A multi-temporal and spatial state main multi-source collaborative modeling method and device

Abstract

The application discloses a kind of multi-temporal and spatial state main multi-source collaborative modeling method and device, the method includes the following steps: S1, multi-temporal and spatial state data layering processing;S2, the data after processing is extracted and is associated with analysis, forms multi-temporal and spatial correlation matrix;S3, based on multi-temporal and spatial correlation matrix is carried out main multi-source collaborative modeling and exports result;S4, using output result, prediction and generation scheduling scheme are carried out.The application is based on multi-temporal and spatial state data in work, by combining multi-temporal and spatial correlation analysis and main model-collaborative submodel collaborative modeling method, integrate global and local features, realize the accurate prediction and optimization of distribution network operating state by dynamic weight adjustment, ultimately provide comprehensive support for dispatching, optimization and emergency response.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power distribution network intelligent scheduling, and in particular to a multi-time-space-state main multi-source collaborative modeling method and device. BACKGROUND

[0002] With the continuous expansion of the scale of the power distribution network and the increase of the operation complexity, the traditional power distribution network state modeling method is facing great challenges. Most of the existing methods rely on static models or simplified calculations based on linear assumptions, and cannot accurately cope with various dynamic situations that occur in the time-space change process of the power distribution network. In addition, the data sources involved in the operation process of the power distribution network are diverse, and how to efficiently integrate, analyze and make decisions is still a difficult problem to be solved.

[0003] Traditional power distribution network modeling is usually based on a single model, and the calculation time is long, which is difficult to realize real-time scheduling. In the case of complex time-space changes such as load fluctuation and equipment failure, the reaction speed and accuracy of the traditional model cannot meet the actual demand. Therefore, how to combine multi-source data and design a multi-time-space-state modeling method that can dynamically and adaptively adjust has become a problem to be solved in the field. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art.

[0005] The technical solution of the present application is: a multi-time-space-state main multi-source collaborative modeling method, comprising the following steps:

[0006] Step S1, multi-time-space-state data layering processing;

[0007] Step S2, multi-time-space-state feature extraction and correlation analysis are performed on the processed data to form a multi-time-space correlation matrix;

[0008] Step S3, main multi-source collaborative modeling is performed based on the multi-time-space correlation matrix and the results are output;

[0009] Step S4, using the output results, prediction is performed and a scheduling scheme is generated.

[0010] In step S1, further comprising multi-time-space-state data layering acquisition and preprocessing;

[0011] The multi-time-space-state data acquired by layering includes short-time-state data, medium-time-state data and long-time-state data,

[0012] The preprocessing of the data acquired by layering includes missing value filling, outlier removal and time alignment.

[0013] The short-time-state data includes real-time operation data related to load fluctuation, voltage change and power flow, and the time scale is seconds to minutes.

[0014] Medium time scale data includes device status, power dispatch records, time scale is hours to days;

[0015] Long time scale data includes historical operation records, power grid planning parameters, time scale is months to years.

[0016] In step S2, further comprising time feature extraction and spatial feature extraction;

[0017] The time feature extraction comprises:

[0018] Based on time series analysis method, the following time features are extracted from short time scale, medium time scale and long time scale data:

[0019] Short time scale features: load fluctuation frequency, voltage change trend;

[0020] Medium time scale features: device utilization rate change, fault record frequency;

[0021] Long time scale features: historical load growth rate, dispatch efficiency change;

[0022] The spatial feature extraction comprises:

[0023] Combined with power grid topology information and regional operation data, the following spatial features are extracted: node load distribution, line congestion situation and inter-regional power flow characteristics.

[0024] The multi-time-space correlation matrix A is:

[0025] A={a i,j,k},i∈V,j∈V,k∈τ

[0026] In the formula, a i,j,k : the correlation value of node i and j at time k; V: all node sets in the distribution network; τ: time range; i and j are natural numbers.

