A power dispatching method and system based on artificial intelligence

Abstract

The present invention proposes an artificial intelligence-based power dispatching method and system, the method includes collecting multimodal data, preprocessing, feature extraction and feature fusion of multimodal data to obtain a comprehensive feature vector; constructing a multimodal deep learning model to evaluate the health status; using a spatiotemporal convolutional neural network combined with a long short-term memory network to predict the operating status of power equipment; using a self-supervised learning method to identify defects and diagnose faults, and updating the equipment information and status data in the knowledge base; using a heuristic intelligent algorithm to generate the optimal maintenance plan; dynamically adjusting and optimizing the maintenance strategy according to the actual operation results and fault feedback; integrating and deploying it into the intelligent power dispatching system, and applying and verifying it in the actual distribution network. The present invention effectively improves the operating quality of distribution network automation equipment, ensures the safety and stability of the power grid, and comprehensively solves the deficiencies in the prior art.

Description

Technical Field

[0001] The present invention belongs to the field of electric power dispatching, and in particular relates to an electric power dispatching method and system based on artificial intelligence. Background Art

[0002] The power dispatching system plays a vital role in the modern power grid. Its main task is to ensure the safe and stable operation of the power grid and the efficient use of power resources. With the continuous expansion of the scale and complexity of the power grid, the traditional power dispatching method faces many challenges and bottlenecks. At present, the power dispatching system mainly relies on manual experience and rule-based decision-making methods, lacking accurate evaluation and prediction of the operating status and health of equipment. This method has the following main disadvantages:

[0003] 1. Insufficient data utilization: A large amount of multi-source data is generated during the operation of the power grid, including equipment operation data, environmental data, and historical fault data. However, traditional methods often only focus on some key data and fail to fully utilize these rich multi-source data for comprehensive analysis and decision-making, resulting in incomplete decision-making information and insufficient accuracy.

[0004] 2. Limited status assessment and prediction capabilities: Existing technologies usually evaluate equipment status based on static or single-type data, lacking dynamic monitoring and multi-dimensional analysis of equipment operating status. This makes it difficult for the power grid dispatching system to timely and accurately assess the health status of equipment and predict the operating status when facing a complex and ever-changing power grid operating environment.

[0005] 3. Low efficiency of defect identification and fault diagnosis: Traditional defect identification and fault diagnosis methods mainly rely on rule bases and expert experience, and have problems such as high false alarm rate and high missed alarm rate. At the same time, the update of the rule base depends on manual maintenance and cannot quickly adapt to changes in the power grid operating environment, resulting in serious restrictions on diagnostic efficiency and accuracy.

[0006] 4. Insufficient optimization of maintenance strategies: Most existing maintenance strategies are based on preventive maintenance or fixed-cycle maintenance plans. They lack real-time monitoring and feedback mechanisms for equipment operating status, resulting in the inability to dynamically adjust maintenance strategies based on the actual status of the equipment. This may result in excessive maintenance and waste of resources, and may also lead to equipment failures due to untimely maintenance.

[0007] 5. Low technical integration: Although technologies such as artificial intelligence and big data are gradually being applied in the field of power dispatching, existing systems often lack deep integration of these technologies and are unable to form a highly integrated and collaborative system, limiting the overall performance improvement and intelligence level of the system.

[0008] The above problems not only restrict the efficiency and reliability of the power grid dispatching system, but also increase the cost of power grid operation and maintenance. Therefore, there is an urgent need for a new power dispatching method and system based on artificial intelligence to comprehensively improve the operation quality of distribution network automation equipment and the safety and stability of the power grid. Summary of the invention

[0009] The purpose of the present invention is to design an artificial intelligence-based power dispatching method and system. First, through multimodal data collection and preprocessing, the comprehensiveness and accuracy of the data are ensured; then, through multimodal health status assessment and spatiotemporal data fusion, accurate assessment and prediction of the health status and operating status of the equipment are achieved; then, through self-supervised learning, defect identification and fault diagnosis are performed to improve the diagnostic efficiency and accuracy; the update and maintenance of the dynamic knowledge base ensures the real-time and accuracy of the data; through the heuristic intelligent algorithm to optimize the maintenance strategy, the closed-loop feedback system realizes the dynamic adjustment and continuous optimization of the maintenance strategy; finally, through the integration and application of the system, a highly intelligent and collaborative overall system is formed. Through these innovative steps, the present invention effectively improves the operating quality of distribution network automation equipment, ensures the safety and stability of the power grid, and comprehensively solves the deficiencies in the prior art.

[0010] In order to achieve the above object, in a first aspect of the present invention, a power dispatching method based on artificial intelligence is provided, the method comprising:

[0011] S1. Deploy multimodal sensors to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor;

[0012] S2. Build a multimodal deep learning model and input the preprocessed multimodal data into the multimodal deep learning model to evaluate health status;

[0013] S3, using spatiotemporal convolutional neural network combined with long short-term memory network to predict the operating status of power equipment based on health status assessment results and original multimodal data;

[0014] S4, perform defect identification and fault diagnosis through self-supervised learning methods, and update the equipment information and status data in the knowledge base;

[0015] S5. Generate the optimal maintenance plan using heuristic intelligent algorithm based on the equipment information and status data in the knowledge base;

[0016] S6. Real-time detection of the operation results of the power dispatching strategy, and dynamic adjustment and optimization of the maintenance strategy based on the actual operation results and fault feedback;

[0017] S7, integrating steps S1-S6, deploying them into the intelligent power dispatching system, and applying and verifying them in the actual distribution network;

[0018] The method of using a spatiotemporal convolutional neural network combined with a long short-term memory network includes:

[0019] Define the space-time matrix X(t,s), where t represents the time step and s represents the spatial position, which means:

[0020] X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)],

[0021] Among them, F img 、F snd 、F vib 、F tmp They represent image, sound, vibration and temperature characteristics respectively, and H represents the health status indicator;

[0022] Define the spatiotemporal convolution kernel K, and perform convolution operation on the spatiotemporal matrix X(t,s), which is expressed as follows:

[0023]

[0024] Among them, * represents the convolution operation, k represents half of the size of the convolution kernel, K(i,j) represents the weight of the spatiotemporal features in the spatiotemporal convolution kernel; X(t,s) represents the input spatiotemporal matrix, and Y(t,s) represents the output matrix after convolution processing;

[0025] Build a time series prediction model, define the input of the time series prediction model as the feature matrix Y(t,s) after convolution processing, and the output as the future equipment operation status It is expressed as follows:

[0026]

[0027] Among them, f(Y) represents the time series prediction function, which maps the feature matrix Y after convolution processing to the future health state; Δt represents the predicted time step.

