Ies situation awareness method and system based on integration of graph model library and case reasoning

Abstract

An IES (Environmentally Integrated System) situational awareness method and system based on integrated graph-model-database and case-based reasoning includes: collecting various energy data from sources, grids, loads, and storage within an integrated energy system to construct an original dataset; extracting key features from the original dataset and, combined with integrated graph-model-database modeling technology, constructing a process network model of the integrated energy system; publishing the model to an integrated management and control system, storing the operational status of the networked integrated energy system in the form of case profiles; using the integrated management and control system as a carrier, identifying future state characteristics of the networked integrated energy system and performing physical constraint verification to obtain target state characteristics; employing a case-based reasoning method, searching through a set of historical case profiles for historical cases with a similarity greater than a preset similarity threshold to the target state characteristics as reference cases; and using situational awareness techniques to achieve system situation prediction. This invention can improve the comprehensiveness of situational awareness in integrated energy systems.

Description

Technical Field

[0001] This invention belongs to the field of integrated energy systems, and in particular relates to an IES situational awareness method and system based on integrated graph model library and case reasoning. Background Technology

[0002] As the core carrier of the energy internet, the Integrated Energy System (IES) exhibits complex characteristics of "multi-energy coupling, spatiotemporal dynamics, and topological variability" due to the continuous increase in the penetration rate of renewable energy sources such as wind power and photovoltaics, as well as the large-scale access of distributed energy and energy storage devices. However, the development of its situational awareness technology lags behind the increase in system complexity. Among the existing IES situational awareness technologies, there are already various specific implementation schemes, but the following limitations still exist:

[0003] Chinese patent application CN115022187B focuses on coupled systems of specific energy types. It collects measurement data from the electricity-gas system, calculates system deviations after state estimation filtering, constructs feature images based on the temporal changes of these deviations, and then predicts future operating trends using a convolutional neural network model. While this approach can mine historical data information from electricity-gas coupling, it only targets electricity and gas, failing to cover multiple energy types such as heat and storage. It lacks sufficient multi-source data fusion across the entire "source-grid-load-storage" chain and relies on a single convolutional neural network algorithm, lacking a mechanism for reusing experience from similar historical cases. When the system experiences sudden load increases or equipment failures, the prediction error increases significantly, making it difficult to meet the full-scenario situational awareness needs of complex energy systems. Chinese patent application CN115063058B adopts a layered architecture to achieve energy situational awareness in industrial parks. It integrates equipment operation information through a perception layer, judges the current state through an understanding layer, and predicts future events through a prediction layer. It also employs a decentralized model with users as nodes for energy dispatching. While the scheme has made some improvements in timeliness, the modeling process does not fully consider the systemic nature of energy conversion across links (such as the synergistic coupling of electricity, heat and gas), making it difficult to depict the global synergistic characteristics of "material flow-energy flow-information flow". Furthermore, the decentralized model may lead to insufficient adaptability to changes in the overall system topology, and the model is prone to failure under complex conditions such as extreme weather, thus failing to meet the precise control requirements of large-scale integrated energy systems.

[0004] This invention, by constructing a collaborative model of the entire "source-grid-load-storage" process and combining case-based reasoning and situational awareness technologies, can overcome limitations in areas such as insufficient coverage of multiple energy types, data fusion across all processes, adaptability to complex operating conditions, and prediction accuracy, thereby improving the comprehensiveness, adaptability, and accuracy of situational awareness in integrated energy systems. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an IES situational awareness method and system based on integrated graph model library and case reasoning. This invention is based on a comprehensive energy system network and achieves iterative interactive optimization of the perception layer, understanding layer and prediction layer.

[0006] To achieve the above-mentioned objectives, the present invention specifically adopts the following technical solution.

[0007] In a first aspect, the present invention discloses an IES situational awareness method based on graph-model-library integration and case-based reasoning, the method comprising the following steps:

[0008] S1. Collect data on various energy sources and loads in the integrated energy system, including sources, grids, loads, and storage, and construct the original dataset;

[0009] S2. Extract key features from the original dataset and construct a process network model of the integrated energy system; publish the process network model to the integrated management and control system, and store the operating status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections;

[0010] S3. Using the integrated management and control system as a carrier, identify the future state characteristics of the integrated energy system in the process network model, and perform physical constraint verification to obtain the target state characteristics; in the historical case cross-section set, search for historical cases with a similarity greater than a preset similarity threshold with the target state characteristics as reference cases; extract the evolution law template of the key features from the reference cases, and modify it based on the current system characteristics to obtain the modified key feature evolution law template; perform weighted fusion of the target state characteristics and the modified key feature evolution law template to obtain the final key feature prediction value, thereby realizing system situation prediction.

[0011] More preferably,

[0012] In S2, the extraction of key features from the original dataset specifically includes:

[0013] An initial candidate feature set is selected from the original dataset. The initial candidate feature set includes energy production fluctuation characteristics, load change trend characteristics, network transmission efficiency characteristics, and energy storage system status characteristics. Principal component analysis is used to reduce the dimensionality of the initial candidate feature set to obtain a core principal component subset. Based on the correlation analysis method, the correlation features in the core principal component subset are analyzed to extract the key features from the original dataset.

[0014] More preferably,

[0015] In S2, the process network model of the integrated energy system is constructed in the following ways:

[0016] The integrated energy system is abstracted as a directed graph containing several nodes and edges. The nodes include source nodes, load nodes, network nodes, and storage nodes, and the edges are represented by arrow lines indicating the direction of energy flow. Each node and edge is assigned corresponding attributes, and a complete process network diagram is drawn based on the actual physical connections in the integrated energy system to generate a graphical information file.

[0017] The database filling program is invoked to convert the graphic information file into a structured logical database model containing a node table, an edge table, and an association table, thereby realizing the mapping from graphics to data. The node table includes attribute data of all nodes, the edge table includes attribute data of all edges, and the association table includes the topological association relationship between nodes and edges.

[0018] More preferably,

[0019] In S2, storing the operational status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections specifically includes:

[0020] The operational status of the integrated energy system in the process network model is collected at fixed intervals to form case profiles. Each case profile includes a graph model instance, key features, and system indicators at the time of collection. Case profiles are stored based on different operating conditions to obtain independent sub-libraries under different operating condition categories. For each independent sub-library, a high-dimensional spatial index is constructed using a KD-tree based on the key features in the case profile, so that each node of the KD-tree corresponds to a key feature of a case profile. The space is divided by the dimension of the key features, so that case profiles with similar key features are clustered in the index. The case profiles are then updated according to a preset update rule.

[0021] More preferably,

[0022] The graph model instance includes the attributes of all nodes and edges in the directed graph; the system indicators include: total energy utilization efficiency, supply and demand balance deviation rate, network transmission reliability coefficient, energy storage synergy support coefficient, and comprehensive operating cost.

[0023] More preferably,

[0024] The preset update rule is as follows:

[0025] For newly added process network model running instances under special operating conditions, increase the collection frequency of their corresponding case segments; set the life cycle of case segments, and archive case segments whose storage time exceeds the life cycle to the cold backup library; for process network model running instances corresponding to operating conditions that fail due to equipment retirement, their associated case segments are marked as failed and no longer participate in similarity search.

[0026] More preferably,

[0027] In S3, the future state characteristics of the integrated energy system are specifically as follows:

[0028] The current state features are obtained by calculating the key features using a sliding window. F 当前 The time series extrapolation algorithm is used to analyze the... F 当前 Extrapolate to generate the future The state feature vector at hour t represents the future state feature of the integrated energy system.

[0029] More preferably,

[0030] In S3, the step of extracting the evolutionary pattern template of the key features from the reference case specifically involves:

[0031] The historical sequence of each key feature of the case section in the reference case is decomposed into trend term and period term using a time series decomposition model. After removing the residual term, the evolution law template of the key feature is formed.

