New energy operation multi-temporal-spatial-scale situation awareness method

Abstract

The invention discloses a new energy operation multi-temporal-spatial-scale situation awareness method. The method comprises the steps of multi-source heterogeneous data acquisition and preprocessing, multi-temporal-spatial-scale feature extraction and fusion, uncertainty quantification and probability prediction, edge cloud collaborative real-time situation assessment and multi-target collaborative optimization and feedback control. The invention belongs to the technical field of smart power grids, and particularly relates to a new energy operation multi-temporal-spatial-scale situation awareness method, which adopts multi-source heterogeneous data acquisition and preprocessing, clarifies data sources, preprocesses data through temporal-spatial alignment and anomaly detection, reconstructs an error to capture a complex anomaly mode by using a variational auto-encoder, and achieves the multi-temporal-spatial-scale situation awareness. Detecting a hidden equipment fault; edge cloud is adopted to cooperate with real-time situation assessment, a lightweight anomaly detection model is deployed on the edge side through a knowledge distillation technology, a local outlier factor algorithm is adopted to carry out millisecond anomaly detection, update parameters are issued through cloud digital twin modeling, and the communication efficiency is optimized.

Description

Technical Field

[0001] The present invention belongs to the field of smart grid technology, and specifically provides a multi-temporal and spatial scale situation awareness method for new energy operation. Background Art

[0002] Multi-spatiotemporal-scale situational awareness for new energy systems aims to improve their safety, stability, and economic efficiency by enabling real-time perception, dynamic analysis, and collaborative prediction of their operating status across different time dimensions and spatial levels. However, existing multi-spatiotemporal-scale situational awareness methods for new energy systems suffer from technical issues such as mismatched spatiotemporal scales, difficulty in data fusion, and significant noise interference, resulting in insufficient anomaly detection accuracy, difficulty in detecting hidden faults, and the tendency to miss early equipment failures. They also suffer from insufficient real-time edge computing, insufficient consistency across spatiotemporal models, and significant latency in massive data transmission, leading to a greater risk of fault propagation, ineffective energy efficiency optimization, and distorted state perception. Summary of the Invention

[0003] In response to the above situation, in order to overcome the defects of the existing technology, the present invention provides a multi-spatiotemporal scale situational awareness method for new energy operation. In order to solve the technical problems of spatiotemporal scale mismatch, difficult data fusion, large noise interference, insufficient anomaly detection accuracy, difficult detection of hidden faults, and easy omission of early equipment faults, multi-source heterogeneous data collection and preprocessing are adopted to clarify the data source, pre-process the data through spatiotemporal alignment and anomaly detection, and use variational autoencoder reconstruction error to capture complex abnormal patterns and detect hidden equipment failures; in order to solve the technical problems of insufficient real-time edge computing, insufficient consistency of cross-spatiotemporal models, large delay in massive data transmission, resulting in a large risk of fault propagation, failure of energy efficiency optimization, and distortion of state perception, edge-cloud collaborative real-time situation assessment is adopted, and a lightweight anomaly detection model is deployed on the edge side through knowledge distillation technology. The local outlier factor algorithm is used for millisecond-level anomaly detection, and updated parameters are issued through cloud-based digital twin modeling to optimize communication efficiency.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides a method for multi-temporal and spatial scale situational awareness of new energy operation, which includes the following steps:

[0005] Step S1: Multi-source heterogeneous data collection and preprocessing: clarify the data source, preprocess the data through spatiotemporal alignment and anomaly detection, and use variational autoencoder reconstruction error to capture complex abnormal patterns and detect hidden equipment failures;

[0006] Step S2: Extract and fuse multi-spatiotemporal scale features, perform time scale hierarchical processing, and achieve spatial feature fusion of multiple new energy stations through federated learning to generate a comprehensive situation representation;

[0007] Step S3: Uncertainty quantification and probabilistic prediction, using Bayesian neural network combined with Monte Carlo method, generating multiple sets of prediction curves through random forward propagation, calculating the uncertainty of the prediction expectation and variance quantification model cognition, and forming a probabilistic prediction result with confidence interval;

[0008] Step S4: Edge-cloud collaborative real-time situation assessment, deploying a lightweight anomaly detection model on the edge side through knowledge distillation technology, using a local outlier factor algorithm for millisecond-level anomaly detection, and issuing updated parameters through cloud-based digital twin modeling to optimize communication efficiency;

[0009] Step S5: Multi-objective collaborative optimization and feedback control, feeding back actual operation data to the digital twin platform to achieve continuous optimization and calibration of the model.