[0028] Furthermore, the preprocessing includes data format standardization processing, time synchronization processing, image data preprocessing, sound data preprocessing, vibration data preprocessing and temperature data preprocessing.

[0029] Furthermore, the step S2 specifically includes:

[0030] S21. Standardize each feature, as shown below:

[0031]

[0032] Among them, μ i Representation feature F i The mean value, σ i Representation feature F i The standard deviation of i,standardized Represents the features after standardization;

[0033] S22. Calculate the health status index of each feature and observe the current health status of the device, as shown below:

[0034]

[0035] Among them, w j Representation feature F j,standardized The weight of n represents the number of features;

[0036] S23. The health status indicators of each feature are converted into health scores and normalized, which are expressed as follows:

[0037]

[0038] Among them, H min and H max Respectively represent the minimum and maximum values ​​of the health status indicators;

[0039] S24. Generate an equipment health status assessment report.

[0040] Furthermore, in step S3, the operating state of the power equipment is predicted based on the health status assessment result and the original multimodal data, the predicted operating state is compared with the actual monitoring data, and the prediction error E(t+Δt,s) is calculated, which is expressed as follows:

[0041]

[0042] Among them, E(t+Δt,s) represents the prediction error, which reflects the difference between the actual health status and the predicted result; H(t+Δt,s) represents the actual health status indicator; Represents the predicted health status.

[0043] Furthermore, the defect identification and fault diagnosis by self-supervised learning method includes:

[0044] S41. Define the input data matrix Z(t,s), which contains multimodal features and prediction results, as follows:

[0045]

[0046] in, is the predicted health status index, which represents the predicted health status index at time t and location s; F img (t,s) represents the image features at time t and position s; F snd (t,s) represents the sound features at time t and position s; F vib (t,s) represents the vibration characteristics at time t and position s; F tmp (t,s) represents the temperature characteristic at time t and position s; H(t,s) represents the health status index at time t and position s;

[0047] S42. Define the difference matrix D(t,s) to represent the difference between the actual health status H(t,s) and the predicted health status The characteristic differences are used to identify potential defects and failures, as shown below:

[0048]

[0049] Among them, H(t,s) represents the actual health status indicator, represents the predicted health status index, D(t,s) represents the difference matrix;

[0050] S43, based on the difference matrix D(t,s), set a threshold to perform abnormality detection, identify whether there is a deviation from the normal operating state, set a threshold θ, and when the difference D(t,s) exceeds the threshold, it is determined to be abnormal, as shown below:

[0051]

[0052] Among them, A(t,s) represents the anomaly matrix, which identifies the anomaly points;

[0053] S44. Define the fault mode matrix F(t,s), which contains the eigenvalues ​​and positions of all abnormal points, as follows:

[0054] FF(t,s)=[F img (t,s),F vib (t,s),F tmp (t,s),D(t,s),A(t,s)].

[0055] Furthermore, the knowledge base defines an entry K i , containing various types of data and analysis results, are expressed as follows:

[0056]

[0057] Among them, t represents time and s represents spatial position.

[0058] Furthermore, we define the update function U(Ki ,N i ), used to update the existing entry K i , which is expressed as follows:

[0059]

[0060] Among them, N i Represents a new data entry;

[0061] At the same time, the knowledge base is cleaned and the cleaning function C(K) is defined as follows:

[0062] K clean = C(K),

[0063] Among them, K is all the entries in the knowledge base;

[0064] Construct a dynamic knowledge base maintenance mechanism and define the maintenance function M(K) to regularly maintain the knowledge base, which is expressed as follows:

[0065] K maintained =M(K),

[0066] Among them, M includes data updating, cleaning and optimization.

[0067] Furthermore, the use of a heuristic intelligent algorithm to generate an optimal maintenance plan includes:

[0068] S51. Define a maintenance data matrix M, which includes a variety of maintenance related data:

[0069] M=[t,s,H(t,s),F(t,s),repair type ,repair effect ],

[0070] Among them, repair type Indicates the maintenance type, repair effect Indicates the maintenance effect;

[0071] S52, using a multi-objective optimization method, based on the maintenance data matrix, evaluate the effects of different maintenance strategies, and define a multi-objective optimization function f(p), where p represents a maintenance strategy parameter, which is expressed as follows:

[0072] f(p)=α·cost(p)+β·time(p)+γ·effect(p),

[0073] Among them, α, β, and γ represent weight parameters, cost(p) represents the maintenance cost, time(p) represents the maintenance time, and effect(p) represents the maintenance effect;

[0074] S53, use heuristic algorithm to adjust maintenance strategy parameters, conduct global search, find the parameter combination that minimizes f(p), and output the optimal maintenance strategy parameter p * .

[0075] Furthermore, the step S6 specifically includes:

[0076] S61. Define real-time monitoring data R(t), including health status assessment and fault diagnosis information, as follows:

[0077] R(t)=[H(t)F(t)E(t)],

[0078] Among them, H(t) represents the health state, F(t) represents the failure mode, and E(t) represents the maintenance effect;

[0079] S62. Based on the real-time monitoring data R(t), the physical model and evaluation algorithm are used to evaluate the equipment status. According to the evaluation results, feedback signals are generated to adjust and optimize the maintenance strategy, which is expressed as follows:

[0080] Define the state evaluation function S(R(t)) and output the evaluation result S(t):

[0081] S(t)=f(R(t)),

[0082] Where f(R) represents the state evaluation function;

[0083] S63. According to the status evaluation results, the maintenance strategy is dynamically adjusted, as shown below:

[0084] Define the maintenance strategy adjustment function A(S(t),p) to generate a new maintenance strategy parameter p new :

[0085] p new =A(S(t),p),

[0086] Where p represents the current maintenance strategy parameter;

[0087] S64, implement the adjusted maintenance strategy, monitor the maintenance effect in real time, and feed the maintenance effect back to the system to form a closed-loop control, as shown below:

[0088] Define the effect verification function V(p new ,R(t)), verify the effect of the new maintenance strategy V(t):

[0089] V(t)=g(p new ,R(t)),

[0090] Among them, g(p,R) represents the effect verification function.