[0032] More preferably,

[0033] The revised template for the evolution of key features is determined as follows:

[0034] The relative deviation between the core parameters of the current system nodes and the core parameters of the nodes in the original key feature evolution law template is calculated as the node attribute correction coefficient. The ratio of the edge transformation efficiency of the current system to the edge transformation efficiency of the original key feature evolution law template is calculated as the edge attribute correction coefficient. The overlap ratio of nodes and edges in the current system topology and the original key feature evolution law template is calculated as the topology structure correction coefficient. The node attribute correction coefficient, edge attribute correction coefficient, and topology structure correction coefficient are weighted and fused to obtain the correction coefficient of the key feature evolution law template. The product of the correction coefficient and the original key feature evolution law template is the corrected key feature evolution law template.

[0035] Secondly, the present invention discloses an IES situation awareness system based on the aforementioned method, including an original dataset construction module, a process network model construction module, and a system situation prediction module;

[0036] The raw dataset construction module collects various energy sources and load data from the integrated energy system, including sources, grids, loads, and storage, to construct the raw dataset.

[0037] The process network model construction module extracts key features from the original dataset and constructs a process network model of the integrated energy system; the process network model is published to the integrated management and control system, and the operating status of the integrated energy system in the process network model is stored in the integrated management and control system in the form of case cross-sections;

[0038] The system situation prediction module, using the integrated management and control system as a carrier, identifies the future state characteristics of the integrated energy system in the process network model and performs physical constraint verification to obtain target state characteristics. In the historical case cross-section set, it searches for historical cases with a similarity greater than a preset similarity threshold to the target state characteristics as reference cases. It extracts the evolutionary pattern template of the key features from the reference cases and corrects it based on the current system characteristics to obtain a corrected key feature evolutionary pattern template. Finally, it performs a weighted fusion of the target state characteristics and the corrected key feature evolutionary pattern template to obtain the final key feature prediction value, thus realizing system situation prediction.

[0039] The beneficial effects of this invention are compared with those of the prior art:

[0040] (1) In terms of modeling methods, existing technologies are difficult to characterize the collaborative characteristics of multiple energy sources across links, and there are also problems such as asynchronous time of multi-source data and low retrieval efficiency of large-scale case library. This invention constructs an integrated graph-model-library model, which abstracts IES into a directed graph containing four types of nodes: source, network, load, and storage. Combined with structured database mapping and real-time dynamic synchronization of the library, it can simultaneously characterize the conversion and flow relationships of multiple energy sources such as electricity, heat, and gas. At the same time, the case cross section adopts three-dimensional tensor storage and "working condition classification first-level index + KD tree high-dimensional second-level index", which solves the problems of asynchronous time of multi-source data and low retrieval efficiency of large-scale case library, further improving the sufficiency of data fusion and the dynamic adaptability of the model. Moreover, the model stability is improved in extreme weather scenarios.

[0041] (2) In terms of situation prediction, existing technologies rely on a single algorithm and lack a historical case reuse mechanism, resulting in large errors under sudden changes in operating conditions. This invention combines case reasoning and situation awareness. First, it uses a time series extrapolation algorithm to extrapolate and cover new operating conditions without similar cases. Then, it uses a process network model to verify and eliminate infeasible prediction values ​​through physical constraints, thus solving the problem that traditional single case reuse cannot cover new operating conditions and that prediction values ​​are prone to exceed physical limits. It uses a similarity measurement algorithm to automatically calculate feature weights, solving the problem of mismatch of similar cases caused by subjective weights, reducing prediction errors under sudden changes in operating conditions. Finally, by reusing historical similar sections, it is more effective than a single neural network prediction method in reducing the short-term situation prediction error of the system, thus gaining more response time for fault warning and scheduling decisions. Attached Figure Description

[0042] Figure 1This is the IES situational awareness structure model in this invention;

[0043] Figure 2 This is the integrated modeling process of the image and model library in this invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0045] The invention proposes an IES situational awareness method based on integrated graph model library and case reasoning, the method comprising the following steps:

[0046] S1. Collect data on various energy sources and loads in the integrated energy system, including sources, grids, loads, and storage, and construct the original dataset;

[0047] S2. Extract key features from the original dataset and construct a process network model of the integrated energy system; publish the process network model to the integrated management and control system, and store the operating status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections;

[0048] In S2, the extraction of key features from the original dataset specifically includes:

[0049] An initial candidate feature set is selected from the original dataset. The initial candidate feature set includes energy production fluctuation characteristics, load change trend characteristics, network transmission efficiency characteristics, and energy storage system status characteristics. Principal component analysis is used to reduce the dimensionality of the initial candidate feature set to obtain a core principal component subset. Based on correlation analysis methods, the correlation features in the core principal component subset are analyzed to extract the key features from the original dataset. The correlation analysis methods include the Pearson correlation coefficient method and the mutual information entropy method.

[0050] In S2, the process network model of the integrated energy system is constructed in the following ways:

[0051] The integrated energy system is abstracted as a directed graph containing several nodes and edges. The nodes include source nodes, load nodes, network nodes, and storage nodes, and the edges are represented by arrow lines indicating the direction of energy flow. Each node and edge is assigned corresponding attributes, and a complete process network diagram is drawn based on the actual physical connections in the integrated energy system to generate a graphical information file.

[0052] The database filling program is invoked to convert the graphic information file into a structured logical database model containing a node table, an edge table, and an association table, thereby realizing the mapping from graphics to data. The node table includes attribute data of all nodes, the edge table includes attribute data of all edges, and the association table includes the topological association relationship between nodes and edges.

[0053] In S2, storing the operational status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections specifically includes:

[0054] The operational status of the integrated energy system in the process network model is collected at fixed intervals to form case profiles. Each case profile includes a graph model instance, key features, and system indicators at the time of collection. Case profiles are stored based on different operating conditions to obtain independent sub-libraries under different operating condition categories. For each independent sub-library, a high-dimensional spatial index is constructed using a KD-tree based on the key features in the case profile, so that each node of the KD-tree corresponds to a key feature of a case profile. The space is divided by the dimension of the key features, so that case profiles with similar key features are clustered in the index. The case profiles are then updated according to a preset update rule.

[0055] The preset update rule is as follows:

[0056] For newly added process network model running instances under special operating conditions, increase the collection frequency of their corresponding case segments; set the life cycle of case segments, and archive case segments whose storage time exceeds the life cycle to the cold backup library; for process network model running instances corresponding to operating conditions that fail due to equipment retirement, their associated case segments are marked as failed and no longer participate in similarity search.

[0057] Specifically, the special operating conditions include a composite operating condition model of power grid failure and extreme cold wave;

[0058] S3. Using the integrated management and control system as a carrier, identify the future state characteristics of the integrated energy system in the process network model, and perform physical constraint verification to obtain the target state characteristics; in the historical case cross-section set, search for historical cases with a similarity greater than a preset similarity threshold with the target state characteristics as reference cases; extract the evolution law template of the key features from the reference cases, and modify it based on the current system characteristics to obtain the modified key feature evolution law template; perform weighted fusion of the target state characteristics and the modified key feature evolution law template to obtain the final key feature prediction value, thereby realizing system situation prediction.

[0059] In S3, the future state characteristics of the integrated energy system are specifically as follows:

[0060] The current state features are obtained by calculating the key features using a sliding window. F 当前 The time series extrapolation algorithm is used to analyze the... F 当前 Extrapolate to generate the future The state feature vector at hour t represents the future state feature of the integrated energy system.

[0061] The time series extrapolation algorithms include, but are not limited to, the ARIMA time series algorithm and the Long Short-Term Memory (LSTM) algorithm.

