Cable damp fault diagnosis method and system based on machine learning

Abstract

The invention discloses a cable damp fault diagnosis method and system based on machine learning, and relates to the technical field of intelligent monitoring of power equipment. According to the method, key dimension information such as partial discharge, temperature distribution and leakage current is integrated through multi-source data feature-level deep fusion, the complementary value of each data source is accurately captured in combination with an attention mechanism, and then through the collaborative architecture of the multi-scale convolutional neural network and the bidirectional LSTM, the multi-scale convolutional neural network and the bidirectional LSTM are integrated. Meanwhile, local detail features and long-time-history space-time correlation features of the signals are mined, fine recognition of the damp state is achieved, data deep correlation is effectively mined, the early damp recognition accuracy can reach 95% or above, the SHAP interpretability analysis technology is innovatively integrated, the contribution degree of each feature to the diagnosis result is quantified, and the diagnosis accuracy is improved. And a visual feature importance thermodynamic diagram and a decision path diagram are generated, physical significance interpretation of a diagnosis basis is provided in combination with the insulation characteristics of the power equipment, and targeted operation and maintenance suggestions are directly output.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for power equipment, specifically to a method and system for diagnosing cable dampness faults based on machine learning. Background Technology

[0002] Power cables are critical equipment in power systems, and their insulation condition directly affects the reliability of power supply. Moisture absorption is one of the main factors leading to insulation degradation in cables, and early detection and diagnosis of moisture-induced cable faults are of great significance in preventing major accidents.

[0003] Existing cable moisture fault diagnosis technologies have the following prominent problems:

[0004] First, traditional diagnostic methods rely on single parameters such as insulation resistance and dielectric loss factor, which have low sensitivity to early, slight moisture damage and are prone to missed diagnoses. Second, although existing online monitoring systems can collect data from multiple sources, data fusion remains superficial and fails to fully explore the spatiotemporal correlation characteristics between various data sources, resulting in low data utilization efficiency. Third, diagnostic models are mostly black-box architectures, and the output results lack clear physical meaning, making it difficult for maintenance personnel to understand the diagnostic basis and affecting decision-making trust. Fourth, diagnostic models are statically trained and cannot adapt to dynamic scenarios such as changes in the field environment and equipment aging, and their diagnostic accuracy is prone to decline after long-term operation.

[0005] Therefore, developing a cable moisture fault diagnosis technology that combines high accuracy, strong interpretability, and adaptability has become a pressing technical challenge for the power industry. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for diagnosing cable moisture-induced faults based on machine learning, so as to solve the technical problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for diagnosing cable moisture-induced faults based on machine learning, comprising at least the following steps:

[0008] S1: Collect multi-source monitoring data on the cable's operating status, wherein the multi-source monitoring data includes at least two of the following: partial discharge data, temperature distribution data, and leakage current data;

[0009] S2: Perform data fusion processing on the collected multi-source monitoring data to construct a comprehensive feature vector, which includes at least a comprehensive feature vector of time domain features, frequency domain features and spatiotemporal correlation features;

[0010] S3: Input the comprehensive feature vector into the pre-trained deep learning diagnostic model. The deep learning diagnostic model adopts an architecture that combines a multi-scale convolutional neural network with an attention mechanism and outputs the diagnostic results of the cable's moisture status.

[0011] S4: Based on the diagnostic results and model confidence assessment, establish an interpretability analysis report for the diagnostic results;

[0012] S5: Online updates of model parameters.

[0013] Furthermore, the data fusion process includes at least the following steps:

[0014] S201: Perform time series alignment and standardization on multi-source monitoring data to ensure data temporal consistency and dimensional uniformity;

[0015] S202: Extracting multi-scale features from various data sources: extracting pulse waveform features and phase distribution features from partial discharge data; extracting axial temperature gradient features and thermal diffusion features from temperature distribution data; extracting harmonic features and trend change features from leakage current data;

[0016] S203: A feature extraction network constructed from multiple one-dimensional convolutional layers and an attention mechanism is used to process the data sources for the first... There are 1 data source, and the characteristics of its output are represented as follows:

[0017]

[0018] in, Input data; and Here are the parameters for the convolutional layer; Conv1D is the one-dimensional convolution operation; Attn(·) is the attention weighting function;

[0019] S204: The extracted features are adaptively weighted and fused using a feature attention mechanism. The attention weights are calculated as follows:

[0020]

[0021] in, For the first One characteristic, and For learnable parameters, j represents the corresponding attention weight.