[0091] In a second aspect of the present invention, there is provided an artificial intelligence-based power dispatching system, the system comprising:

[0092] A data collection unit, used to deploy multimodal sensors to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor;

[0093] The state monitoring unit is used to build a multimodal deep learning model, input the preprocessed multimodal data into the multimodal deep learning model to evaluate the health status; use the spatiotemporal convolutional neural network combined with the long short-term memory network to predict the operating status of the power equipment based on the health status evaluation results and the original multimodal data; perform defect identification and fault diagnosis through self-supervised learning methods, and update the equipment information and state data in the knowledge base;

[0094] The maintenance optimization unit is used to generate the optimal maintenance plan based on the equipment information and status data in the knowledge base using a heuristic intelligent algorithm; to detect the operation results of the power dispatching strategy in real time, and to dynamically adjust and optimize the maintenance strategy based on the actual operation results and fault feedback;

[0095] System integration unit, used to integrate the above units, deploy them into the intelligent power dispatching system, and apply and verify them in the actual distribution network;

[0096] The method of using a spatiotemporal convolutional neural network combined with a long short-term memory network includes:

[0097] Define the space-time matrix X(t,s), where t represents the time step and s represents the spatial position, which means:

[0098] X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)],

[0099] Among them, F img 、F snd 、F vib 、F tmp They represent image, sound, vibration and temperature characteristics respectively, and H represents the health status indicator;

[0100] Define the spatiotemporal convolution kernel K, and perform convolution operation on the spatiotemporal matrix X(t,s), which is expressed as follows:

[0101]

[0102] Among them, * represents the convolution operation, k represents half of the size of the convolution kernel, K(i,j) represents the weight of the spatiotemporal features in the spatiotemporal convolution kernel; X(t,s) represents the input spatiotemporal matrix, and Y(t,s) represents the output matrix after convolution processing;

[0103] Build a time series prediction model, define the input of the time series prediction model as the feature matrix Y(t,s) after convolution processing, and the output as the future equipment operation status It is expressed as follows:

[0104]

[0105] Among them, f(Y) represents the time series prediction function, which maps the feature matrix Y after convolution processing to the future health state; Δt represents the predicted time step.

[0106] The beneficial technical effects of the present invention are at least as follows:

[0107] (1) This invention uses multimodal deep learning technology to integrate and process multiple sensor data such as images, sounds, and vibrations, and uses the Transformer model to evaluate the health status of the equipment. This method overcomes the problem that traditional evaluation methods rely only on a single data source and lack evaluation accuracy, and achieves a comprehensive and accurate evaluation of the health status of the equipment.

[0108] (2) The present invention uses spatiotemporal convolutional neural networks (STCNs) combined with long short-term memory networks (LSTMs) to predict the operating status of the equipment at multiple time scales. Through spatiotemporal data fusion, the prediction accuracy of the equipment operating status is improved, solving the problem of limited prediction ability of traditional methods.

[0109] (3) The present invention introduces a self-supervised learning method. In the absence of a large amount of labeled data, defect identification and fault diagnosis are performed through a self-supervised model. A multi-task learning framework is integrated to simultaneously process defect identification and fault diagnosis tasks, thus solving the problem of low diagnostic efficiency due to reliance on manual experience and rule bases in the prior art.

[0110] (4) The present invention uses heuristic intelligent algorithms (such as genetic algorithms and ant colony algorithms) to optimize equipment maintenance strategies, and dynamically adjusts and optimizes maintenance strategies based on maintenance results through a closed-loop feedback system. This method effectively solves the problem of traditional maintenance strategies lacking real-time monitoring and dynamic adjustment mechanisms, and improves the optimization efficiency of maintenance strategies and the reliability of equipment operation.

[0111] (5) The present invention deeply integrates multimodal learning, spatiotemporal convolutional networks, self-supervised learning and heuristic intelligent algorithms to construct a highly integrated and collaborative intelligent system, which realizes comprehensive monitoring, accurate evaluation, real-time prediction, intelligent diagnosis and dynamic optimization of distribution network automation equipment, and significantly improves the overall performance and intelligence level of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0113] Figure 1 The present invention is a flow chart of an electric power dispatching method based on artificial intelligence. DETAILED DESCRIPTION

[0114] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0115] In one or more embodiments, Figure 1 As shown, a method for monitoring the status of a knife switch of the present invention is disclosed, and the method comprises steps S1-S7:

[0116] S1. Deploy multimodal sensors to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor;

[0117] Among them, the image sensor is used to capture visual information about the appearance and environment of the device. The sound sensor is used to collect sound signals generated during the operation of the device. The vibration sensor is used to detect the mechanical vibration characteristics of the device. The temperature sensor is used to monitor the temperature changes of the device.

[0118] Furthermore, the sampling frequencies of different sensors are set according to the characteristics of the device operation. For example, the sampling frequency of vibration and sound sensors can be higher (such as 10kHz) to capture rapidly changing signals, while the sampling frequency of image sensors can be lower (such as 1Hz) because visual information changes more slowly.

[0119] Processing of data format:

[0120] Image data: RGB three-channel image, size is W×H;

[0121] Sound data: time series signal, sampling rate is f s ;

[0122] Vibration data: time series signal with sampling rate f v ;

[0123] Temperature data: scalar time series with sampling rate f t .

[0124] Furthermore, the timestamp is used for data synchronization to ensure that the data of each sensor is collected under the same time reference. The collection time of each sensor data is recorded, T i ={T img ,T snd ,T vib ,T tmp}, T i Indicates a timestamp;

[0125] Specifically, the preprocessing includes data format standardization processing, time synchronization processing, image data preprocessing, sound data preprocessing, vibration data preprocessing and temperature data preprocessing.