[0062] In S3, the step of extracting the evolutionary pattern template of the key features from the reference case specifically involves:

[0063] The historical sequence of each key feature of the case section in the reference case is decomposed into trend term and period term using a time series decomposition model. After removing the residual term, the evolution law template of the key feature is formed.

[0064] The time series decomposition model includes, but is not limited to, the STL model.

[0065] The revised template for the evolution of key features is determined as follows:

[0066] The relative deviation between the core parameters of the current system nodes and the core parameters of the nodes in the original key feature evolution law template is calculated as the node attribute correction coefficient. The ratio of the edge transformation efficiency of the current system to the edge transformation efficiency of the original key feature evolution law template is calculated as the edge attribute correction coefficient. The overlap ratio of nodes and edges in the current system topology and the original key feature evolution law template is calculated as the topology structure correction coefficient. The node attribute correction coefficient, edge attribute correction coefficient, and topology structure correction coefficient are weighted and fused to obtain the correction coefficient of the key feature evolution law template. The product of the correction coefficient and the original key feature evolution law template is the corrected key feature evolution law template.

[0067] This invention also claims protection for an IES situation awareness system based on the aforementioned method, including an original dataset construction module, a process network model construction module, and a system situation prediction module;

[0068] The raw dataset construction module collects various energy sources and load data from the integrated energy system, including sources, grids, loads, and storage, to construct the raw dataset.

[0069] The process network model construction module extracts key features from the original dataset and constructs a process network model of the integrated energy system; the process network model is published to the integrated management and control system, and the operating status of the integrated energy system in the process network model is stored in the integrated management and control system in the form of case cross-sections;

[0070] The system situation prediction module, using the integrated management and control system as a carrier, identifies the future state characteristics of the comprehensive energy system in the process network model and performs physical constraint verification to obtain target state characteristics. In the historical case cross-section set, it searches for historical cases with a similarity greater than a preset similarity threshold to the target state characteristics as reference cases. It extracts the evolutionary pattern template of the key features from the reference cases and corrects it based on the current system characteristics to obtain a corrected key feature evolutionary pattern template. Finally, it performs a weighted fusion of the target state characteristics and the corrected key feature evolutionary pattern template to obtain the final key feature prediction value, thus realizing system situation prediction. Example 1:

[0071] IES situational awareness structure model, such as Figure 1 As shown, based on the IES situational awareness structure model, this invention proposes an IES situational awareness method based on graph-model-library integration and case-based reasoning. The method includes the following steps:

[0072] S1. Collect data on various energy sources and loads from the integrated energy system, including sources, grids, loads, and storage, and construct the original dataset;

[0073] S1 includes the following steps:

[0074] S101: For each link of "source-grid-load-storage", determine the scope and type of data collection, including data of energy production links such as solar and wind power, data of energy network links such as power transmission network operation parameters, data of energy load links such as industrial and residential, and data of energy storage links such as the charging and discharging status of energy storage equipment.

[0075] (1) Data collection and definition for energy production links (sources), covering renewable energy and traditional energy production scenarios, specifically including:

[0076] Solar power generation system: Collects real-time data on photovoltaic array irradiance, module operating temperature, output voltage / current of individual photovoltaic panels and combiner boxes, inverter conversion efficiency, total power generation, and power fluctuation.

[0077] Wind power generation system: Collects data on wind turbine hub speed, wind speed / direction, blade angle, nacelle temperature, generator output power, converter operating status, and cumulative power generation;

[0078] Other energy production systems include fuel consumption and steam output of biomass boilers, geothermal fluid temperature / flow rate of geothermal units, and data such as unit output, fuel consumption, cooling water temperature, and unit operating status of traditional energy sources such as thermal power and hydropower.

[0079] (2) Data collection and definition of energy network links (networks), covering power transmission, distribution and microgrid interaction networks, specifically including:

[0080] Transmission network: Collect real-time voltage, current, active / reactive power, power factor, line transmission loss, conductor temperature, tower tilt angle, circuit breaker and disconnector status, as well as load rate, oil temperature and gas relay status data of substation main transformers at each voltage level.

[0081] Distribution network: Collect voltage deviation, three-phase imbalance, power flow distribution, feeder switch status, distribution transformer load rate, and grid connection interface power data of distributed power sources and distribution network.

[0082] Microgrid system: Collects the power of the tie line between the microgrid and the main grid, the coordinated operation data of each distributed power source within the microgrid, and the command response status data of the microgrid control system.

[0083] (3) Data collection and definition of energy load links (loads), covering industrial, residential, and commercial load types, specifically including:

[0084] Industrial load: Collect real-time power consumption, unit product energy consumption, start-up and shutdown status and energy consumption curves of major production equipment for each industrial user, as well as consumption and pressure / flow data of non-electric loads such as industrial steam and gas.

[0085] Residential load: Collect real-time electricity consumption, electricity consumption period distribution, operating power of air conditioning / heating equipment, and energy consumption data of community public facilities for residential users;

[0086] Commercial load: Collect real-time power of lighting, air conditioning, elevators and other equipment in shopping malls and office buildings, as well as load fluctuation curves and peak-valley load difference data during business hours;

[0087] Load characteristic data: Synchronously collect data such as the adjustability parameters of various loads and historical load prediction deviations.

[0088] (4) Data collection and definition for energy storage (storage) links, covering the operating status of various energy storage devices, specifically including:

[0089] Electrochemical energy storage system: Collects data on individual cell voltage / current, SOC (state of charge), SOH (state of health), charge / discharge power, charge / discharge cycle count, battery box internal temperature, fire protection system status, and fault alarm information of the battery pack;

[0090] Mechanical energy storage system: Collects data on flywheel energy storage speed, energy storage capacity, and output power;

[0091] Other energy storage systems: Collect data on compressed air energy storage, including storage pressure / volume and energy release power; and for thermal / cold storage systems, collect data on medium temperature, heat / cold storage capacity, and heat exchange efficiency.

[0092] S102: Perform time-granular alignment on the collected multi-source data to construct the original dataset;

[0093] Specifically, the following steps are included:

[0094] (1) Time granularity alignment

[0095] Data from different acquisition frequencies are standardized to a baseline time granularity (1 minute / time); for low-frequency data (e.g., 5-minute intervals), linear interpolation is used to supplement intermediate time values.

[0096] Suppose the original data is at time... t 0, t The value of 5 is x 0, x 5, then t k ( k The interpolation values ​​for (1, 2, 3, 4 minutes) are:

[0097] (1)

[0098] For high-frequency data (such as 30-second intervals), a sliding window mean reduction method is used to reduce the frequency of data per minute. for:

[0099] (2)

[0100] (2) Construct the original dataset

[0101] Using a three-dimensional tensor structure ∈ T×S×F Building the original dataset ,in:

[0102] T represents the time dimension;

[0103] S is the spatial dimension, containing J = J1 + J2 + J3 + J4 nodes, where J1 is the source node, J2 is the network node, J3 is the load node, and J4 is the storage node;

[0104] F represents the feature dimension.

[0105] Tensor elements ( t , j , f ) indicates the first t Time of the first j The node of the first f Class feature values.

[0106] Through the above three-dimensional tensor ∈ T×S×F After constructing the global raw dataset, to meet the subsequent feature engineering requirements for "independent analysis units," the three-dimensional tensor needs to be converted into a two-dimensional sample set: Define "single time step - single node" as the smallest sample unit, and traverse the tensor according to the rule of "time priority, node priority," reshaping it into an N×M two-dimensional sample matrix (N=T×S is the total number of samples, M=F is the feature dimension). Each sample xi corresponds to the "first"... t Time step - 1 j The F-class feature values ​​of the "node" are obtained, and a mapping table between samples and "timestamp-node ID" is stored simultaneously so that subsequent feature results can be back-linked to the global tensor; the specific steps to convert the three-dimensional tensor into a two-dimensional sample set are as follows:

[0107] Step 1: Determine the sample unit mapping rules;

[0108] Define "the first" t time step ( t =1,2,…,T)+then j Nodes (j =1,2,…,S)” is one independent sample, that is, each “time-node” combination corresponds to one sample, and the total number of samples N=T×S.