[0027] In step S3, further comprising designing a main model and a collaborative sub-model based on the multi-time-space correlation matrix, and fusing the main model and the collaborative sub-model;

[0028] The main model is used to capture global features, and the collaborative sub-model is used to capture local regional features. The main model design comprises:

[0029] Input representation: multi-time-space correlation matrix A;

[0030] Global spatial feature extraction, extracting global spatial features through graph convolution network:

[0031]

[0032] In the formula, The feature of node v in the (l+1)th layer of the graph; σ: Sigmoid activation function; N(v): The neighbor node set of node v vu : The connection weight of node v and node u in the adjacency matrix; w (l) : The weight matrix of the lth layer of graph convolution; h u (l) : The feature of node u in the lth layer; b (l) : The bias vector of the lth layer;

[0033] Global time feature extraction, using LSTM to model the time dimension:

[0034] h t = LSTM({h t-Δ , …, h t})

[0035] In the formula, h t : The hidden state of global time t, representing the feature representation of the current moment; h t-Δ : The hidden state of global time t-Δ;

[0036] Global feature fusion, fusing spatial and temporal features:

[0037]

[0038] In the formula, H global : Global feature vector; W g ,b g : Parameters of global feature fusion; The feature of node v in the lth layer of the graph;

[0039] Output global prediction, predict the target according to the global feature:

[0040]

[0041] In the formula, Global prediction value.

[0042] The collaborative sub-model design includes:

[0043] Input sub-matrix, extract the sub-matrix A sub of the local region from the multi-spatiotemporal correlation matrix A:

[0044] A sub = A[i:j,i:j,1:K]

[0045] In the formula,

[0046] i:j: Node range, representing the node subset of the local region;

[0047] 1:K: time range, representing the selected time interval;

[0048] Local spatial feature extraction, extract global spatial features through graph convolution network:

[0049]

[0050] In the formula,

[0051] c vu : the connection weight of node v and node u in the adjacency matrix;

[0052] N sub (v): the neighbor set of node v in the sub-area;

[0053] Local weight and bias;

[0054] Local time feature extraction, using LSTM to model the time dimension:

[0055] h t，sub =LSTM({h t-Δ，sub ,···,h t，sub})

[0056] In the formula, h t，sub : the hidden state of local time t; h t-Δ，sub : the hidden state of local time t-Δ;

[0057] Local feature fusion, fusion of spatial and temporal features:

[0058]

[0059] In the formula,

[0060] H local : local feature vector; W l : weight matrix of local feature fusion, b l : bias term of local feature fusion;

[0061] Output local prediction, predict target according to global feature:

[0062]

[0063] In the formula,

[0064] Local prediction value

[0065] The fusion of the main model and the collaborative sub-model includes:

[0066] According to the weight of the space-time correlation matrix, dynamically adjust the output of the main model and the collaborative sub-model :

[0067]

[0068] wherein,

[0069] w global : weight of the main model;

[0070] w local : weight of each collaborative sub-model.

[0071] Main model weight:

[0072]

[0073] Collaborative sub-model weight:

[0074]

[0075] wherein,

[0076] gamma: adjustment parameter;

[0077] sub: represents under local feature;

[0078] Var(A): variance of global matrix A;

[0079] Var(A sub ): variance of sub-matrix A sub .

[0080] A main multi-source collaborative modeling device of multi-time-space state comprises:

[0081] A hierarchical module is used for multi-time-space state data hierarchical processing;

[0082] An association module is used for multi-time-space state feature extraction and association analysis on the processed data, and forms a multi-time-space association matrix;

[0083] A modeling module is used for main multi-source collaborative modeling based on the multi-time-space association matrix and outputs a result;

[0084] An output module is used for prediction and generation of a scheduling scheme by using the output result.

[0085] In the working process, the present application is based on multi-time-space state data, combines multi-time-space association analysis and main model-collaborative sub-model collaborative modeling method, integrates global and local features, realizes accurate prediction and optimization of power distribution network operation state through dynamic weight adjustment, and finally provides comprehensive support for scheduling, optimization and emergency response. BRIEF DESCRIPTION OF DRAWINGS

[0086] In order to more clearly illustrate the technical solutions in the specific embodiments or the prior art of the present application, the drawings required to be used in the description of the specific embodiments or the prior art will be briefly introduced. In the drawings, the parts are not necessarily drawn according to the actual proportion.