[0126] Wherein, the image data is preprocessed:

[0127] Convert to grayscale image to reduce computational complexity:

[0128] I gray =0.2989×I R +0.5870×I G +0.1140×I B (1)

[0129] Among them, I R ,I G ,I B are the pixel values ​​of RGB channels respectively;

[0130] Image normalization, normalize the pixel values ​​to the [0,1] interval:

[0131]

[0132] The sound data preprocessing:

[0133] Denoising, using filters to remove high-frequency noise:

[0134] S filtered =ButterworthFilter(S,f c ,order) (3),

[0135] Among them, S is the original sound signal, f cis the cutoff frequency of the filter, and order is the order of the filter;

[0136] Calculate Mel-frequency cepstral coefficients (MFCC):

[0137] MFCC=MelFrequencyCepstralCoefficients(S filtered ) (4),

[0138] Among them, MFCC is used to extract the characteristics of sound signals.

[0139] The vibration data preprocessing:

[0140] Denoising, using wavelet transform to remove noise from vibration signals:

[0141] V denoised = WaveletDenoising(V) (5),

[0142] Wherein, V is the original vibration signal;

[0143] Calculate the frequency domain features and use fast Fourier transform (FFT) to convert the time domain signal into the frequency domain signal: V fft =FFT(V denoised ) (6).

[0144] The temperature data preprocessing:

[0145] Temperature data smoothing, using moving average method to eliminate short-term fluctuations in temperature data: T smoothed =MovingAverage(T,window_size) (7),

[0146] Among them, T is the original temperature data, and window_size is the size of the moving window.

[0147] The image feature extraction:

[0148] Use image processing algorithms to extract key feature points (such as edge detection, SIFT, SURF): F img =ExtractKeyPoints(I norm ) (8);

[0149] Sound feature extraction:

[0150] Extract MFCC features:

[0151] F snd =MFCC (9);

[0152] Vibration feature extraction:

[0153] Extract frequency domain features (such as spectrum peaks, frequency centers):

[0154] F vib =ExtractFrequencyFeatures(V fft ) (10);

[0155] Temperature feature extraction:

[0156] Calculate the statistical characteristics of temperature variation (such as mean, variance):

[0157] F tmp ={mean(T smoothed ),var(T smoothed )} (11);

[0158] Data fusion and storage

[0159] Data Fusion:

[0160] The features of all sensors are integrated to form a comprehensive feature vector:

[0161] F combined =[F img ,F snd ,F vib ,F tmp ] (12);

[0162] Furthermore, the fused feature vector is stored in a dynamic knowledge base for use in subsequent steps.

[0163] S2. Construct a multimodal deep learning model and input the preprocessed multimodal data into the multimodal deep learning model to evaluate the health status, including:

[0164] S21, standardize each feature, where for each feature F i (including image features F img 、Sound characteristics F snd , vibration characteristics F vib 、Temperature characteristics F tmp ), and is standardized as follows:

[0165]

[0166] Among them, μ i Representation feature F i The mean value, σ i Representation feature F i The standard deviation of i,standardized Represents the standardized feature; the standardized feature value F i,standardizedIt has zero mean and unit variance, so that different features can be compared on the same scale.

[0167] S22. Calculate the health status index of each feature, observe the current health status of the device, and define the health status index H i is the weighted sum of the standardized eigenvalues, expressed as follows:

[0168]

[0169] Among them, w j Representation feature F j,standardized The weight of n represents the number of features; the health status index H i Combining the information of multimodal features, the weight w j It can be adjusted based on the importance of the features to accurately reflect the health status of the device.

[0170] S23. The health status indicators of each feature are converted into health scores and normalized, which are expressed as follows:

[0171]

[0172] Among them, H min and H max They represent the minimum and maximum values ​​of the health status indicator respectively; the health score S ranges from 0 to 100, reflecting the health status of the device. The higher the value, the better the health status.

[0173] S24. Generate an equipment health status assessment report, including health status indicators, health scores and key feature descriptions.

[0174] S3, using spatiotemporal convolutional neural network combined with long short-term memory network to predict the operating status of power equipment based on health status assessment results and original multimodal data;

[0175] The multimodal data of step S1 and the health status assessment result of step S2 are integrated to ensure the consistency of time and space information.

[0176] Define the spatiotemporal matrix X(t,s), where t represents the time step and s represents the spatial position. The integrated spatiotemporal matrix contains all standardized multimodal features and health status assessment results, as shown below:

[0177] X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)] (16),

[0178] Among them, F img 、F snd 、F vib 、F tmp are image, sound, vibration and temperature features, respectively, and H is the health status indicator. The spatiotemporal matrix X(t,s) integrates all features and evaluation results at different time steps and spatial positions for subsequent processing.

[0179] Furthermore, a custom spatiotemporal convolution algorithm is used to process the spatiotemporal matrix to capture the correlation in the time and space dimensions. Define the spatiotemporal convolution kernel K and perform convolution operation on the spatiotemporal matrix X:

[0180]

[0181] Where * represents the convolution operation, k is half the size of the convolution kernel, and K(i,j) is the weight of the spatiotemporal features in the spatiotemporal convolution kernel. X(t,s) is the input spatiotemporal matrix. Y(t,s) is the output matrix after convolution. The convolution operation can extract local features in the spatiotemporal matrix and capture the correlation of features in time and space dimensions.

[0182] Based on the features processed by spatiotemporal convolution, a custom time series prediction model is built to predict the future equipment operation status. The input of the prediction model is defined as the feature matrix Y(t,s) after convolution processing, and the output is the future equipment operation status.

[0183]

[0184] Where f(·) is the time series prediction function, Δt is the prediction time step, To predict the future health status. The prediction model can predict the future operating status of the equipment at different spatial locations based on the current spatiotemporal feature matrix;

[0185] According to the predicted operating status, the health status of the equipment is evaluated and the status information is updated. The predicted operating status is compared with the actual monitoring data to calculate the prediction error E(t+Δt,s):

[0186]

[0187] Among them, E(t+Δt,s) represents the prediction error, which reflects the difference between the actual health status and the predicted result. H(t+Δt,s) represents the actual health status indicator. represents the predicted health status;

[0188] The prediction error reflects the accuracy of the prediction results and can be used to optimize and adjust the model.