[0022] Furthermore, the deep learning diagnostic model in S3 includes a multi-scale feature extraction module, a spatiotemporal feature fusion module, an attention weighting module, and a classification output module;

[0023] S301: Extracts local and global features of the signal in parallel using convolution kernels of different sizes through a multi-scale feature extraction module;

[0024] S302: A spatiotemporal feature fusion module is used to capture the temporal dependence of features through a bidirectional LSTM network;

[0025] S303: Employs an attention-weighted module to enhance key features using a self-attention mechanism;

[0026] S304: Output the moisture status level and corresponding confidence level of the cable through the classification output module.

[0027] Furthermore, S4 includes at least the following steps:

[0028] S401: The SHAP method was used to analyze the contribution of each feature to the diagnostic results;

[0029] S402: Generate feature importance heatmaps and decision path maps;

[0030] S403: Explanation of the physical meaning of the diagnostic basis and maintenance suggestions provided by S402.

[0031] Furthermore, S5 includes at least the following steps:

[0032] S501: Collect new diagnostic samples and their validation results to construct an incremental learning dataset;

[0033] S502: The knowledge distillation method is used to transfer the knowledge of the old model to the new model, so as to realize the online updating of model parameters.

[0034] A system for diagnosing cable moisture-induced faults based on machine learning includes a data acquisition module, an edge computing module, a cloud analysis platform, an operation and maintenance management module, and a model verification module.

[0035] The data acquisition module is used to collect multi-source monitoring data of the cable in real time.

[0036] The edge computing module is deployed at the cable site for data preprocessing and feature extraction;

[0037] The cloud-based analysis platform includes a feature fusion unit, an intelligent diagnostic unit, and a visualization unit;

[0038] The feature fusion unit is used for feature-level fusion of multi-source data; the intelligent diagnosis unit includes a deep learning diagnosis model and a model management module; the visualization unit is used to display the diagnosis results and interpretable analysis reports.

[0039] The operation and maintenance management module is used to receive diagnostic results and generate maintenance work orders;

[0040] The model verification module includes an on-site verification interface, a model performance evaluation unit, and a model version management unit;

[0041] The on-site verification interface is used to receive diagnostic results feedback from maintenance personnel; the model performance evaluation unit is used to continuously monitor the diagnostic accuracy and false alarm rate of the model; and the model version management unit is used to manage different versions of the diagnostic model.

[0042] Furthermore, the data acquisition module includes a distributed fiber optic temperature sensor, a high-frequency partial discharge sensor, a high-precision current transformer, and a multi-sensor time synchronization unit;

[0043] The distributed optical fiber temperature sensor is used to continuously monitor the temperature distribution of the cable body.

[0044] The high-frequency partial discharge sensor is used to collect partial discharge pulse signals;

[0045] The high-precision current transformer is used to monitor the leakage current of the grounding wire.

[0046] The multi-sensor time synchronization unit is used to ensure the time consistency of multi-source data.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] 1. This invention integrates key dimensions such as partial discharge, temperature distribution, and leakage current through deep fusion of multi-source data features. It combines attention mechanism to accurately capture the complementary value of each data source. Then, through the collaborative architecture of multi-scale convolutional neural network and bidirectional LSTM, it simultaneously mines the local detailed features and long-term spatiotemporal correlation features of the signal, so as to achieve refined identification of the damp state, effectively mine deep data correlation, and achieve an early damp identification accuracy of over 95%.