[0087] Figure 1 is a method flowchart of the present application. DETAILED DESCRIPTION

[0088] As shown in the present application Figure 1 , a multi-time-space state main multi-source collaborative modeling method is provided, comprising the following steps:

[0089] Step S1, multi-time-space state data hierarchical processing;

[0090] Step S2, multi-time-space state feature extraction and correlation analysis are performed on the processed data to form a multi-time-space correlation matrix;

[0091] Step S3, main multi-source collaborative modeling is performed based on the multi-time-space correlation matrix and the result is output;

[0092] Step S4, using the output result, prediction is performed and a scheduling scheme is generated.

[0093] The present application adopts hierarchical data processing and standardization preprocessing technology to ensure data consistency and integrity. Through multi-time-space correlation analysis, a multi-time-space state feature matrix is constructed to dynamically capture the time-space changes in power grid operation.

[0094] Through feature fusion and dynamic weight optimization of the main model and the collaborative sub-model, multi-time-space data is integrated and the modeling efficiency is improved.

[0095] At the same time, the main model and the collaborative sub-model can jointly process global and local dynamic features to adapt to complex changes in different scenarios.

[0096] The present application can predict the tidal flow changes, load trend and fault possibility in real time to provide fast support for scheduling optimization.

[0097] In step S1, further comprising multi-time-space state data hierarchical acquisition and preprocessing;

[0098] The data is acquired hierarchically based on the time scale and spatial attributes of the data from the operation data of the distribution network;

[0099] The multi-time-space state data acquired hierarchically includes short-time state data, medium-time state data and long-time state data,

[0100] The acquired data is preprocessed, including missing value filling, outlier removal and time alignment.

[0101] Missing value filling: fill missing values with interpolation or historical data;

[0102] Outlier removal: remove outliers that do not conform to reality by setting thresholds;

[0103] Time alignment: unify different data time scales to ensure the synchronization of hierarchical data.

[0104] Short-term data includes load fluctuation, voltage change and real-time operation data related to power flow, with a time scale of seconds to minutes;

[0105] Medium-term data includes device status and power dispatch records, with a time scale of hours to days;

[0106] Long-term data includes historical operation records and power grid planning parameters, with a time scale of months to years.

[0107] Step S1 realizes the systematic management of multi-source data, ensuring the integrity and consistency of model input data; hierarchical processing of data with different time scales provides a basis for subsequent spatio-temporal feature extraction.

[0108] In step S2, further including time feature extraction and spatial feature extraction;

[0109] The time feature extraction includes:

[0110] Based on time series analysis method, the following time features are extracted from short-term, medium-term and long-term data:

[0111] Short-term features: load fluctuation frequency, voltage change trend;

[0112] Medium-term features: device utilization rate change, fault record frequency;

[0113] Long-term features: historical load growth rate, dispatch efficiency change;

[0114] The spatial feature extraction includes:

[0115] Combined with grid topology information and regional operation data, the following spatial features are extracted: node load distribution; line congestion; power flow characteristics between regions.

[0116] The time series analysis method includes the following steps:

[0117] (1) Constructing a space-time graph

[0118] Space: Construct the graph structure δ = (v, ε) of the distribution network, where v represents the substation and equipment, and ε represents the line connection relationship;

[0119] Time: For each time step t, generate a snapshot graph δt .

[0120] (2) Graph Neural Network Model, where graph convolution is used to extract spatial features:

[0121]

[0122] (3) Time series modeling: Combining the output of graph neural networks, LSTM is used to capture dynamic changes in the time dimension.