[0189] S4, performing defect identification and fault diagnosis through self-supervised learning methods, and updating the equipment information and status data in the knowledge base, integrating the multimodal data features obtained from step S1 and the spatiotemporal prediction results of step S3, specifically including:

[0190] S41. Define the input data matrix Z(t,s), which contains multimodal features and prediction results, as follows:

[0191]

[0192] in, is the predicted health status index, which represents the predicted health status index at time t and location s; F img (t,s) represents the image features at time t and position s; F snd (t,s) represents the sound features at time t and position s; F vib (t,s) represents the vibration characteristics at time t and position s; F tmp (t,s) represents the temperature characteristics at time t and position s; H(t,s) represents the health status index at time t and position s. The input data matrix Z(t,s) integrates various features and prediction results to provide rich input information for defect identification and fault diagnosis.

[0193] S42. Calculate the difference between the actual health state and the predicted state using multimodal features and prediction results to identify potential defects and failures. Define the difference matrix D(t,s) to represent the difference between the actual health state H(t,s) and the predicted health state The characteristic differences are used to identify potential defects and failures, as shown below:

[0194]

[0195] Among them, H(t,s) represents the actual health status indicator, represents the predicted health status indicator, and D(t,s) represents the difference matrix; the difference matrix D(t,s) can reveal the deviation between the actual operating status of the equipment and the predicted status, and identify potential defects and failures.

[0196] S43, based on the difference matrix D(t,s), set a threshold to perform abnormality detection, identify whether there is a deviation from the normal operating state, set a threshold θ, and when the difference D(t,s) exceeds the threshold, it is determined to be abnormal, as shown below:

[0197]

[0198] Among them, A(t,s) represents the anomaly matrix, which identifies the anomalies; the anomaly matrix A(t,s) identifies all anomalies that exceed the threshold, which is convenient for subsequent analysis and processing.

[0199] S44. According to the abnormality detection results, a fault mode analysis is performed to identify the specific fault type and location, and a fault mode matrix F(t,s) is defined, which includes the eigenvalues ​​and locations of all abnormal points, as shown below:

[0200] F(t,s)=[F img (t,s),F vib (t,s),F tmp (t,s),D(t,s),A(t,s)] (23),

[0201] Among them, the failure mode matrix F(t,s) provides detailed information of all abnormal points, which is convenient for failure mode analysis and specific diagnosis;

[0202] Specifically, the device information and status data in the updating knowledge base includes:

[0203] In a dynamic knowledge base, define the knowledge base entry K i , containing various data and analysis results:

[0204]

[0205] Where t is time, s is spatial position, and the meanings of other symbols are the same as above. i It includes multimodal features, evaluation and prediction results, differential analysis and failure modes to form a comprehensive data entry;

[0206] Dynamically update knowledge base entries based on new data input to ensure the timeliness and accuracy of data.

[0207] Furthermore, we define the update function U(K i ,N i ), used to update the existing entry K i :

[0208]

[0209] Among them, N i is the new data entry. The update function U converts the new data entry N i Merge into existing entry K i Generate updated knowledge base entries

[0210] In order to maintain the quality of the knowledge base, the data needs to be cleaned and filtered regularly to remove redundant and outdated data;

[0211] Define the cleaning function C(K) to clean the knowledge base entries:

[0212] K clean = C(K) (26),

[0213] Where K is all the entries in the knowledge base. The cleaning function C checks the validity and relevance of each entry, removes unnecessary or duplicate entries, and ensures the high quality and timeliness of the knowledge base data.

[0214] Build a dynamic maintenance mechanism to ensure that the knowledge base can be continuously updated and optimized. Define the maintenance function M(K) to regularly maintain the knowledge base:

[0215] K maintained =M(K) (27),

[0216] Among them, M includes functions such as data updating, cleaning and optimization. The maintenance function M ensures that the knowledge base can remain accurate and efficient during long-term operation.

[0217] S5. Generate the optimal maintenance plan based on the equipment information and status data in the knowledge base using heuristic intelligent algorithms, including:

[0218] S51. Analyze the maintenance-related data in the dynamic knowledge base in the previous step to identify the relationship between historical maintenance strategies and actual results. The data to be analyzed include equipment operating status, fault records, maintenance records, and equipment status after maintenance. Define the maintenance data matrix M, which contains a variety of maintenance-related data:

[0219] M=[t,s,H(t,s),F(t,s),repair type ,repair effect ] (28),

[0220] Among them, repair type Indicates the maintenance type, repair effect Indicates the maintenance effect; the maintenance data matrix M includes the health status, failure mode, maintenance type and maintenance effect of the equipment, providing data support for subsequent maintenance strategy optimization.

[0221] S52. According to the maintenance data matrix, the effects of different maintenance strategies are evaluated. A multi-objective optimization method is used to evaluate the effects of different maintenance strategies according to the maintenance data matrix, and a multi-objective optimization function f(p) is defined, where p represents a maintenance strategy parameter, which is expressed as follows:

[0222] f(p)=α·cost(p)+β·time(p)+γ·effect(p)(29),

[0223] Among them, α, β, and γ represent weight parameters, cost(p) represents the maintenance cost, time(p) represents the maintenance time, and effect(p) represents the maintenance effect. The optimization function f(p) comprehensively considers various factors of the maintenance strategy, which is convenient for evaluating the overall effect of different maintenance strategies.

[0224] S53. Based on the multi-objective optimization results, optimize the maintenance strategy parameters and determine the optimal maintenance strategy. Use heuristic algorithms (such as genetic algorithms, simulated annealing algorithms, etc.) to perform global search on the basis of multi-objective optimization to find the optimal maintenance strategy parameters. Use heuristic algorithms to adjust the maintenance strategy parameters, perform global search, find the parameter combination that minimizes f(p), and output the optimal maintenance strategy parameters p * .