[0049] 2. This invention innovatively integrates SHAP interpretability analysis technology, which generates intuitive feature importance heatmaps and decision path diagrams by quantifying the contribution of each feature to the diagnostic results. At the same time, it combines the insulation characteristics of power equipment to provide a physical interpretation of the diagnostic basis and directly outputs targeted operation and maintenance suggestions. This not only solves the problem of operation and maintenance personnel trusting the intelligent diagnostic results, but also transforms the diagnostic results into operational guidelines that can be directly implemented, significantly reducing the threshold for operation and maintenance decisions and improving the efficiency of fault diagnosis.

[0050] 3. This invention utilizes a knowledge distillation-driven online update mechanism to integrate newly labeled samples into the incremental learning dataset without interrupting system operation. This enables efficient transfer of knowledge from the old model to the new model, ensuring that the model can dynamically adapt to changes in the field environment and the aging characteristics of equipment. The system is set to automatically trigger the update process when 1000 new labeled data are added. Combined with the model version management unit, version traceability and rollback are achieved. The system can maintain stable diagnostic accuracy during long-term operation, avoiding missed diagnoses and false alarms caused by model failure, and significantly reducing system maintenance costs.

[0051] 4. This invention adopts a cloud-edge collaborative architecture. The edge computing module is deployed at the cable site, enabling rapid data preprocessing and preliminary feature extraction. This effectively reduces the data transmission bandwidth consumption of 5G private networks / fiber optic lines, adapting to cable monitoring needs in remote areas or complex terrains. The cloud analysis platform provides centralized feature fusion, model inference, and visualization, supporting access from multiple terminals including web and mobile apps, facilitating remote real-time monitoring of equipment status by maintenance personnel. Furthermore, this technical solution is not limited to power cables. By adjusting the feature extraction dimensions and model parameters, it can be quickly adapted to insulation status diagnosis of other power equipment such as transformers and GIS systems without requiring system architecture reconstruction, significantly expanding industrial application scenarios and providing an integrated solution for intelligent monitoring of power system equipment. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0054] Figure 2 This is a flowchart illustrating the deep learning diagnostic model of the present invention.

[0055] Figure 3 This is a schematic diagram of the cloud-based analytics platform of the present invention;

[0056] Figure 4 This is a schematic diagram illustrating the technical effects of the present invention. Detailed Implementation

[0057] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0058] Example 1:

[0059] See Figure 1 and Figure 2 A machine learning-based method for diagnosing cable moisture problems includes at least the following steps:

[0060] S1: Collect multi-source monitoring data on the cable's operating status. The multi-source monitoring data shall include at least two of the following: partial discharge data, temperature distribution data, and leakage current data.

[0061] S2: Perform data fusion processing on the collected multi-source monitoring data to construct a comprehensive feature vector. The comprehensive feature vector shall include at least a comprehensive feature vector of time domain features, frequency domain features and spatiotemporal correlation features.

[0062] S3: Input the comprehensive feature vector into the pre-trained deep learning diagnostic model. The deep learning diagnostic model adopts an architecture that combines multi-scale convolutional neural networks with attention mechanisms, and outputs the diagnostic results of the cable's moisture status.

[0063] S4: Based on the diagnostic results and model confidence assessment, establish an interpretability analysis report for the diagnostic results;

[0064] S5: Online updates of model parameters.

[0065] Data fusion processing includes at least the following steps:

[0066] S201: Perform time series alignment and standardization on multi-source monitoring data to ensure data temporal consistency and dimensional uniformity;

[0067] S202: Extracting multi-scale features from various data sources: extracting pulse waveform features and phase distribution features from partial discharge data; extracting axial temperature gradient features and thermal diffusion features from temperature distribution data; extracting harmonic features and trend change features from leakage current data;

[0068] S203: A feature extraction network constructed from multiple one-dimensional convolutional layers and an attention mechanism is used to process the data sources for the first... There are 1 data source, and the characteristics of its output are represented as follows:

[0069]

[0070] in, Input data; and Here are the parameters for the convolutional layer; Conv1D is the one-dimensional convolution operation; Attn(·) is the attention weighting function;

[0071] S204: The extracted features are adaptively weighted and fused using a feature attention mechanism. The attention weights are calculated as follows:

[0072]

[0073] in, For the first One characteristic, and For learnable parameters, j represents the corresponding attention weight.