[0123] h t =LSTM({h t-Δ ,···,h t})

[0124] (4) Multi-task loss function, which combines the errors of spatial and temporal prediction:

[0125]

[0126] (5) Prediction and optimization: Based on spatiotemporal characteristics, output the future operating state.

[0127] in:

[0128] h t The hidden state at time t represents the feature representation at the current moment;

[0129] σ: Sigmoid activation function;

[0130] The features of node v in the diagram at layer l;

[0131] N(v): The set of neighboring nodes of node v;

[0132] c vu The connection weights between nodes v and u in the adjacency matrix;

[0133] w (l) The weight matrix of the graph convolution in the l-th layer;

[0134] b (l) : The bias vector of the l-th layer;

[0135] Δ: Time step;

[0136] These represent the errors in spatial feature prediction and temporal feature prediction, respectively.

[0137] λ1, λ2: Weight parameters in the loss function;

[0138] The predicted value output by the model.

[0139] Multi-temporal-spatial correlation analysis includes:

[0140] By analyzing the correlation between different time scales and spatial dimension data, a multi-temporal-spatial correlation matrix is constructed to describe the operation characteristics of the power distribution network at different times and spaces.

[0141] The hierarchical data (short-time, medium-time, long-time) and spatial topology data are uniformly processed and represented as input feature set X:

[0142] X={X T ,X S}

[0143] X T : Time dimension features, including short-time (such as real-time load fluctuation), medium-time (such as device utilization change), and long-time (such as historical load growth rate).

[0144] X S : Spatial dimension features, including grid node attributes (such as load) and edge attributes (such as power flow).

[0145] Time feature extraction; the change in time dimension is modeled through time series analysis to extract the following features:

[0146] F T，short : Short-time features, such as load fluctuation frequency;

[0147] F T，mid : Medium-time features, such as scheduling changes in device state;

[0148] F T，long : Long-time features, such as load growth trend;

[0149] These features are represented as time feature vector F T :

[0150] F T =[F T，short ,F T，mid ,F T，long ]

[0151] Spatial feature extraction; the change in spatial dimension is extracted through grid topology graph analysis:

[0152] Node features F S，node : Node load, voltage level, etc.

[0153] Edge features F S，edge : Power flow, grid line impedance, etc.

[0154] Spatial feature vector F S is represented as:

[0155] F S =[FS，node ,F S，edge ]

[0156] The spatiotemporal correlation feature is constructed by time series analysis and spatial feature fusion, and the spatiotemporal correlation feature F is calculated T,S :

[0157]

[0158] The cross calculation (such as inner product or convolution) of the spatiotemporal feature.

[0159] A comprehensive multi-spatiotemporal correlation matrix is constructed to describe the spatiotemporal relationship of the power distribution network:

[0160] A={a i,j,k},i∈V,j∈V,k∈τ

[0161] In the formula,

[0162] A: multi-spatiotemporal correlation matrix, describing the spatiotemporal correlation relationship between nodes;

[0163] a i,j,k : the correlation value of node i and j at time k, usually represented as correlation or weight;

[0164] V: the set of all nodes in the power distribution network;

[0165] τ: time range (such as short time state seconds to long time state annual level); i and j are natural numbers.

[0166] The calculation formula of the correlation value; the spatiotemporal correlation value a i,j,k Based on the fusion calculation of node features and time features:

[0167]

[0168] In the formula,

[0169] The correlation of node i and j in the time dimension;

[0170] The similarity of node i and j in the spatial dimension.

[0171] Step S2 extracts the key spatiotemporal features of power grid operation, and comprehensively describes the dynamic behavior of the power grid;

[0172] Establishing spatiotemporal correlation provides input features for collaborative modeling and optimization.

[0173] In step S3, further including designing the main model and the collaborative sub-model based on the multi-spatiotemporal correlation matrix, and the fusion of the main model and the collaborative sub-model;

[0174] Among them, the main model: capture global characteristics, such as the distribution of tidal flow, voltage stability, etc.