[0010] Furthermore, in step S1, the multi-source heterogeneous data collection and preprocessing includes the following steps:

[0011] Step S11: Data collection, data sources include real-time monitoring data, equipment operation logs and external environment data;

[0012] The real-time monitoring data includes SCADA, PMU, meteorological sensors and satellite remote sensing;

[0013] The equipment operation logs, including photovoltaic inverters, wind turbine controllers, and energy storage SOCs;

[0014] The external environmental data includes irradiance, wind speed, temperature and grid dispatch instructions;

[0015] Step S12: data preprocessing, including spatiotemporal alignment and anomaly detection;

[0016] The spatiotemporal alignment includes using a sliding time window and spatial interpolation to unify the data granularity of different sampling frequencies;

[0017] Align time and space to establish a unified time and space coordinate system. The formula used is as follows:

[0018] ;

[0019] Where, represents the standardized time point after alignment, t represents the timestamp of the original data, represents the base time resolution, represents Gaussian weight, Represents spatial interpolation, target position The interpolation result at , represents the measured value of the i-th neighboring observation point, and n represents the number of neighboring observation points involved in interpolation;

[0020] The anomaly detection includes removing noise data based on isolation forest and variational autoencoder;

[0021] Anomaly detection quickly identifies obvious anomalies using the following formula:

[0022] ;

[0023] Where x represents the input feature vector, represents the path length, represents the expected value of the path length, represents the normalization factor, Indicates the anomaly score. The closer it is to 1, the higher the probability of anomaly. When it is greater than 0.6, it is considered an anomaly.

[0024] Step S13: The variational autoencoder reconstructs the error, captures complex nonlinear abnormal patterns, and detects hidden equipment degradation failures that are difficult to detect with the isolation forest. The formula used is as follows:

[0025] ;

[0026] Where, represents the loss function of the variational autoencoder, represents the encoder network, which is used to map the input to the latent space, and D() is used to reconstruct the data from the latent space. represents the approximate posterior distribution, Represents the prior distribution, KL() represents the KL divergence, which is used to constrain the latent space distribution to prevent overfitting and set the abnormality judgment threshold m. It is judged as abnormal.

[0027] Furthermore, in step S2, the multi-spatiotemporal scale feature extraction and fusion includes the following steps:

[0028] Step S21: time scale hierarchical processing, including equipment level processing, station level processing and regional level processing;

[0029] The device-level processing extracts transient features such as high-frequency vibration, voltage and current signals through a dynamic graph convolutional network, providing fine-grained features for device health status assessment;

[0030] The station-level processing uses a spatiotemporal attention network to capture local correlations, analyze power output and load trends, establish spatiotemporal correlations between equipment clusters, and optimize station operation strategies.

[0031] The regional-level processing uses Transformer encoders to model global dependent climate patterns and cross-regional load trends, capture atmospheric circulation, and cross-regional power dispatch to support wide-area energy coordination.

[0032] Step S22: Spatial scale fusion, aggregating the characteristics of multiple new energy sites through federated learning, generating regional level situation representation, and realizing cross-spatial scale information fusion.

[0033] Furthermore, in step S3, the uncertainty quantification and probability prediction are specifically performed by using a Bayesian neural network to output a confidence interval of a prediction value and a Monte Carlo simulation of a random scenario to generate multiple sets of curves;

[0034] Monte Carlo sampling obtains probability output through T random forward propagation. The formula used is as follows:

[0035] ;

[0036] Where, represents the network weight of the t-th sampling, represents the network weight of the t-th sampling, Represents the forward propagation of a neural network with random weights, randomly discarding some neurons each time it is executed, and T represents the total number of sampling times. represents the expected value of the forecast, Represents the prediction variance, which is used to characterize the uncertainty of model cognition.

[0037] Furthermore, in step S4, the edge cloud collaborates with real-time situation assessment, including the following steps:

[0038] Step S41: deploying a lightweight model to process high-frequency data;

[0039] Model deployment uses knowledge distillation technology to compress the model and deploy a lightweight anomaly detector. The formula used is as follows:

[0040] ;

[0041] Where, represents the knowledge distillation loss function, y represents the true label, represents the student model prediction output, represents the high-precision teacher model on the cloud, represents the edge-side student model, Used to balance task loss and knowledge transfer loss, represents the task loss function, represents KL divergence loss;

[0042] Step S42: millisecond-level anomaly detection, based on the local outlier factor algorithm of the sliding window, real-time calculation of density deviation and low-latency instantaneous anomaly detection, to meet the timeliness requirements of rapid response to power equipment failures. The formula used is as follows:

[0043] ;

[0044] Where LOF represents the local outlier factor function, p represents the data point to be detected, k represents the number of neighbors, represents the set of k nearest neighbors of p, represents the local reachability density;

