; The warning level is determined by combining the risk probability and the scope of impact, and a simulation report with interactive parameter adjustment function is output.

10. A large-scale watershed clean energy historical data interactive visualization analysis system based on multi-dimensional correlation for implementing the method as described in any one of claims 1-9, characterized in that, include: The static aggregation module is used to construct a hexagonal grid based on the geographical boundary of the watershed and to complete the static aggregation and initial screening of multi-source heterogeneous data, outputting a static index matrix; The feature extraction module, connected to the static aggregation module, is used to perform streaming computation on the static index matrix and output a multi-level resolution feature pyramid. The data distribution module, connected to the feature extraction module, is used to perform viewport-aware data distribution based on the feature pyramid and output incremental data packet streams and front-end rendering instruction sets. The rendering scheduling module, connected to the data distribution module, is used to perform adaptive texture mapping and chunked rendering scheduling according to the incremental data packet stream and output a frame buffer image. The correlation mining module, connected to the rendering scheduling module, is used to respond to user interactions, reconstruct the dynamic thermal field based on the underlying feature data, and output multi-dimensional correlation analysis results. The situation simulation module connects the correlation mining module and the feature extraction module. It is used to perform time-series evolution simulation based on historical features and correlation relationships, and output the future situation simulation results.

Images