[0225] S6. Real-time detection of the operation results of the power dispatch strategy, dynamic adjustment and optimization of the maintenance strategy based on the actual operation results and fault feedback, including:

[0226] S61. Use the deployed multimodal sensor network to collect equipment operation status data in real time, including health status assessment, fault diagnosis and maintenance effect. These data will serve as input to the closed-loop feedback system. Define real-time monitoring data R(t), including health status assessment and fault diagnosis information, as follows:

[0227] R(t)=[H(t),F(t),E(t)](30),

[0228] Among them, H(t) represents the health status, F(t) represents the failure mode, and E(t) represents the maintenance effect; the real-time monitoring data R(t) provides the latest information on the equipment operation status and maintenance effect, providing basic data for the feedback system.

[0229] S62. Based on the real-time monitoring data R(t), the physical model and evaluation algorithm are used to evaluate the equipment status. According to the evaluation results, feedback signals are generated to adjust and optimize the maintenance strategy, which is expressed as follows:

[0230] Define the state evaluation function S(R(t)) and output the evaluation result S(t):

[0231] S(t)=f(R(t))(31),

[0232] Where f(R) represents the state evaluation function; the state evaluation function S(t) determines the current state of the equipment and the maintenance strategy that needs to be adjusted by analyzing real-time data.

[0233] S63. Dynamically adjust the maintenance strategy based on the status assessment results. Ensure that the maintenance strategy can respond to changes in equipment status in real time and optimize equipment operation and maintenance, as shown below:

[0234] Define the maintenance strategy adjustment function A(S(t),p) to generate a new maintenance strategy parameter p new :

[0235] p new =A(S(t),p)(32),

[0236] Where p represents the current maintenance strategy parameter;

[0237] S64, implement the adjusted maintenance strategy, monitor the maintenance effect in real time, and feed the maintenance effect back to the system to form a closed-loop control, as shown below:

[0238] Define the effect verification function V(p new ,R(t)), verify the effect of the new maintenance strategy V(t):

[0239] V(t)=g(p new ,R(t))(33),

[0240] Among them, g(p,R) represents the effect verification function.

[0241] S7. Integrate steps S1-S6, deploy them into the intelligent power dispatching system, and apply and verify them in the actual distribution network, including:

[0242] Integrate data preprocessing, feature extraction and data storage functions to ensure that multimodal data can be effectively processed and managed;

[0243] Deploy algorithms for health status assessment and fault diagnosis to ensure that equipment status can be analyzed and assessment results generated in real time;

[0244] Integrate algorithms for maintenance strategy optimization and dynamically adjust maintenance strategies based on real-time evaluation results;

[0245] Develop a user-friendly interface to facilitate operators to monitor equipment status, view diagnostic results, and adjust maintenance strategies;

[0246] Furthermore, the integrated system is fully tested and verified to ensure that its functions and performance meet the expected requirements, including:

[0247] Each software module is tested independently to ensure correct functionality;

[0248] Test the overall integration of hardware and software to ensure that the parts work together seamlessly;

[0249] Evaluate the system's real-time performance, data processing capabilities, and stability to ensure that the system can run stably in actual applications;

[0250] Invite actual operators to use the system, collect feedback and make optimizations to ensure the practicality and ease of use of the system.

[0251] Furthermore, the integrated system is deployed to the actual application site and continuously optimized and improved based on the feedback from the site, including:

[0252] Deploy system hardware and software to the actual application site to ensure that all equipment operates normally;

[0253] Monitor system operation in real time, record equipment status data and diagnostic results;

[0254] Collect feedback from field operators to identify problems and improvement needs in system operation;

[0255] Based on on-site feedback, the system is continuously optimized and improved to enhance its reliability and practicality.

[0256] In a second aspect of the present invention, a knife switch state monitoring system is provided, comprising the following units:

[0257] The data collection unit 101 is used to deploy a multimodal sensor to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor;

[0258] The state monitoring unit 102 is used to construct a multimodal deep learning model, input the preprocessed multimodal data into the multimodal deep learning model to evaluate the health status; use a spatiotemporal convolutional neural network combined with a long short-term memory network to predict the operating status of the power equipment based on the health status evaluation results and the original multimodal data; perform defect identification and fault diagnosis through a self-supervised learning method, and update the equipment information and state data in the knowledge base;

[0259] The maintenance optimization unit 103 is used to generate the optimal maintenance plan based on the equipment information and status data in the knowledge base using a heuristic intelligent algorithm; detect the operation results of the power dispatching strategy in real time, and dynamically adjust and optimize the maintenance strategy based on the actual operation results and fault feedback;

[0260] The system integration unit 104 is used to integrate the above units, deploy them into the intelligent power dispatching system, and apply and verify them in the actual distribution network;

[0261] The method of using a spatiotemporal convolutional neural network combined with a long short-term memory network includes:

[0262] Define the space-time matrix X(t,s), where t represents the time step and s represents the spatial position, which means:

[0263] X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)] (16),

[0264] Among them, F img 、F snd 、F vib 、F tmp They represent image, sound, vibration and temperature characteristics respectively, and H represents the health status indicator;

[0265] Define the spatiotemporal convolution kernel K, and perform convolution operation on the spatiotemporal matrix X(t,s), which is expressed as follows:

[0266]

[0267] Among them, * represents the convolution operation, k represents half of the size of the convolution kernel, K(i,j) represents the weight of the spatiotemporal features in the spatiotemporal convolution kernel; X(t,s) represents the input spatiotemporal matrix, and Y(t,s) represents the output matrix after convolution processing;

[0268] Build a time series prediction model, define the input of the time series prediction model as the feature matrix Y(t,s) after convolution processing, and the output as the future equipment operation status It is expressed as follows:

[0269]

[0270] Among them, f(Y) represents the time series prediction function, which maps the feature matrix Y after convolution processing to the future health state; Δt represents the predicted time step.