[0045] Step S43: Cloud-based digital twin modeling and regular parameter updates to edge devices;

[0046] High-precision twin construction, dynamic coupling of physical models and data-driven models, the formula used is as follows:

[0047] ;

[0048] Where, represents the device mechanism equation, Represents the adaptive weight, with an online adjustment range of 0.2 to 0.8. represents the residual network;

[0049] Step S44: Communication optimization, using compressed sensing to reduce data transmission volume and lower latency;

[0050] The compression rate is dynamically adjusted based on channel quality. The formula used is as follows:

[0051] ;

[0052] Where, represents the compression ratio, and SNR represents the channel noise ratio.

[0053] Furthermore, in step S5, the scheme is optimized and fed back, specifically by establishing a multi-objective optimization framework, with the optimization objectives including safety, economy and low carbon; using multi-agent reinforcement learning to coordinate resources, generating dynamic adjustment strategies, and realizing distributed free collaborative optimization; and transmitting actual operation data back to the digital twin platform to continuously calibrate the model.

[0054] The beneficial results achieved by the present invention using the above scheme are as follows:

[0055] (1) In order to solve the technical problems of insufficient anomaly detection accuracy, difficulty in detecting hidden faults, and easy omission of early equipment faults due to the mismatch of time and space scales, difficulty in data fusion, and large noise interference, we adopt multi-source heterogeneous data collection and preprocessing, clarify the data source, preprocess the data through time and space alignment and anomaly detection, and use variational autoencoder reconstruction error to capture complex abnormal patterns and detect hidden equipment faults;

[0056] (2) In order to address the technical problems of insufficient real-time performance of edge computing, insufficient consistency of cross-temporal and spatial models, large delay in massive data transmission, resulting in a high risk of fault propagation, failure of energy efficiency optimization, and distortion of state perception, edge cloud collaborative real-time situation assessment is adopted. A lightweight anomaly detection model is deployed on the edge side through knowledge distillation technology, and a local outlier factor algorithm is used for millisecond-level anomaly detection. Updated parameters are issued through cloud-based digital twin modeling to optimize communication efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a process for a multi-temporal and spatial scale situation awareness method for new energy operation provided by the present invention;

[0058] Figure 2 Schematic diagram of the process of step S1;

[0059] Figure 3 Schematic diagram of the process of step S2;

[0060] Figure 4 Schematic diagram of the process of step S4.

[0061] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0063] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0064] Example 1, see Figure 1 The present invention provides a method for multi-temporal and spatial scale situation awareness of new energy operation, which includes the following steps:

[0065] Step S1: Multi-source heterogeneous data collection and preprocessing: clarify the data source, preprocess the data through spatiotemporal alignment and anomaly detection, and use variational autoencoder reconstruction error to capture complex abnormal patterns and detect hidden equipment failures;

[0066] Step S2: Extract and fuse multi-spatiotemporal scale features, perform time scale hierarchical processing, and achieve spatial feature fusion of multiple new energy stations through federated learning to generate a comprehensive situation representation;

[0067] Step S3: Uncertainty quantification and probabilistic prediction, using Bayesian neural network combined with Monte Carlo method, generating multiple sets of prediction curves through random forward propagation, calculating the uncertainty of the prediction expectation and variance quantification model cognition, and forming a probabilistic prediction result with confidence interval;

[0068] Step S4: Edge-cloud collaborative real-time situation assessment, deploying a lightweight anomaly detection model on the edge side through knowledge distillation technology, using a local outlier factor algorithm for millisecond-level anomaly detection, and issuing updated parameters through cloud-based digital twin modeling to optimize communication efficiency;

[0069] Step S5: Multi-objective collaborative optimization and feedback control, feeding back actual operation data to the digital twin platform to achieve continuous optimization and calibration of the model.

[0070] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the multi-source heterogeneous data collection and preprocessing includes the following steps:

[0071] Step S11: Data collection, data sources include real-time monitoring data, equipment operation logs and external environment data;

[0072] The real-time monitoring data includes SCADA, PMU, meteorological sensors and satellite remote sensing;

[0073] The equipment operation logs, including photovoltaic inverters, wind turbine controllers, and energy storage SOCs;

[0074] The external environmental data includes irradiance, wind speed, temperature and grid dispatch instructions;

[0075] Step S12: data preprocessing, including spatiotemporal alignment and anomaly detection;

[0076] The spatiotemporal alignment includes using a sliding time window and spatial interpolation to unify the data granularity of different sampling frequencies;

[0077] Align time and space to establish a unified time and space coordinate system. The formula used is as follows:

[0078] ;

[0079] Where, represents the standardized time point after alignment, t represents the timestamp of the original data, represents the base time resolution, represents Gaussian weight, Represents spatial interpolation, target position The interpolation result at , represents the measured value of the i-th neighboring observation point, and n represents the number of neighboring observation points involved in interpolation;

[0080] The anomaly detection includes removing noise data based on isolation forest and variational autoencoder;

[0081] Anomaly detection quickly identifies obvious anomalies using the following formula:

[0082] ;

[0083] Where x represents the input feature vector, represents the path length, represents the expected value of the path length, represents the normalization factor, Indicates the anomaly score. The closer it is to 1, the higher the probability of anomaly. When it is greater than 0.6, it is considered an anomaly.

[0084] Step S13: The variational autoencoder reconstructs the error, captures complex nonlinear abnormal patterns, and detects hidden equipment degradation failures that are difficult to detect with the isolation forest. The formula used is as follows:

[0085] ;

[0086] Where, represents the loss function of the variational autoencoder, represents the encoder network, which is used to map the input to the latent space, and D() is used to reconstruct the data from the latent space. represents the approximate posterior distribution, Represents the prior distribution, KL() represents the KL divergence, which is used to constrain the latent space distribution to prevent overfitting and set the abnormality judgment threshold m. It is judged as abnormal.

[0087] By performing the above operations, multi-source heterogeneous data collection and preprocessing are adopted to clarify the data source, and the data is preprocessed through spatiotemporal alignment and anomaly detection. The variational autoencoder reconstruction error is used to capture complex abnormal patterns and detect hidden equipment failures. This solves the technical problems of insufficient anomaly detection accuracy, difficulty in detecting hidden faults, and easy omission of early equipment failures due to spatiotemporal scale mismatch, difficult data fusion, and large noise interference.

[0088] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S2, the multi-spatiotemporal scale feature extraction and fusion includes the following steps:

[0089] Step S21: time scale hierarchical processing, including equipment level processing, station level processing and regional level processing;

[0090] The device-level processing extracts transient features such as high-frequency vibration, voltage and current signals through a dynamic graph convolutional network, providing fine-grained features for device health status assessment;

[0091] The station-level processing uses a spatiotemporal attention network to capture local correlations, analyze power output and load trends, establish spatiotemporal correlations between equipment clusters, and optimize station operation strategies.

[0092] The regional-level processing uses Transformer encoders to model global dependent climate patterns and cross-regional load trends, capture atmospheric circulation, and cross-regional power dispatch to support wide-area energy coordination.

[0093] Step S22: Spatial scale fusion, aggregating the characteristics of multiple new energy sites through federated learning, generating regional level situation representation, and realizing cross-spatial scale information fusion.

[0094] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the uncertainty quantification and probability prediction are specifically performed by using a Bayesian neural network to output a confidence interval of a prediction value and Monte Carlo simulation of a random scenario to generate multiple sets of curves.

[0095] Monte Carlo sampling obtains probability output through T random forward propagation. The formula used is as follows:

[0096] ;

[0097] Where, represents the network weight of the t-th sampling, represents the network weight of the t-th sampling, Represents the forward propagation of a neural network with random weights, randomly discarding some neurons each time it is executed, and T represents the total number of sampling times. represents the expected value of the forecast, Represents the prediction variance, which is used to characterize the uncertainty of model cognition.

[0098] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the edge cloud collaborates with the real-time situation assessment, including the following steps:

[0099] Step S41: deploying a lightweight model to process high-frequency data;

[0100] Model deployment uses knowledge distillation technology to compress the model and deploy a lightweight anomaly detector. The formula used is as follows:

[0101] ;

[0102] Where, represents the knowledge distillation loss function, y represents the true label, represents the student model prediction output, represents the high-precision teacher model on the cloud, represents the edge-side student model, Used to balance task loss and knowledge transfer loss, represents the task loss function, represents KL divergence loss;

[0103] Step S42: millisecond-level anomaly detection, based on the local outlier factor algorithm of the sliding window, real-time calculation of density deviation and low-latency instantaneous anomaly detection, to meet the timeliness requirements of rapid response to power equipment failures. The formula used is as follows:

[0104] ;

[0105] Where LOF represents the local outlier factor function, p represents the data point to be detected, k represents the number of neighbors, represents the set of k nearest neighbors of p, represents the local reachability density;

[0106] Step S43: Cloud-based digital twin modeling and regular parameter updates to edge devices;

[0107] High-precision twin construction, dynamic coupling of physical models and data-driven models, the formula used is as follows:

[0108] ;

[0109] Where, represents the device mechanism equation, Represents the adaptive weight, with an online adjustment range of 0.2 to 0.8. represents the residual network;

[0110] Step S44: Communication optimization, using compressed sensing to reduce data transmission volume and lower latency;

[0111] The compression rate is dynamically adjusted based on channel quality. The formula used is as follows:

[0112] ;

[0113] Where, represents the compression ratio, and SNR represents the channel noise ratio.