[0074] The deep learning diagnostic model in S3 includes a multi-scale feature extraction module, a spatiotemporal feature fusion module, an attention weighting module, and a classification output module;

[0075] S301: Extracts local and global features of the signal in parallel using convolution kernels of different sizes through a multi-scale feature extraction module;

[0076] S302: A spatiotemporal feature fusion module is used to capture the temporal dependence of features through a bidirectional LSTM network;

[0077] S303: Employs an attention-weighted module to enhance key features using a self-attention mechanism;

[0078] S304: Output the moisture status level and corresponding confidence level of the cable through the classification output module.

[0079] S4 includes at least the following steps:

[0080] S401: The SHAP method was used to analyze the contribution of each feature to the diagnostic results;

[0081] S402: Generate feature importance heatmaps and decision path maps;

[0082] S403: Explanation of the physical meaning of the diagnostic basis and maintenance suggestions provided by S402.

[0083] S5 includes at least the following steps:

[0084] S501: Collect new diagnostic samples and their validation results to construct an incremental learning dataset;

[0085] S502: The knowledge distillation method is used to transfer the knowledge of the old model to the new model, so as to realize the online updating of model parameters.

[0086] Specifically, one implementation application is proposed:

[0087] Taking a 110kV cross-linked polyethylene cable as the monitoring object, the implementation steps are as follows:

[0088] Data acquisition deployment: A distributed fiber optic temperature measurement system (one temperature measurement point per meter, sampling frequency 1Hz), an ultra-high frequency partial discharge sensor (deployed at the cable joint, sampling frequency 100MHz), and a high frequency current transformer (inserted on the grounding wire, sampling frequency 10kHz) are installed on the cable body and joints. Multi-sensor data time synchronization is achieved through a GPS / BeiDou time synchronization unit.

[0089] Feature extraction and fusion:

[0090] Twenty-dimensional features, including pulse amplitude, phase distribution, and repetition rate, were extracted from partial discharge data.

[0091] Calculate 15-dimensional features such as axial temperature gradient, rate of change of temperature difference, and thermal diffusivity from temperature data;

[0092] The leakage current data was analyzed for 10-dimensional characteristics, including fundamental amplitude, harmonic content, and trend slope.

[0093] A one-dimensional convolutional-attention feature extraction network is used to process features from various data sources. The 45-dimensional features are adaptively weighted and fused through a feature attention mechanism to generate a comprehensive feature vector.

[0094] Model diagnostic process: The comprehensive feature vector is input into the pre-trained multi-scale CNN-LSTM diagnostic model. After the model extracts local / global features through multi-scale convolutional kernels, captures time dependence through bidirectional LSTM, and strengthens key features through self-attention mechanism, the diagnostic results are output through Softmax, including four state levels: normal, mildly damp, moderately damp, and severely damp, and their corresponding confidence scores (the confidence score threshold is set to 0.85).

[0095] Interpretability Analysis and Maintenance Recommendations: When the system detects that the cable is in a moderately damp state (confidence level 0.92), the SHAP method analysis shows that the temperature gradient feature contribution is 45% and the partial discharge phase concentration feature contribution is 30%. A feature importance heatmap and decision path map are generated. Combined with the cable structural characteristics, the maintenance recommendation is to "focus on checking the sealing performance of the cable intermediate joints and investigate the moisture intrusion channels".

[0096] Example 2:

[0097] See Figure 3 This embodiment proposes a system for diagnosing cable moisture-induced faults based on machine learning, based on the above embodiment one. The system includes a data acquisition module, an edge computing module, a cloud analysis platform, an operation and maintenance management module, and a model verification module.

[0098] The data acquisition module is used to collect multi-source monitoring data of the cable in real time;

[0099] Edge computing modules are deployed at the cable site for data preprocessing and feature extraction;

[0100] The cloud-based analytics platform includes a feature fusion unit, an intelligent diagnostic unit, and a visualization unit;

[0101] The feature fusion unit is used for feature-level fusion of multi-source data; the intelligent diagnosis unit includes a deep learning diagnostic model and a model management module; the visualization unit is used to display diagnostic results and interpretable analysis reports.