[0271] In summary, the present invention systematically solves many defects of traditional power dispatching methods through a series of innovative steps. First, through multimodal data collection and preprocessing, the comprehensiveness and accuracy of the data are ensured; then, through multimodal health status assessment and spatiotemporal data fusion, accurate assessment and prediction of equipment health status and operating status are achieved; then, through self-supervised learning, defect identification and fault diagnosis are performed to improve diagnostic efficiency and accuracy; the update and maintenance of the dynamic knowledge base ensures the real-time and accuracy of the data; through the heuristic intelligent algorithm to optimize the maintenance strategy, the closed-loop feedback system realizes the dynamic adjustment and continuous optimization of the maintenance strategy; finally, through the integration and application of the system, a highly intelligent and collaborative overall system is formed. Through these innovative steps, the present invention effectively improves the operating quality of distribution network automation equipment, ensures the safety and stability of the power grid, and comprehensively solves the deficiencies in the prior art.

[0272] The above disclosures are only some preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention are still within the scope of the invention.

Claims

1. A power dispatching method based on artificial intelligence, characterized in that: The method comprises: S1. Deploy multimodal sensors to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor; S2. Build a multimodal deep learning model and input the preprocessed multimodal data into the multimodal deep learning model to evaluate health status; S3, using spatiotemporal convolutional neural network combined with long short-term memory network to predict the operating status of power equipment based on health status assessment results and original multimodal data; S4, perform defect identification and fault diagnosis through self-supervised learning methods, and update the equipment information and status data in the knowledge base; S5. Generate the optimal maintenance plan using heuristic intelligent algorithm based on the equipment information and status data in the knowledge base; S6. Real-time detection of the operation results of the power dispatching strategy, and dynamic adjustment and optimization of the maintenance strategy based on the actual operation results and fault feedback; S7, integrating steps S1-S6, deploying them into the intelligent power dispatching system, and applying and verifying them in the actual distribution network; The method of using a spatiotemporal convolutional neural network combined with a long short-term memory network includes: Define the space-time matrix X(t,s), where t represents the time step and s represents the spatial position, which means: X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)], Among them, F img 、F snd 、F vib 、F tmp They represent image, sound, vibration and temperature characteristics respectively, and H represents the health status indicator; Define the spatiotemporal convolution kernel K, and perform convolution operation on the spatiotemporal matrix X(t,s), which is expressed as follows: Among them, * represents the convolution operation, k represents half of the size of the convolution kernel, K(i,j) represents the weight of the spatiotemporal features in the spatiotemporal convolution kernel; X(t,s) represents the input spatiotemporal matrix, and Y(t,s) represents the output matrix after convolution processing; Build a time series prediction model, define the input of the time series prediction model as the feature matrix Y(t,s) after convolution processing, and the output as the future equipment operation status It is expressed as follows: Where f(Y) represents the time series prediction function, which maps the convolutional feature matrix Y(t,s) to the future health state; Δt represents the predicted time step; The defect identification and fault diagnosis by self-supervised learning method includes: S41. Define the input data matrix Z(t,s), which contains multimodal features and prediction results, as follows: in, is the predicted health status index, which represents the predicted health status index at time t and location s; F img (t,s) represents the image features at time t and position s; F snd (t,s) represents the sound features at time t and position s; F vib (t,s) represents the vibration characteristics at time t and position s; F tmp (t,s) represents the temperature characteristic at time t and position s; H(t,s) represents the health status index at time t and position s; S42. Define the difference matrix D(t,s) to represent the difference between the actual health status H(t,s) and the predicted health status The characteristic differences are used to identify potential defects and failures, as shown below: Among them, H(t,s) represents the actual health status indicator, represents the predicted health status index, D(t,s) represents the difference matrix; S43, based on the difference matrix D(t,s), set a threshold to perform abnormality detection, identify whether there is a deviation from the normal operating state, set a threshold θ, and when the difference D(t,s) exceeds the threshold, it is determined to be abnormal, as shown below: Among them, A(t,s) represents the anomaly matrix, which identifies the anomaly points; S44. Define the fault mode matrix F(t,s), which contains the eigenvalues ​​and positions of all abnormal points, as follows: F(t,s)=[F img (t,s),F vib (t,s),F tmp (t,s),D(t,s),A(t,s)]。 2. The power dispatching method based on artificial intelligence according to claim 1, characterized in that: The preprocessing includes data format standardization processing, time synchronization processing and preprocessing of this type, and the preprocessing of this type includes image data preprocessing, sound data preprocessing, vibration data preprocessing and temperature data preprocessing.

3. The power dispatching method based on artificial intelligence according to claim 1 is characterized in that: The step S2 specifically includes: S21. Standardize each feature, as shown below: Among them, μ i Representation feature F i The mean value, σ i Representation feature F i The standard deviation of i,standardized Represents the features after standardization; S22. Calculate the health status index of each feature and observe the current health status of the device, as shown below: Among them, w j Representation feature F j,standardized The weight of n represents the number of features; S23. The health status indicators of each feature are converted into health scores and normalized, which are expressed as follows: Among them, H min and H max Respectively represent the minimum and maximum values ​​of the health status indicators; S24. Generate an equipment health status assessment report.

4. The power dispatching method based on artificial intelligence according to claim 1 is characterized in that: In step S3, the operating state of the power equipment is predicted based on the health status assessment result and the original multimodal data, the predicted operating state is compared with the actual monitoring data, and the prediction error E(t+Δt,s) is calculated, which is expressed as follows: Among them, E(t+Δt,s) represents the prediction error, which reflects the difference between the actual health status and the predicted result; H(t+Δt,s) represents the actual health status indicator; Represents the predicted health status.

5. The power dispatching method based on artificial intelligence according to claim 1 is characterized in that: The knowledge base defines an entry K i , containing various types of data and analysis results, are expressed as follows: Among them, s represents the spatial position.

6. The power dispatching method based on artificial intelligence according to claim 5 is characterized in that: Define the update function U(K i ,N i ), used to update the existing entry K i , which is expressed as follows: Among them, N i Represents a new data entry; At the same time, the knowledge base is cleaned and the cleaning function C(K) is defined as follows: K clean =C(K), Among them, K is all the entries in the knowledge base; Construct a dynamic knowledge base maintenance mechanism and define the maintenance function M(K) to regularly maintain the knowledge base, which is expressed as follows: K maintained =M(K), Among them, M(K) includes data updating, cleaning and optimization.