[0114] By performing the above operations, edge cloud collaborative real-time situation assessment is adopted, lightweight anomaly detection models are deployed on the edge side through knowledge distillation technology, local outlier factor algorithm is used for millisecond-level anomaly detection, and updated parameters are issued through cloud-based digital twin modeling to optimize communication efficiency. This solves the technical problems of insufficient real-time performance of edge computing, insufficient consistency of cross-temporal and spatial models, large delay in massive data transmission, resulting in a greater risk of fault propagation, failure of energy efficiency optimization, and distorted state perception.

[0115] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the scheme optimization feedback is specifically to establish a multi-objective optimization framework, and the optimization objectives include safety, economy and low carbon; use multi-agent reinforcement learning to coordinate resources, generate dynamic adjustment strategies, and realize distributed free collaborative optimization; send actual operation data back to the digital twin platform to continuously calibrate the model.

[0116] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0117] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0118] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. A multi-temporal and spatial scale situational awareness method for new energy operation, characterized by: The method comprises the following steps: Step S1: Multi-source heterogeneous data collection and preprocessing: clarify the data source, preprocess the data through spatiotemporal alignment and anomaly detection, and use variational autoencoder reconstruction error to capture complex abnormal patterns and detect hidden equipment failures; Step S2: multi-temporal and spatial scale feature extraction and fusion; Step S3: uncertainty quantification and probability prediction; Step S4: Edge-cloud collaborative real-time situation assessment, deploying a lightweight anomaly detection model on the edge side through knowledge distillation technology, using a local outlier factor algorithm for millisecond-level anomaly detection, and issuing updated parameters through cloud-based digital twin modeling; Step S5: Multi-objective collaborative optimization and feedback control.

2. The multi-temporal and spatial scale situation awareness method for renewable energy operation according to claim 1 is characterized by: In step S1, the multi-source heterogeneous data collection and preprocessing includes the following steps: Step S11: data collection; Step S12: data preprocessing, including spatiotemporal alignment and anomaly detection; The spatiotemporal alignment includes using a sliding time window and spatial interpolation to unify the data granularity of different sampling frequencies; Align time and space to establish a unified space-time coordinate system; The anomaly detection includes removing noise data based on isolation forest and variational autoencoder; Step S13: The variational autoencoder reconstructs the error, captures complex nonlinear abnormal patterns, and detects hidden equipment degradation faults that are difficult to detect with the isolation forest.

3. The multi-temporal and spatial scale situation awareness method for renewable energy operation according to claim 1 is characterized by: In step S4, the edge cloud collaborates with real-time situation assessment, including the following steps: Step S41: deploying a lightweight model to process high-frequency data; Model deployment: using knowledge distillation technology to compress the model and deploy lightweight anomaly detectors; Step S42: millisecond-level anomaly detection, based on the local outlier factor algorithm of the sliding window, real-time calculation of density deviation and low-latency instantaneous anomaly detection, meeting the timeliness requirements of rapid response to power equipment failures; Step S43: Cloud-based digital twin modeling and regular parameter updates to edge devices; High-precision twin construction, dynamic coupling of physical models and data-driven models; Step S44: Communication optimization.

4. The multi-temporal and spatial scale situation awareness method for renewable energy operation according to claim 1 is characterized by: In step S2, the multi-spatiotemporal scale feature extraction and fusion includes the following steps: Step S21: time scale layering processing; Step S22: Spatial scale fusion, aggregating the characteristics of multiple new energy sites through federated learning, generating regional level situation representation, and realizing cross-spatial scale information fusion.

5. The multi-temporal and spatial scale situation awareness method for renewable energy operation according to claim 1 is characterized by: In step S3, the uncertainty quantification and probability prediction are specifically performed by using a Bayesian neural network to output a confidence interval of a predicted value, and Monte Carlo simulation of a random scenario to generate a curve.

6. The multi-temporal and spatial scale situation awareness method for renewable energy operation according to claim 1, characterized in that: In step S5, the solution optimization feedback is specifically to establish a multi-objective optimization framework; Multi-agent reinforcement learning is used to coordinate resources, generate dynamic adjustment strategies, and achieve distributed free collaborative optimization; actual operation data is transmitted back to the digital twin platform to continuously calibrate the model.

Images