[0102] The operation and maintenance management module is used to receive diagnostic results and generate maintenance work orders;

[0103] The model validation module includes a field validation interface, a model performance evaluation unit, and a model version management unit;

[0104] The on-site verification interface is used to receive diagnostic results feedback from operations and maintenance personnel; the model performance evaluation unit is used to continuously monitor the diagnostic accuracy and false alarm rate of the model; and the model version management unit is used to manage different versions of the diagnostic model.

[0105] The data acquisition module includes a distributed fiber optic temperature sensor, a high-frequency partial discharge sensor, a high-precision current transformer, and a multi-sensor time synchronization unit.

[0106] Distributed fiber optic temperature sensors are used to continuously monitor the temperature distribution within the cable body;

[0107] High-frequency partial discharge sensors are used to acquire partial discharge pulse signals;

[0108] High-precision current transformers are used to monitor leakage current in grounding wires;

[0109] The multi-sensor time synchronization unit is used to ensure the time consistency of multi-source data.

[0110] Based on Embodiment 2, a specific system deployment and implementation method is proposed. This system adopts a cloud-edge collaborative architecture for deployment, as detailed below:

[0111] Edge-side deployment: An industrial control computer is deployed at the cable site as an edge computing gateway to run data acquisition drivers, time synchronization protocols and feature extraction algorithms, enabling real-time preprocessing and preliminary feature extraction of multi-source data, reducing transmission bandwidth usage;

[0112] Cloud Deployment: Deploy a cloud-based analysis platform in the power operation and maintenance center, including a feature fusion server, intelligent diagnostic server, visualization server, and cloud storage array. Ensure data security through a security gateway / firewall, and enable device interconnection through industrial switches.

[0113] Communication Link: A 5G private network is used as the data transmission channel to achieve low-latency and high-reliability data interaction between the edge and the cloud, and to support the stable transmission of cable monitoring data in remote areas;

[0114] Terminal display: Display diagnostic results, feature visualization charts, operation and maintenance suggestions and model status to operation and maintenance personnel through the web management interface (adapted to PC) and mobile APP (supports iOS / Android systems), and support online generation and circulation of work orders;

[0115] Model update mechanism: The system sets an incremental data threshold (1000 labeled samples). When the newly added valid labeled data reaches the threshold, the model update process is automatically triggered. Knowledge distillation technology is used to migrate the knowledge of the old model to the new model. The update process does not affect the normal operation of the system. After the update, the version information is recorded through the model version management unit, and rollback operation is supported.

[0116] Based on the above technical effect verification:

[0117] See Figure 4 A six-month trial run was conducted in a cable tunnel in a certain city. During this period, the system monitored a total cable length of 12km and successfully issued early warnings for three early moisture-related faults. On-site verification confirmed that all three faults were consistent with the actual conditions, achieving a diagnostic accuracy rate of 96.3%. The average warning time was 15 days earlier than traditional insulation resistance monitoring methods. The interpretability analysis function effectively assisted maintenance personnel in quickly locating fault points, improving maintenance efficiency by more than 40%.

[0118] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for diagnosing cable moisture-induced faults based on machine learning, characterized in that: At least the following steps are included: S1: Collect multi-source monitoring data on the cable's operating status, wherein the multi-source monitoring data includes at least two of the following: partial discharge data, temperature distribution data, and leakage current data; S2: Perform data fusion processing on the collected multi-source monitoring data to construct a comprehensive feature vector, which includes at least a comprehensive feature vector of time domain features, frequency domain features and spatiotemporal correlation features; S3: Input the comprehensive feature vector into the pre-trained deep learning diagnostic model. The deep learning diagnostic model adopts an architecture that combines a multi-scale convolutional neural network with an attention mechanism and outputs the diagnostic results of the cable's moisture status. S4: Based on the diagnostic results and model confidence assessment, establish an interpretability analysis report for the diagnostic results; S5: Online updates of model parameters.