7. The power dispatching method based on artificial intelligence according to claim 1 is characterized in that: The method of generating the optimal maintenance plan by using a heuristic intelligent algorithm includes: S51. Define a maintenance data matrix M, which includes a variety of maintenance related data: M=[t,s,H(t,s),F(t,s),repair type ,repair effect ], Among them, repair type Indicates the maintenance type, repair effect Indicates the maintenance effect; S52, using a multi-objective optimization method, based on the maintenance data matrix, evaluate the effects of different maintenance strategies, and define a multi-objective optimization function f(p), where p represents a maintenance strategy parameter, which is expressed as follows: f(p)=α·cost(p)+β·time(p)+γ·effect(p), Among them, α, β, and γ represent weight parameters, cost(p) represents the maintenance cost, time(p) represents the maintenance time, and effect(p) represents the maintenance effect; S53, use heuristic algorithm to adjust maintenance strategy parameters, conduct global search, find the parameter combination that minimizes f(p), and output the optimal maintenance strategy parameter p * .

8. The power dispatching method based on artificial intelligence according to claim 1, characterized in that: The step S6 specifically includes: S61. Define real-time monitoring data R(t), including health status assessment and fault diagnosis information, as follows: R(t)=[H(t),F(t),E(t)], Among them, H(t) represents the health state, F(t) represents the failure mode, and E(t) represents the maintenance effect; S62. Based on the real-time monitoring data R(t), the physical model and evaluation algorithm are used to evaluate the equipment status. According to the evaluation results, feedback signals are generated to adjust and optimize the maintenance strategy, which is expressed as follows: Define the state evaluation function S(R(t)) and output the evaluation result S(t): S(t)=f(R(t)), Where f(R) represents the state evaluation function; S63. According to the status evaluation results, the maintenance strategy is dynamically adjusted, as shown below: Define the maintenance strategy adjustment function A(S(t),p) to generate a new maintenance strategy parameter p new : p new =A(S(t),p); S64, implement the adjusted maintenance strategy, monitor the maintenance effect in real time, and feed the maintenance effect back to the system to form a closed-loop control, as shown below: Define the effect verification function V(p new ,R(t)), verify the effect of the new maintenance strategy V(t): V(t)=g(p new ,R(t)), Among them, g(p,R) represents the effect verification function.

9. An artificial intelligence-based power dispatching system, characterized in that: The system comprises: A data collection unit, used to deploy multimodal sensors to collect multimodal data, perform preprocessing, feature extraction and feature fusion on the multimodal data, and obtain a comprehensive feature vector; wherein the multimodal sensor includes an image sensor, a sound sensor, a vibration sensor and a temperature sensor; The power dispatch detection unit is used to build a multimodal deep learning model, input the preprocessed multimodal data into the multimodal deep learning model to evaluate the health status; use the spatiotemporal convolutional neural network combined with the long short-term memory network to predict the operating status of the power equipment based on the health status evaluation results and the original multimodal data; perform defect identification and fault diagnosis through self-supervised learning methods, and update the equipment information and status data in the knowledge base; The maintenance optimization unit is used to generate the optimal maintenance plan based on the equipment information and status data in the knowledge base using a heuristic intelligent algorithm; to detect the operation results of the power dispatching strategy in real time, and to dynamically adjust and optimize the maintenance strategy based on the actual operation results and fault feedback; System integration unit, used to integrate the data collection unit, power dispatch detection unit and maintenance optimization unit, deploy them into the intelligent power dispatch system, and apply and verify them in the actual distribution network; The method of using a spatiotemporal convolutional neural network combined with a long short-term memory network includes: Define the space-time matrix X(t,s), where t represents the time step and s represents the spatial position, which means: X(t,s)=[F img (t,s),F snd (t,s),F vib (t,s),F tmp (t,s),H(t,s)], Among them, F img 、F snd 、F vib 、F tmp They represent image, sound, vibration and temperature characteristics respectively, and H represents the health status indicator; Define the spatiotemporal convolution kernel K, and perform convolution operation on the spatiotemporal matrix X(t,s), which is expressed as follows: Among them, * represents the convolution operation, k represents half of the size of the convolution kernel, K(i,j) represents the weight of the spatiotemporal features in the spatiotemporal convolution kernel; X(t,s) represents the input spatiotemporal matrix, and Y(t,s) represents the output matrix after convolution processing; Build a time series prediction model, define the input of the time series prediction model as the feature matrix Y(t,s) after convolution processing, and the output as the future equipment operation status It is expressed as follows: Where f(Y) represents the time series prediction function, which maps the convolutional feature matrix Y(t,s) to the future health state; Δt represents the predicted time step; The defect identification and fault diagnosis by self-supervised learning method includes: S41. Define the input data matrix Z(t,s), which contains multimodal features and prediction results, as follows: in, is the predicted health status index, which represents the predicted health status index at time t and location s; F img (t,s) represents the image features at time t and position s; F snd (t,s) represents the sound features at time t and position s; F vib (t,s) represents the vibration characteristics at time t and position s; F tmp (t,s) represents the temperature characteristic at time t and position s; H(t,s) represents the health status index at time t and position s; S42. Define the difference matrix D(t,s) to represent the difference between the actual health status H(t,s) and the predicted health status The characteristic differences are used to identify potential defects and failures, as shown below: Among them, H(t,s) represents the actual health status indicator, represents the predicted health status index, D(t,s) represents the difference matrix; S43, based on the difference matrix D(t,s), set a threshold to perform abnormality detection, identify whether there is a deviation from the normal operating state, set a threshold θ, and when the difference D(t,s) exceeds the threshold, it is determined to be abnormal, as shown below: Among them, A(t,s) represents the anomaly matrix, which identifies the anomaly points; S44. Define the fault mode matrix F(t,s), which contains the eigenvalues ​​and positions of all abnormal points, as follows: F(t,s) = [F img (t,s),F vib (t,s),F tmp (t,s),D(t,s),A(t,s)].

Images