[0109] Step 2: Reconstruct the 3D tensor into a 2D matrix;

[0110] Traverse the 3D tensor in "time priority, node priority" order. (T×S×F), reshape it into an N×M two-dimensional matrix (M=F):

[0111] The first two-dimensional matrix i Row (corresponding to sample xi): If i =( t 1)×S+ j ( t For time step index, j If the node index is used, then the M elements of this row are [ ( t , j ,1), ( t , j ,2),…, ( t , j ,F)];

[0112] If T=2 (2 time steps), S=3 (3 nodes), and F=2 (2 types of features), the 3D tensor is: Then the two-dimensional sample matrix is There are 6 samples (N=2×3=6).

[0113] Step 3: Retain the association identifiers between the retained samples and tensors;

[0114] During the transformation process, the timestamp corresponding to each sample xi is recorded synchronously. t "and nodes" j (Form a mapping table) to ensure that after the subsequent feature selection is completed, the key feature results can be mapped back to the "time-space" position of the three-dimensional tensor, providing data association basis for S202 to build the process network model and S203 to generate case cross-sections.

[0115] S2. Extract key features from the original dataset and construct a process network model of the integrated energy system; publish the process network model to the integrated management and control system, and store the operating status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections;

[0116] S2 includes the following steps:

[0117] S201: From the original dataset using feature engineering methods An initial candidate feature set is selected, and key features are obtained by further screening of the initial candidate feature set; the initial candidate feature set includes energy output fluctuation characteristics, load change trend characteristics, network transmission efficiency characteristics, and energy storage system status characteristics; specifically:

[0118] 1) Energy output fluctuation characteristics, characterizing the stability and dynamic changes in energy production output at the "source" stage (photovoltaic, wind power, thermal power, etc.), reflecting the impact of renewable energy output fluctuations on the system, specifically including:

[0119] Photovoltaic: power fluctuation amplitude of photovoltaic array, irradiance change rate, and inverter conversion efficiency fluctuation value;

[0120] Wind power: wind turbine output fluctuation rate, wind speed change frequency, generator power fluctuation period;

[0121] Traditional energy categories: output adjustment rate of thermal power units, fluctuation coefficient of inflow water for hydropower, and fluctuation of steam output from biomass energy.

[0122] 2) Load variation trend characteristics, characterizing the time-series variation patterns and demand elasticity of energy consumption in the "load" segment (industrial, residential, and commercial), reflecting the impact of load fluctuations on the system's supply and demand balance, specifically including:

[0123] Time-series trend data: Hourly rate of change of industrial load, daily peak-valley difference of residential load, and time-period fluctuation coefficient of commercial load;

[0124] Elasticity characteristics include: load adjustability margin, response delay time, and historical load forecast deviation rate.

[0125] Mutation characteristics: magnitude of sudden increase / decrease in load, duration of mutation.

[0126] 3) Network transmission efficiency characteristics, characterizing the efficiency and topological stability of energy transmission and conversion in the "network" links (transmission, distribution, microgrids), reflecting the impact of network losses and power flow distribution on system economy, specifically including:

[0127] Transmission efficiency categories: power loss rate of transmission lines, voltage deviation rate of distribution networks, and power exchange efficiency of microgrid tie lines;

[0128] Topology and power flow: uniformity of power flow distribution in transmission lines, three-phase imbalance in distribution networks, and power fluctuation at the grid connection interface of distributed generation in microgrids;

[0129] Equipment status categories: substation main transformer load rate fluctuation, circuit breaker switching frequency, and tower tilt angle change.

[0130] 4) Energy storage system state characteristics, characterizing the energy storage capacity, health status, and charge / discharge dynamics of the "storage" stage (electrochemical, mechanical, thermal / cold storage), reflecting the energy storage's ability to mitigate system fluctuations, specifically including:

[0131] Energy status categories: electrochemical energy storage SOC (state of charge), thermal storage system heat storage capacity remaining rate, compressed air energy storage gas pressure;

[0132] Health and safety: Battery pack SOH (State of Health), internal temperature gradient of the battery box, and wear coefficient of the flywheel energy storage bearing;

[0133] Charge and discharge dynamics: rate of change of charge and discharge power of electrochemical energy storage, fluctuation of cold release efficiency of cold storage system, and stability of flywheel energy storage speed.

[0134] The above are the four initial candidate features in the original parameter set. They need to be filtered as follows:

[0135] (1) Principal Component Analysis (PCA) is used to reduce the dimensionality of highly correlated parameters of the four initial candidate features, retaining principal components with a cumulative contribution rate ≥95% and reducing redundant features. The original quantized parameter set of the four initial candidate features is input, and redundancy is removed from the highly correlated parameters within / between the four initial candidate features. Let the original dataset be... ={x1,x2,…,x N},in x i =[ x i1 , xi2 ,…, x iM Let be the M-dimensional feature vector of the i-th sample. First, standardize the features to eliminate the influence of dimensions:

[0136] (3)

[0137] in, m k For the first k The mean of the dimensional features, For the first k Standard deviation of dimensional features For the first i The first sample k The original values ​​of the dimensional features, These are the standardized feature values.

[0138] The standardized data matrix is Calculate the covariance matrix that reflects the correlation between features:

[0139] (4)

[0140] Where C is the covariance matrix, and C∈ M×M C jk For the first j peacekeeping k The covariance of the dimensional features; a larger value indicates a stronger correlation. N This represents the number of samples.

[0141] Perform eigenvalue decomposition on the covariance matrix C:

[0142] (5)

[0143] in, l p For the first p There are eigenvalues, and they satisfy the following conditions: l 1≥ l 2≥...≥ l M ≥0, u p This is the corresponding unit eigenvector (principal component loading).

[0144] No. p Contribution rate of each principal component for:

[0145] (6)

[0146] Cumulative contribution rate is the highest P The sum of the contribution rates of the principal components must satisfy the following condition: ≥95%, before extraction P The principal components are used as features after dimensionality reduction:

[0147] (7)

[0148] Among them, z i For the first i one sample P Principal component eigenvectors ( P ≤ M ).

[0149] After the above processing, the core principal component subsets of the four initial candidate features are obtained. Each principal component corresponds to a linear combination of one or more original parameters in the four features (such as "load trend principal component = 0.6 × industrial load change rate + 0.4 × civilian load peak-valley difference"), thus realizing the information condensation of the four features.

[0150] (2) Based on correlation analysis, the association features in the core principal component subset are analyzed to extract the key features in the original dataset; specifically including:

[0151] For the linear correlation features in the core principal component subset, calculate their linear correlation with the core system status indicators, and select features with a correlation coefficient ≥ 0.6 as key features; the core system status indicators are key evaluation parameters for the global operation of IES, including: overall system efficiency (the ratio of total output energy to total consumption energy), fault early warning accuracy (the matching ratio between actual faults and early warning faults), and supply-demand balance deviation rate (the ratio of the difference between total power generation and total load to total load).