[0175] The synergy sub-model: capture local area characteristics, such as the load state of a specific node, line congestion. The main model design includes:

[0176] (1) Input representation: multi-temporal and spatial correlation matrix A;

[0177] (2) Global spatial feature extraction, extract global spatial features through graph convolution network:

[0178]

[0179] In the formula, The feature of node v in the l+1 layer; σ: Sigmoid activation function; N(v): the neighbor node set of node v; c vu : the connection weight of node v and node u in the adjacency matrix; w (l) : the weight matrix of the l-th layer graph convolution; h u (l) : the feature of node u in the l-th layer; b (l) : the bias vector of the l-th layer;

[0180] (3) Global time feature extraction, use LSTM to model the time dimension:

[0181] h t =LSTM({h t-Δ ,···,h t})

[0182] In the formula, h t : the hidden state of time t, representing the feature representation of the current moment; h t-Δ : the hidden state of global time t-Δ;

[0183] (4) Global feature fusion, fuse spatial and temporal features:

[0184]

[0185] In the formula, H global : global feature vector; W g ,b g : parameters of global feature fusion; The feature of node v in the l-th layer;

[0186] (5) Output global prediction, predict the target according to the global feature:

[0187]

[0188] In the formula, Global prediction value.

[0189] The collaborative sub-model design includes:

[0190] (1) Input sub-matrix, extract the sub-matrix A of the local area from the multi-temporal and spatial correlation matrix A sub :

[0191] A sub = A[i:j, i:j, 1:K]

[0192] In the formula,

[0193] i:j: node range, indicating the node subset of the local area;

[0194] 1:K: time range, indicating the selected time interval;

[0195] (2) Local spatial feature extraction, extract the global spatial feature through the graph convolution network:

[0196]

[0197] In the formula,

[0198] c vu : the connection weight of node v and node u in the adjacency matrix;

[0199] N sub (v): the neighbor set of node v in the sub-area;

[0200] Local weight and bias;

[0201] (3) Local time feature extraction, use LSTM to model the time dimension:

[0202] h t，sub = LSTM({h t-Δ，sub ,···,h t，sub})

[0203] In the formula, h t，sub : the hidden state of local time t; h t-Δ，sub : the hidden state of local time t-Δ;

[0204] (4) Local feature fusion, fuse spatial and temporal features:

[0205]

[0206] In the formula,

[0207] H local : local feature vector; W l : weight matrix of local feature fusion, b l: bias term of local feature fusion

[0208] (5) Output local prediction, predict target according to global feature:

[0209]

[0210] In the formula,

[0211] Local prediction value

[0212] The fusion of the main model and the collaborative sub-model includes:

[0213] According to the weight of the spatio-temporal correlation matrix, dynamically adjust the output of the main model and the collaborative sub-model

[0214]

[0215] In the formula,

[0216] w global : weight of the main model, reflecting the importance of global features;

[0217] w local : weight of each collaborative sub-model, reflecting the importance of local features.

[0218] Main model weight:

[0219]

[0220] Collaborative sub-model weight:

[0221]

[0222] In the formula,

[0223] γ: adjustment parameter, used to control the degree of volatility on the weight, when γ> 0, the weight is more inclined to smooth the feature with smaller volatility.

[0224] γ controls the sensitivity of volatility to weight.

[0225] Large value (γ≥1): volatility has more significant impact, weight distribution is sensitive to volatility.

[0226] Small value (γ≤1): weight distribution is more smooth, ignoring small fluctuations.

[0227] sub: indicates under local feature;

[0228] Var(A): variance of global matrix A

[0229] Var(A sub ): sub-matrix A subvariance of A

[0230] Var(A) and Var(A sub ): volatility

[0231] By calculating the variance of the matrix, the stability or volatility of global or local spatio-temporal features is measured.

[0232] Large variance: indicates that the feature has strong dynamic changes or uncertainties.

[0233] Small variance: indicates that the feature is relatively stable and can provide more reliable modeling basis.

[0234] exp(-γ·Var(A)): suppression of volatility;

[0235] Use a negative exponential function to suppress volatility, the greater the volatility, the smaller the impact of the weight.