2. The method for diagnosing cable moisture-induced faults based on machine learning according to claim 1, characterized in that: The data fusion process includes at least the following steps: S201: Perform time series alignment and standardization on multi-source monitoring data to ensure data temporal consistency and dimensional uniformity; S202: Extracting multi-scale features from various data sources: extracting pulse waveform features and phase distribution features from partial discharge data; extracting axial temperature gradient features and thermal diffusion features from temperature distribution data; extracting harmonic features and trend change features from leakage current data; S203: A feature extraction network constructed from multiple one-dimensional convolutional layers and an attention mechanism is used to process the data sources for the first... There are 1 data source, and the characteristics of its output are represented as follows: in, Input data; and Here are the parameters for the convolutional layer; Conv1D is the one-dimensional convolution operation; Attn(·) is the attention weighting function; S204: The extracted features are adaptively weighted and fused using a feature attention mechanism. The attention weights are calculated as follows: in, For the first One characteristic, and For learnable parameters, j represents the corresponding attention weight.

3. The method for diagnosing cable moisture-induced faults based on machine learning according to claim 1, characterized in that: The deep learning diagnostic model in S3 includes a multi-scale feature extraction module, a spatiotemporal feature fusion module, an attention weighting module, and a classification output module. S301: Extracts local and global features of the signal in parallel using convolution kernels of different sizes through a multi-scale feature extraction module; S302: A spatiotemporal feature fusion module is used to capture the temporal dependence of features through a bidirectional LSTM network; S303: Employs an attention-weighted module to enhance key features using a self-attention mechanism; S304: Output the moisture status level and corresponding confidence level of the cable through the classification output module.

4. The method for diagnosing cable moisture-induced faults based on machine learning according to claim 1, characterized in that: The S4 includes at least the following steps: S401: The SHAP method was used to analyze the contribution of each feature to the diagnostic results; S402: Generate feature importance heatmaps and decision path maps; S403: Explanation of the physical meaning of the diagnostic basis and maintenance suggestions provided by S402.

5. The method for diagnosing cable moisture-induced faults based on machine learning according to claim 1, characterized in that: The S5 includes at least the following steps: S501: Collect new diagnostic samples and their validation results to construct an incremental learning dataset; S502: The knowledge distillation method is used to transfer the knowledge of the old model to the new model, so as to realize the online updating of model parameters.

6. A system for diagnosing cable moisture-induced faults based on machine learning, used in the machine learning-based cable moisture-induced fault diagnosis method described in any one of claims 1-5, characterized in that: It includes a data acquisition module, an edge computing module, a cloud analytics platform, an operations and maintenance management module, and a model validation module; The data acquisition module is used to collect multi-source monitoring data of the cable in real time. The edge computing module is deployed at the cable site for data preprocessing and feature extraction; The cloud-based analysis platform includes a feature fusion unit, an intelligent diagnostic unit, and a visualization unit; The feature fusion unit is used for feature-level fusion of multi-source data; The intelligent diagnostic unit includes a deep learning diagnostic model and a model management module; the visualization unit is used to display diagnostic results and interpretable analysis reports. The operation and maintenance management module is used to receive diagnostic results and generate maintenance work orders; The model verification module includes an on-site verification interface, a model performance evaluation unit, and a model version management unit; The on-site verification interface is used to receive diagnostic results feedback from maintenance personnel; the model performance evaluation unit is used to continuously monitor the diagnostic accuracy and false alarm rate of the model; and the model version management unit is used to manage different versions of the diagnostic model.

7. The method and system for diagnosing cable moisture-induced faults based on machine learning according to claim 6, characterized in that: The data acquisition module includes a distributed fiber optic temperature sensor, a high-frequency partial discharge sensor, a high-precision current transformer, and a multi-sensor time synchronization unit. The distributed optical fiber temperature sensor is used to continuously monitor the temperature distribution of the cable body. The high-frequency partial discharge sensor is used to collect partial discharge pulse signals; The high-precision current transformer is used to monitor the leakage current of the grounding wire. The multi-sensor time synchronization unit is used to ensure the time consistency of multi-source data.

Images