[0152] Let the linear correlation characteristic variables in the core principal component subset be... F The core status indicators of the system are I Then the Pearson correlation coefficient is:

[0153] (8)

[0154] in, , These represent the specific values ​​of the feature variables and the core state indicators of the system in the i-th sample, respectively. , They are respectively F and I The mean, N The total number of samples involved in the calculation; when At that time, retain features F。

[0155] For the nonlinear correlation features (such as load abrupt changes and weather factors) in the core principal component subset, the dependency is evaluated using mutual information entropy. Features with mutual information entropy ≥ 0.5 are retained as key features. The specific calculation method is as follows:

[0156] Let characteristic variables be defined. F and system core indicators I The joint probability distribution is p ( F , I The marginal probability distribution is p ( F ), p ( I If the mutual information entropy is:

[0157] (9)

[0158] To facilitate comparison with linear indices, mutual information can be normalized (value range [0,1]):

[0159] (10)

[0160] in, for F Information entropy H ( I )for I Information entropy; when At that time, retain features F .

[0161] The final output yields the key features of the four initial candidate features.

[0162] S202: Based on the key features output by S201, and using the integrated modeling technology of graph model library, the energy production, transmission, consumption and storage links are abstracted into process nodes and connection relationships with clear attributes to construct a process network model of the integrated energy system.

[0163] Specifically, the following steps are included:

[0164] (1) Draw the IES process network diagram

[0165] Based on the "source-network-load-storage" node, the primitives and connections of the process network are generated, and each node and edge is assigned corresponding attributes. These attributes include, but are not limited to, the key feature variables extracted in S201, the type of each node, and the direction of each edge. The IES process network graph drawing process specifically includes:

[0166] The integrated energy system is abstracted as a directed graph G=(J,E), where node J includes:

[0167] Source node J1 (abstracted as a single-endpoint primitive): Attributes include energy type (photovoltaic / wind power / thermal power, etc.), rated output, real-time efficiency, and fault probability;

[0168] Network node J2 (abstracted as two-endpoint primitives): Attributes include network type (transmission / distribution / microgrid), transmission capacity, loss coefficient, and topology connection status;

[0169] Load node J3 (abstracted as a single-endpoint primitive): Attributes include load type (industrial / residential / commercial), real-time demand, adjustment elasticity coefficient, and response delay;

[0170] Storage node J4 (abstracted as two endpoint primitives): Attributes include energy storage type (electrochemical / mechanical, etc.), current SOC, maximum charge / discharge power, and cycle life.

[0171] Directed edges E are represented by arrows, indicating the direction of energy flow and transformation relationships, such as transmission lines from power plants to load centers, and heating pipelines from combined heat and power plants to heat users. Attributes include:

[0172] Flow direction (e.g., source → grid, grid → load);

[0173] Conversion efficiency (such as transmission line efficiency, thermoelectric conversion efficiency);

[0174] Constraints (such as maximum transmission power, minimum pressure / voltage threshold).

[0175] Based on the actual physical connections of the IES (such as photovoltaic power station → substation → transmission line → industrial load, thermal power plant → heating pipeline → civil building), draw a complete flow network diagram and generate a graphic information file (including element coordinates, attribute labels, connection relationships, etc.).

[0176] (2) Constructing a logical database model

[0177] The database filling program is invoked to convert the graphic information file into a structured model in the logical database, realizing the "graphic-data" mapping, specifically including:

[0178] Node table: Stores attribute data for all nodes. Fields include: Node ID, Node type (Source / Network / Load / Storage), Physical location, Key characteristic parameters, and Status identifier (Normal / Fault / Maintenance).

[0179] Edge table: Stores attribute data for directed edges. Fields include: edge ID, starting node ID, ending node ID, energy type, flow direction, conversion efficiency, and constraints;

[0180] Association table: Records the topological relationships between nodes and edges, ensuring that the "node-edge" connections in the graph are consistent with the logical relationships in the database.

[0181] (3) Model validation

[0182] Verify the completeness, consistency, and parameter rationality of the process network model to avoid "empty models" (graphs exist but there is no corresponding data in the database) and ensure that the graph model is consistent with the actual system.

[0183] (4) The model is published to the real-time database.

[0184] The validated logical database model is published to the real-time database of the integrated management and control system, enabling integration with real-time data. When the IES topology changes or parameters are adjusted, it is automatically synchronized to the real-time database.

[0185] The integrated modeling process of IES's graph library is as follows: Figure 2 As shown.

[0186] S203: Based on the process network model, the case section management module stores the system operation status of each link of the integrated energy system in the process network model at different times in the form of case sections. The case sections include the graph model instance, key features and system indicators at the time of acquisition.

[0187] (1) Cross-section data collection: Record the system operation status every 5 minutes to form a case cross-section, which includes the graphical model instance, key feature variable values ​​and system indicators (such as total efficiency and failure rate) at that moment.

[0188] (11)

[0189] G(t) is a graph model instance, including the real-time attributes of all nodes and edges in the directed graph, which is called from the model library; F(t) is a key feature vector, which is the feature value extracted at this moment from the key features selected by S201; I(t) is a system index, which is calculated and generated based on the node-edge collaborative relationship of the model.

[0190] All system metrics are calculated based on the node attributes (such as source node output, load node demand, and storage node SOC) and edge attributes (such as transmission efficiency and maximum transmission power) of the S202 process network model. The calculation is achieved through the logic of "node parameters → edge transmission conversion → multi-node collaborative aggregation." These system metrics include total energy utilization efficiency, supply-demand balance deviation rate, network transmission reliability coefficient, energy storage collaborative support coefficient, and comprehensive operating cost, specifically calculated as follows:

[0191] 1) Total energy efficiency

[0192] The energy utilization level of the entire IES from "energy output of source nodes" to "effective consumption of load nodes" is quantified. The transmission loss between network nodes and edges and the charging and discharging loss of energy storage nodes are deducted to reflect the collaborative efficiency of "source-network-load-storage".

[0193] First, calculate the total energy output of the source node. Summarize the real-time energy output of all source nodes J1, taking into account the efficiency of the source nodes themselves:

[0194] (12)

[0195] Where J1 is the set of all source nodes in S202; For the first Each source node in t Real-time output power at any given moment; Δt is the calculation time interval (consistent with the acquisition cycle of section S203, taken as 5 minutes = 300 seconds);

[0196] For the first Each source node in t The efficiency of its own conversion at any given moment.

[0197] Then calculate the total energy loss across the entire link. E loss ( t This includes two types of losses, both of which are related to the transmission characteristics of the edges and the collaborative operation of the nodes.

[0198] Transmission loss at the edge E loss,edge ( t ): Summarize all edges E Transmission loss, calculated based on edge transmission efficiency:

[0199] (13)

[0200] in, P e,trans ( t ) is the first e The edge is t Transmission power at any given moment; or e ( t ) is the first e The edge is t Transmission efficiency at any given moment.

[0201] Energy storage node charging and discharging losses E loss,storage ( t ): Summarize the charging and discharging losses of all storage nodes J4, reflecting the coordinated losses between energy storage and source / load:

[0202] (14)

[0203] in, For the first Each storage node t The charging and discharging power at any given moment (positive for charging, negative for discharging). For the first Each storage node t The charging and discharging efficiency at any given time.

[0204] Total loss for:

[0205] (15)

[0206] The effective energy consumption of the load node = total output of the source node - total loss; therefore, the efficiency formula is:

[0207] (16)

[0208] 2) Supply and demand balance deviation rate

[0209] Quantifying IES t The discrepancy between the coordinated matching of "total output of source nodes + regulation capacity of storage nodes" and "total demand of load nodes + transmission loss of network nodes" reflects the risk of supply and demand imbalance in the system.

[0210] Calculate total supply capacity This includes real-time power output from the source node and adjustable power from the storage node.

[0211] (17)

[0212] in, For the first Each storage node t The adjustable power at any given time is a storage node attribute, determined by the current SOC and the maximum charge / discharge power constraints, specifically:

[0213] (18)

[0214] in, For the first The node capacity of each storage node, For the first The current state of each storage node t SOC at any moment For the first The maximum charge and discharge power of each storage node.