[0236] When Var is very small, exp(-γ·Var(A))≈1, indicating that the feature is more reliable and gives higher weight.

[0237] Dynamic balance of main model and collaborative sub-model:

[0238] When the global feature (A) has less volatility and higher stability, the main model weight w global will be higher;

[0239] When the local feature (A sub ) has less volatility and higher stability in some areas, the corresponding sub-model weight w local will be higher.

[0240] Step S3 realizes the linkage optimization of the main model and the collaborative sub-model, improves the adaptability of modeling; ensures that the model can capture both global and local features.

[0241] The results of collaborative modeling (i.e. the comprehensive output of the main model and the collaborative sub-model, including both global analysis of power grid operation state and detailed modeling of local area, finally forming a comprehensive analysis result of global + local, static + dynamic, description + prediction) as the input of dynamic response, supporting power grid operation optimization and decision-making.

[0242] Step S4 includes: real-time state prediction, using the output results of collaborative modeling to predict the future operation state of the power grid, including: power flow changes; load distribution trends; fault possibilities.

[0243] According to the prediction results, dynamically generate scheduling schemes: optimize line power distribution; start standby power; reduce the priority of non-critical load.

[0244] The step S4 provides real-time power grid state prediction, supports accurate scheduling decision; in the case of emergency, an emergency response scheme is quickly generated to ensure power supply safety.

[0245] The application further provides a multi-time-space-state main multi-source collaborative modeling device, comprising:

[0246] A hierarchical module is configured to perform multi-time-space-state data hierarchical processing.

[0247] An association module is configured to perform multi-time-space-state feature extraction and association analysis on the processed data to form a multi-time-space association matrix.

[0248] A modeling module is configured to perform main multi-source collaborative modeling based on the multi-time-space association matrix and output a result.

[0249] An output module is configured to use the output result to perform prediction and generate a scheduling scheme.

[0250] The application is based on multi-time-space-state data, and through combination of multi-time-space association analysis and main model-collaborative sub-model collaborative modeling method, accurate prediction and optimization of the distribution network operation state are realized through dynamic weight adjustment, and finally comprehensive support is provided for scheduling, optimization and emergency response.

[0251] Although the embodiments of the application have been shown and described, it is to be understood that the application is not limited to these embodiments. It will be obvious to a person skilled in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirit of the application, and the scope of the application is defined by the appended claims and their equivalents.