[0215] Calculate total demand and losses This includes the total load node requirements and the real-time transmission loss of the edges:

[0216] (19)

[0217] in, For the first Each load node is at t Real-time power demand at any given moment; For the first e The edge is t Real-time transmission power loss at any given moment:

[0218] Calculate the supply-demand imbalance rate:

[0219] (20)

[0220] 3) Network transmission reliability coefficient

[0221] The reliability of the coordinated operation of network nodes (transmission / distribution nodes) and edges is quantified, reflecting the degree to which the "load capacity of network nodes + transmission constraints of edges" are met, and reflecting the system's ability to resist overload and faults.

[0222] Computing network node load compliance rate Summarize whether the load rate of all network nodes J2 is within the safe threshold:

[0223] (twenty one)

[0224] Computing edge transmission compliance rate Summarize whether the transmission power of all edges E is within the maximum constraint:

[0225] (twenty two)

[0226] Calculate the network transmission reliability coefficient and the compliance rate of weighted fused network nodes and edges:

[0227] (twenty three)

[0228] in, , These are the load compliance rate weighting coefficient and the side transmission compliance rate weighting coefficient, respectively.

[0229] 4) Energy storage synergy support coefficient

[0230] The ability of quantified energy storage nodes to coordinate and mitigate fluctuations in output from source nodes and sudden changes in demand from load nodes demonstrates the "buffering and regulation" role of energy storage in the entire system.

[0231] Calculate source node fluctuation power Summarize the output fluctuations of all source nodes and calculate based on a sliding window:

[0232] (twenty four)

[0233] in, For the first The average output of each source node at time t-T0 reflects short-term fluctuations.

[0234] Calculate the power of load node mutation This summarizes all demand mutations for all load nodes, also based on a sliding window:

[0235] (25)

[0236] in, For the first The average power of each load node at time t-T0.

[0237] Calculate the power of energy storage coordinated regulation Summarize the actual power involved in the smoothing process across all storage nodes:

[0238] (26)

[0239] in, For the first The transmission efficiency of the connection edge of each energy storage node reflects that the energy storage power can only be effective if it is transmitted through the edge.

[0240] Calculate the energy storage synergy support coefficient:

[0241] (27)

[0242] 5) Overall operating costs

[0243] Quantify the unit-time economic cost of the coordinated operation of the entire IES, including the costs of energy production, transmission, energy storage and regulation, to reflect the economic efficiency of system operation.

[0244] Calculate the total production cost of the source node Summarize the power generation costs of all source nodes:

[0245] (28)

[0246] in, For the first The unit energy production cost of each source node.

[0247] Calculate network transmission loss cost Based on edge-based transmission loss and energy baseline cost:

[0248] (29)

[0249] in, This represents the average cost per unit of energy at the source node, reflecting the economic cost of transmission losses.

[0250] Calculate energy storage regulation costs Summarize the operation and maintenance costs of charging and discharging storage nodes:

[0251] (30)

[0252] in, For the first The unit adjustment cost of each storage node.

[0253] Calculate the overall operating cost:

[0254] (31)

[0255] (2) Cross-section index

[0256] Primary Index: Cross-sections are categorized and stored according to the operating condition type (normal / fault / maintenance) of the S202 model library, with an independent sub-library established under each operating condition category. The normal sub-library only stores cross-sections marked as "normal" operating condition, ensuring that cases under the same operating condition are matched first during retrieval.

[0257] Secondary Index: For each working condition sub-database, a high-dimensional spatial index is constructed using a KD-tree based on the key feature vector F(t) of the cross section. Each node of the KD-tree corresponds to the F(t) of a cross section. By dividing the space through feature dimensions, cross sections with similar features are clustered in the index, significantly improving the efficiency of subsequent similar case searches.

[0258] (3) Cross-section update

[0259] For newly added special operating condition models (such as the new "grid fault + extreme cold wave" composite operating condition model), the frequency of cross-section data collection for the corresponding scenario will be increased simultaneously to strengthen the accumulation of cases for that scenario. At the same time, a lifecycle for the cross-sections will be set, and historical cross-sections exceeding their lifecycle will be automatically archived to a cold backup library. For models that fail due to equipment retirement, their associated cross-sections will be synchronously marked as "failed" and will no longer participate in similarity searches.

[0260] S3. Using the integrated management and control system as a carrier, identify the future state characteristics of the integrated energy system in the process network model, and perform physical constraint verification to obtain the target state characteristics; in the historical case cross-section set, search for historical cases with a similarity greater than a preset similarity threshold with the target state characteristics as reference cases; extract the evolution law template of the key features from the reference cases, and modify it based on the current system characteristics to obtain the modified key feature evolution law template; perform weighted fusion of the target state characteristics and the modified key feature evolution law template to obtain the final key feature prediction value, thereby realizing system situation prediction;

[0261] S3 includes the following steps:

[0262] S301: Relying on the integrated management and control system, the system collects the operational data of the networked IES in real time, extracts key features from the S203 case profile, and predicts future state features to provide target state features for case retrieval.

[0263] (1) Sliding window calculation of current state features F 当前

[0264] Let the sampling interval of the real-time data be Δ. t 采样 =1 minute (consistent with the time granularity of S102), sliding window length is L =5 minutes (synchronized with the S203 section acquisition cycle), then the first t Current state characteristics at time step F 当前 ( t ) Calculate by aggregating data within the window:

[0265] (32)

[0266] in, K For feature dimensions (compared to the S203 case cross-section) F ( t (The dimensions are completely consistent). For the first k The aggregated values ​​of class features within the window are as follows:

[0267] (33)

[0268] In the formula, For the first k Class features in t i The real-time sampled values ​​at each minute are aggregated using a method that ensures consistency with those in S203. F ( t The feature calculation logic is consistent.

[0269] (2) ARIMA model extrapolates to generate future state features F 未来

[0270] Based on the current state characteristics, a time series extrapolation algorithm is used to... F 当前 Extrapolate to generate the future The state feature vector F at hour t 未来 (t+ In this embodiment, the preferred algorithm is the ARIMA time series algorithm, whose core formula is:

[0271] (34)

[0272] in: For the first k Class features in The predicted value at any given time; m For constant terms, For the first k Time series of class features conduct d The characteristic sequence after order difference (eliminating non-stationarity); i Autoregressive coefficients ( p (level) i j The moving average coefficient ( q (rank); p Let the order be the autoregressive order. q The moving average order is... This is the white noise error term.

[0273] The extrapolated future feature vector is:

[0274] (35)

[0275] (3) Perform physical constraint verification of the process network model to obtain the target state characteristics.

[0276] Based on the process network model (including node / edge attribute constraints) constructed using S202, the future feature vectors are analyzed. Perform physical constraint verification and eliminate physically infeasible components. Take transmission power constraints as an example:

[0277] Let the first e The maximum transmission power of the line is Pmax e If the predicted transmission power characteristics satisfy:

[0278] (36)

[0279] Then the feature is modified to retain the physically feasible value. Specifically:

[0280] (37)

[0281] The final verified target feature vector is:

[0282] (38)

[0283] in, The verified number K One characteristic.

[0284] S302: Based on target state characteristics For the query vector, a similarity metric is set, and the operating scenarios that are highly similar to the current state are searched in the case library to provide reference cases for situation prediction.

[0285] First, based on The implicit current working condition characteristics are used to locate the corresponding sub-library in the S203 case section library, avoiding invalid cross-scenario searches and reducing computational load.

[0286] Then, with For query vectors, the KD-tree index built by S203 is called, and a similarity metric is set, using weighted Euclidean distance quantization. Historical cross-sectional features The similarity is calculated using the following formula:

[0287] (39)

[0288] in, , The first k The target feature and the first k One historical characteristic; F For feature dimension, For the first k The weights of each feature are determined by their similarity; the higher the similarity (the larger the Sim value), the better the case matching.

[0289] Calculate using the entropy weight method The information entropy of a feature reflects its influence on the system's state; the smaller the entropy value, the greater the weight.

[0290] (40)

[0291] In the formula, M represents the total number of key features; , for the first k The difference coefficients of each feature, where... For the firstk Information entropy of each feature.

[0292] Finally, historical cases with similarity greater than a preset similarity threshold are retained as a reference case set C={C1,C2,…,C…} n Each case contains feature vectors from historical moments. F 历史 The graph model parameters G(t) (node / edge attributes) and situation development data provide direct reference cases for subsequent situation prediction.

[0293] Preferably, in this embodiment, the preset similarity threshold is 80%.

[0294] S303: Analyze the searched reference cases, extract the evolution law template of the key features from the reference cases, and modify it based on the current system features to obtain the modified key feature evolution law template; perform weighted fusion of the future state features and the modified key feature evolution law template to obtain the final key feature prediction value, thereby realizing system situation prediction (extracting historical patterns → adapting to the current system → generating accurate predictions).

[0295] Specifically, the following steps are included:

[0296] (1) Based on reference cases, the evolution pattern template of the core indicators of “source-grid-load-storage” is extracted to provide a basis for subsequent pattern adaptation.

[0297] Prioritize selecting the first m reference cases from the similar case set C, and use the key feature vector F of the S203 case section as an example. 历史 Based on this, a time series decomposition model is used to analyze its time series from historical time t to t+. In this embodiment, the preferred model for understanding the evolution of the situation of t is the STL model. The STL model decomposes the historical change sequence of each key feature into a "trend term + period term." After removing the residual term, a template for the evolution of key features containing only the "key feature dimension" is formed. .

[0298] Meanwhile, the system index I(t) is used as a constraint for pattern extraction: when generating the template for the evolution of key features, it is necessary to ensure that the "system index I(t)" of the historical cases corresponding to the template meets the normal threshold range, so as to avoid including the patterns of abnormal historical cases in the system. This ensures the effectiveness of the pattern template.

[0299] (2) Calculate the correction coefficient to adapt to historical patterns and the current system (based on the template in step (1) to eliminate scenario differences).

[0300] Based on the evolutionary pattern template generated in step (1), compare it with the real-time graph model instance G of the current system.当前 The correction coefficient is calculated using G(t) from the reference case. The core objective is to eliminate the impact of the difference between the historical case scenario and the current system scenario on the reuse of patterns, and to ensure that the corrected historical patterns can match the current system state.

[0301] By comparing the core differences between the current system and historical cases from three dimensions—node attributes, edge attributes, and topology—three types of correction coefficients are calculated:

[0302] Node attribute correction coefficient Template based on the evolution of core parameters of current system nodes and key features before correction. The formula for calculating the relative deviation of the node core parameters is as follows:

[0303] (41)

[0304] in, k =1,2,3,4: These correspond to four types of nodes: source, network, load, and storage, respectively. For the first k Functional weights of class nodes; , The first k The core parameters of the current node and historical nodes.

[0305] The core parameters are specifically: the rated output of the source node, the transmission capacity of the network node, the real-time demand of the load node, and the current SOC of the storage node.

[0306] Edge attribute correction coefficient Based on the current system's edge conversion efficiency and The ratio of historical edge transformation efficiency in the equation is calculated using the following formula:

[0307] (42)

[0308] in, , These are the final edge transformation efficiency obtained by weighted fusion of the transformation efficiencies of all different types of edges in the current system, and the final edge transformation efficiency obtained by weighted fusion of the transformation efficiencies of all different types of edges in the past.

[0309] Topology correction factor : Calculated based on the overlap between nodes / edges in the current and historical topologies, using the following formula:

[0310] (43)

[0311] in, This represents the number of overlapping nodes between the current system topology and the reference case topology. This represents the number of overlapping edges between the current topology and the reference case topology. This represents the total number of nodes in the current system topology. This represents the total number of edges in the current system topology.

[0312] Calculate the overall correction factor Specifically:

[0313]

[0314] in, , , These are the weights corresponding to the node attribute correction coefficient, edge attribute correction coefficient, and topology correction coefficient, respectively, and the weights are set based on the importance of each dimension to the situation.

[0315] (3) Integrate the correction rules with real-time data to generate prediction results.

[0316] Using the key feature evolution law template corrected in step (2) and the target state features obtained in S301 as inputs, the final prediction result is generated through "weighted integration - accuracy verification - feedback adjustment". The core objective is to combine "adapted historical patterns" with "current real-time data" to ensure prediction accuracy and form a closed-loop optimization mechanism. The specific operation is as follows:

[0317] Integrate and correct the key feature evolution law template with the target state feature F of S301 目标 Predicting the future of IES Changes in key IES metrics within t hours.

[0318] Press F 目标 The final predicted value is calculated by weighting the proportion of (real-time data) β1 and the proportion of the corrected key feature evolution law template (empirical data) β2. The formula is as follows:

[0319] (44)

[0320] in, This is the final predicted value. Predicted values ​​from historical patterns obtained based on the modified key feature evolution pattern template This is a template for the evolution of key features based on the correction coefficient α in step (2).

[0321] Waiting for the future After time period t ends, the mean absolute percentage error (MAPE) between the predicted value and the actual value is calculated using the following formula:

[0322] (45)

[0323] in, The actual value is n, and the number of data points used to calculate the error in the future time interval Δt is n.

[0324] If MAPE > (Based on the valid threshold of the S302 reference case), the feedback mechanism is triggered: return to S302 to reduce the preset similarity threshold (from 80% to 75%), and re-execute steps (1)-(3) until MAPE ≤ This ensures that the prediction accuracy continuously adapts to the system's operational needs.

[0325] This invention is based on an integrated energy system network and achieves iterative interactive optimization of the perception layer, understanding layer, and prediction layer. At the awareness layer, various energy data from the IES (Generator-Grid-Load-Storage) are collected, covering energy production data such as solar and wind power, network data such as transmission network operation parameters, industrial and residential load data, and energy storage data such as the charging and discharging status of energy storage devices. This provides a comprehensive data source for subsequent modeling and prediction. At the understanding layer, data fusion is used to establish a mapping relationship between the model and the data, forming a process network model of the IES. Specifically, an integrated graph-model-library modeling technology is adopted to abstract the IES into a directed graph G=(J,E) containing four types of nodes: source, grid, load, and storage. Each type of node is assigned unique attributes, and directed edges E are labeled with energy flow direction, conversion efficiency, and physical constraints. At the same time, a structured logical database is constructed through node tables, edge tables, and association tables to achieve a one-to-one mapping between graph and data to avoid "empty models." The verified model is automatically published to the real-time database of the integrated management and control system to ensure that the model and real-time data are updated synchronously when the IES topology changes or parameters are adjusted. On this basis, the equipment, parameters, and data of each link in the "Generator-Grid-Load-Storage" system are stored in the form of case sections using a three-dimensional tensor structure. ∈ The T×S×F framework constructs the original dataset. First, it aligns the multi-source data at the time granularity, then establishes a two-level indexing mechanism to ensure the completeness of case profiles and retrieval efficiency. At the prediction layer, an integrated management and control system is used to automatically identify the future state characteristics of networked IES: first, the current state characteristic Fcurrent is calculated using a sliding window based on real-time running data; then, the ARIMA time series algorithm is used to extrapolate Fcurrent to generate the future state. The feature vector Ffuture is calculated at time t, and the physical constraints of the process network model are used to verify Ffuture, resulting in the target feature vector Ftarget. Subsequently, a case-based reasoning method is used to search for highly similar cases that have occurred in the historical cross-section set. Using Ftarget as the query vector, the corresponding working condition sub-database is first locked, and then the weights of each feature are calculated using the entropy weight method. The similarity between Ftarget and the historical cross-section features is quantified using the weighted Euclidean distance. A set of highly similar cases with a similarity of ≥80% is selected. Finally, through situational awareness technology, combined with the evolutionary patterns of similar cases and the current system state, system situation prediction is achieved.