Claims

1. A multi-temporal and multi-source collaborative modeling method, characterized in that, Includes the following steps: Step S1: Multi-temporal and spatial data hierarchical processing; Step S2: Extract multi-temporal and spatiotemporal features from the processed data and perform correlation analysis to form a multi-temporal and spatiotemporal correlation matrix; Step S3: Perform principal multi-source collaborative modeling based on multi-temporal correlation matrices and output the results; Step S4: Use the output results to make predictions and generate a scheduling scheme; Step S3 further includes designing a main model and a cooperating sub-model based on a multi-spatiotemporal correlation matrix, as well as the fusion of the main model and the cooperating sub-model; The main model is used to capture global features, while the collaborative sub-models are used to capture local region features. The main model design includes: Input representation: Multi-temporal correlation matrix A; Global spatial feature extraction: Extracting spatial features from the entire network using a graph convolutional network. In the formula, In the diagram, the features of node v at layer l+1; σ: Sigmoid activation function; N(v): the set of neighboring nodes of node v; c vu : The connection weights between nodes v and u in the adjacency matrix; w (l) : The weight matrix of the graph convolution in the l-th layer; h u (l) : Features of node u in layer l; b (l) : The bias vector of the l-th layer; Global temporal feature extraction, using LSTM to model the time dimension: h t =LSTM({h t-Δ ,···,h t }) In the formula, h t The hidden state at global time t represents the feature representation at the current time; h t-Δ The hidden state at global time t-Δ; Global feature fusion, fusing spatial and temporal features: In the formula, H global : Global feature vector; W g ,b g : Parameters for global feature fusion; The features of node v in the diagram at layer l; Output global prediction: Based on global features, predict the target: In the formula, Global predicted value; The collaborative sub-model design includes: Input submatrix, extract local region submatrix A from multi-temporal correlation matrix A. sub : A sub =A[i:j,i:j,1:K] In the formula, i:j: Node range, representing a subset of nodes in a local region; 1:K:Time range, representing the selected time interval; Local spatial feature extraction: Extracting spatial features of the entire network through graph convolutional networks. In the formula, c vu The connection weights between nodes v and u in the adjacency matrix; N sub (v): The set of neighbors of node v within the sub-region; Local weights and biases; Local temporal feature extraction, using LSTM to model the time dimension: h t，sub =LSTM({h t-Δ，sub ,···,h t，sub }) In the formula, h t，sub The hidden state at local time t; h t-Δ，sub The hidden state at local time t-Δ; Local feature fusion, fusing spatial and temporal features: In the formula, H local : Local feature vector; W l : Weight matrix for local feature fusion, b l Bias term for local feature fusion; Output local predictions; based on global features, predict the target: In the formula, Local predicted values; The fusion of the main model and the co-models includes: The outputs of the main model and the cooperating sub-models are dynamically adjusted based on the weights of the spatiotemporal correlation matrix. In the formula, w global : Weights of the main model; w local The weights of each collaborative sub-model; Main model weights: Collaborative sub-model weights: In the formula, γ: Adjustment parameter; sub: indicates local features; Var(A): The variance of the global matrix A; Var(A sub ): Submatrix A sub The variance.

2. The multi-temporal and multi-source collaborative modeling method according to claim 1, characterized in that, Step S1 further includes multi-temporal and spatial data hierarchical acquisition and preprocessing; The multi-temporal and spatial data acquired through stratification includes: short-temporal data, medium-temporal data, and long-temporal data. The data obtained from the stratified data is preprocessed, including missing value imputation, outlier removal, and time alignment.

3. The multi-temporal and multi-source collaborative modeling method according to claim 2, characterized in that, Short-term data includes real-time operational data related to load fluctuations, voltage changes, and power flow, with time scales ranging from seconds to minutes; Mid-temporal data includes equipment status and power dispatch records, with a time scale ranging from hours to days; Long-term data includes historical operation records and power grid planning parameters, with time scales ranging from monthly to annual.

4. The multi-temporal and multi-source collaborative modeling method according to claim 1, characterized in that, Step S2 further includes temporal feature extraction and spatial feature extraction; The time feature extraction includes: Based on time series analysis methods, the following time features are extracted from short-term, medium-term, and long-term data: Short-term characteristics: load fluctuation frequency, voltage change trend; Mid-temporal characteristics: changes in equipment utilization rate, frequency of fault records; Long-term characteristics: historical load growth rate and scheduling efficiency changes; The spatial feature extraction includes: By combining power grid topology information and regional operation data, the following spatial characteristics are extracted: node load distribution, line congestion, and inter-regional power flow characteristics.

5. The multi-temporal and multi-source collaborative modeling method according to claim 4, characterized in that, The multi-temporal correlation matrix A is: A={a i,j,k },i∈V,j∈V,k∈τ In the formula, a i,j,k : The correlation values ​​of nodes i and j at time k; V: The set of all nodes in the distribution network; τ: The time range; i and j are natural numbers.

6. A multi-temporal and multi-source collaborative modeling device, employing a multi-temporal and multi-source collaborative modeling method as described in any one of claims 1-5, characterized in that, include: The layering module is used for layered processing of multi-temporal and spatial data. The correlation module is used to extract multi-temporal and spatiotemporal features from the processed data and perform correlation analysis to form a multi-temporal and spatiotemporal correlation matrix. The modeling module is used to perform master-multi-source collaborative modeling based on multi-temporal correlation matrices and output the results; The output module is used to make predictions and generate scheduling schemes using the output results.

Images