[0326] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0327] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0328] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0329] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0330] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. An IES situational awareness method based on integrated graph model library and case reasoning, characterized in that, The method includes the following steps: S1. Collect data on various energy sources and loads in the integrated energy system, including sources, grids, loads, and storage, and construct the original dataset; S2. Extract key features from the original dataset and construct a process network model of the integrated energy system; publish the process network model to the integrated management and control system, and store the operating status of the integrated energy system in the process network model in the integrated management and control system in the form of case cross-sections; wherein, the integrated energy system is abstracted as a directed graph containing several nodes and edges, the nodes include source nodes, load nodes, network nodes and storage nodes, and the edges are represented by arrow lines to indicate the direction of energy flow; S3. Using the integrated management and control system as a carrier, identify the future state characteristics of the integrated energy system in the process network model, and perform physical constraint verification to obtain the target state characteristics; in the historical case cross-section set, search for historical cases with a similarity greater than a preset similarity threshold with the target state characteristics as reference cases; extract the evolution law template of the key features from the reference cases, and modify it based on the current system characteristics to obtain the modified key feature evolution law template; perform weighted fusion of the target state characteristics and the modified key feature evolution law template to obtain the final key feature prediction value, thereby realizing system situation prediction; wherein, the current system node core is calculated. The relative deviation between the core parameters and the core parameters of the key feature evolution law template nodes before correction is used as the node attribute correction coefficient. The ratio of the edge transformation efficiency of the current system to the edge transformation efficiency of the key feature evolution law template before correction is used as the edge attribute correction coefficient. The overlap ratio of nodes and edges in the current system topology and the key feature evolution law template before correction is used as the topology structure correction coefficient. The node attribute correction coefficient, edge attribute correction coefficient, and topology structure correction coefficient are weighted and fused to obtain the correction coefficient of the key feature evolution law template. The product of the correction coefficient and the original key feature evolution law template is the corrected key feature evolution law template.

2. The IES situational awareness method based on graph model library integration and case reasoning according to claim 1, characterized in that, In S2, the extraction of key features from the original dataset specifically includes: An initial candidate feature set is selected from the original dataset. The initial candidate feature set includes energy production fluctuation characteristics, load change trend characteristics, network transmission efficiency characteristics, and energy storage system status characteristics. Principal component analysis is used to reduce the dimensionality of the initial candidate feature set to obtain a core principal component subset. Based on the correlation analysis method, the correlation features in the core principal component subset are analyzed to extract the key features from the original dataset.

3. The IES situational awareness method based on graph model library integration and case reasoning according to claim 2, characterized in that, In S2, the process network model of the integrated energy system is constructed in the following ways: Assign corresponding attributes to each node and edge in the directed graph of the integrated energy system, and draw a complete flow network diagram based on the actual physical connections in the integrated energy system to generate a graphical information file; The database filling program is invoked to convert the graphic information file into a structured logical database model containing a node table, an edge table, and an association table, thereby realizing the mapping from graphics to data. The node table includes attribute data of all nodes, the edge table includes attribute data of all edges, and the association table includes the topological association relationship between nodes and edges.

4. The IES situational awareness method based on graph model library integration and case reasoning according to claim 3, characterized in that, In S2, storing the operational status of the integrated energy system in the process network model to the integrated management and control system in the form of case cross-sections specifically includes: The operational status of the integrated energy system in the process network model is collected at fixed intervals to form case profiles. Each case profile includes a graph model instance, key features, and system indicators at the time of collection. Case profiles are stored based on different operating conditions to obtain independent sub-libraries under different operating condition categories. For each independent sub-library, a high-dimensional spatial index is constructed using a KD-tree based on the key features in the case profile, so that each node of the KD-tree corresponds to a key feature of a case profile. The space is divided by the dimension of the key features, so that case profiles with similar key features are clustered in the index. The case profiles are then updated according to a preset update rule.

5. The IES situational awareness method based on graph model library integration and case reasoning according to claim 4, characterized in that, The graph model instance includes the attributes of all nodes and edges in the directed graph; the system indicators include: total energy utilization efficiency, supply and demand balance deviation rate, network transmission reliability coefficient, energy storage synergy support coefficient, and comprehensive operating cost.

6. The IES situational awareness method based on graph model library integration and case reasoning according to claim 4, characterized in that, The preset update rule is as follows: For newly added process network model running instances under special operating conditions, increase the collection frequency of their corresponding case segments; set the life cycle of case segments, and archive case segments whose storage time exceeds the life cycle to the cold backup library; for process network model running instances corresponding to operating conditions that fail due to equipment retirement, their associated case segments are marked as failed and no longer participate in similarity search.

7. The IES situational awareness method based on graph model library integration and case reasoning according to claim 4, characterized in that, In S3, the future state characteristics of the integrated energy system are specifically as follows: The current state feature Fcurrent is calculated using a sliding window method based on the key features. A time series extrapolation algorithm is then used to extrapolate Fcurrent to generate the future state feature. The state feature vector at hour t represents the future state feature of the integrated energy system.

8. The IES situational awareness method based on graph model library integration and case reasoning according to claim 7, characterized in that, In S3, the step of extracting the evolutionary pattern template of the key features from the reference case specifically involves: The historical sequence of each key feature of the case section in the reference case is decomposed into trend term and period term using a time series decomposition model. After removing the residual term, the evolution law template of the key feature is formed.

9. An IES situational awareness system based on the method of any one of claims 1-8, integrating graph model library and case reasoning, comprising a raw dataset construction module, a process network model construction module, and a system situational prediction module, characterized in that, The raw dataset construction module collects various energy sources and load data from the integrated energy system, including sources, grids, loads, and storage, to construct the raw dataset. The process network model construction module extracts key features from the original dataset and constructs a process network model of the integrated energy system. The process network model is then published to the integrated management and control system, and the operational status of the integrated energy system within the process network model is stored in the integrated management and control system in the form of case cross-sections. The integrated energy system is abstracted as a directed graph containing several nodes and edges. The nodes include source nodes, load nodes, network nodes, and storage nodes, and the edges are represented by arrow lines indicating the direction of energy flow. The system situation prediction module, using the integrated management and control system as a carrier, identifies the future state characteristics of the integrated energy system in the process network model and performs physical constraint verification to obtain target state characteristics. In the historical case cross-section set, it searches for historical cases with a similarity greater than a preset similarity threshold to the target state characteristics as reference cases. It extracts the evolutionary pattern template of the key features from the reference cases and corrects it based on the current system characteristics to obtain a corrected key feature evolutionary pattern template. It then performs a weighted fusion of the target state characteristics and the corrected key feature evolutionary pattern template to obtain the final key feature prediction value, thus realizing system situation prediction. Specifically, it calculates the current system... The relative deviation between the node core parameters and the core parameters of the key feature evolution law template before correction is used as the node attribute correction coefficient. The ratio of the edge transformation efficiency of the current system to the edge transformation efficiency of the key feature evolution law template before correction is used as the edge attribute correction coefficient. The overlap ratio of nodes and edges in the current system topology and the key feature evolution law template before correction is used as the topology structure correction coefficient. The node attribute correction coefficient, edge attribute correction coefficient, and topology structure correction coefficient are weighted and fused to obtain the correction coefficient of the key feature evolution law template. The product of the correction coefficient and the original key feature evolution law template is the corrected key feature evolution law template.

Images