Power system non-intrusive load identification and simulation modeling analysis method and system

Abstract

The invention discloses a non-intrusive load identification and simulation modeling analysis method and system for a power system, and belongs to the technical field of power system analysis. The method comprises the following steps: acquiring key node data of a power distribution network in a power system through a non-intrusive measurement device, preprocessing the key node data, and extracting and classifying load characteristics for the preprocessed key node data to obtain response characteristics; modeling of static loads and dynamic loads and identification of the dynamic loads are carried out based on the response characteristics, and a load component identification matrix is established according to modeling and identification results so as to distinguish different types of loads; for different types of loads, after a multi-scale space-time diagram attention network MSTG-Attention is adopted to carry out load intelligent classification, a simulation model comprising a static load and a dynamic load is established, and simulation verification is carried out through the simulation model. According to the method, the problems of insufficient load identification precision and limited dynamic modeling capability in the prior art are solved.

Description

Technical Field

[0001] This invention relates to the field of power system analysis technology, and more specifically, to a non-intrusive load identification and simulation modeling analysis method and system for power systems. Background Technology

[0002] With the widespread integration of new power electronic devices into power systems, traditional load modeling methods face significant challenges. Existing technologies suffer from the following problems:

[0003] Limitations of traditional load models: Traditional ZIP static load models cannot accurately describe the dynamic characteristics of power electronic loads, especially their response behavior under grid frequency and voltage disturbances.

[0004] Power electronic loads have complex characteristics: new types of loads such as frequency converters, charging piles, and distributed power sources have characteristics such as fast response and nonlinearity, and the mechanism of their impact on power grid stability has not yet been fully understood.

[0005] Insufficient accuracy in load identification: Existing load identification methods are mainly based on traditional measurement methods and lack the ability to accurately extract and classify dynamic parameters of power electronic loads.

[0006] Modeling accuracy needs improvement: Existing modeling methods have difficulty distinguishing the components and characteristics of different types of loads, which affects the accuracy of power system simulation analysis. Summary of the Invention

[0007] To address the above problems, this invention proposes a non-intrusive load identification and simulation modeling analysis method for power systems, comprising:

[0008] Key node data of the power distribution network in the power system are collected by non-intrusive measurement devices, and the key node data is preprocessed. Load characteristics are extracted and classified from the preprocessed key node data to obtain response characteristics.

[0009] Based on the response characteristics, static load and dynamic load are modeled and dynamic load is identified. Based on the modeling and identification results, a load composition identification matrix is ​​established to distinguish different types of loads.

[0010] For different types of loads, a multi-scale spatiotemporal graph attention network (MSTG-Attention) is used for intelligent load classification. Then, a simulation model including static loads and dynamic loads is established, and the simulation model is used for simulation verification.

[0011] Optional, key node data, including: electrical quantity data.

[0012] Optionally, the key node data may be preprocessed, including:

[0013] After denoising, filtering, and normalizing the key node data, the voltage amplitude and frequency disturbance information are extracted as load response excitation signals.

[0014] Optionally, load characteristics can be extracted from the preprocessed key node data, including: extracting the load response characteristics under voltage and frequency disturbances.

[0015] Optional static load modeling includes:

[0016] Based on the ZIP model framework, three static load components are identified: constant impedance Z, constant current I, and constant power P.

[0017] The proportion coefficients of each component are estimated using the least squares method to establish a static load mathematical model.

[0018] Optionally, a dynamic load model can be established for power electronic loads.

[0019] Optionally, a load composition identification matrix can be established to distinguish different types of loads, including:

[0020] Distinguish between traditional inductive loads such as motors, power electronic loads such as frequency converters and charging piles, and distributed power loads;

[0021] Calculate the proportion of each type of load in the total load;

[0022] Construct a load component database.

[0023] Optional, the multi-scale spatiotemporal graph attention network MSTG-Attention includes:

[0024] Adaptive multi-head attention mechanism, adversarial training to enhance classifier, dynamic weight fusion strategy and online incremental learning mechanism;

[0025] The adaptive multi-head attention mechanism includes a three-layer attention structure, specifically:

[0026] Time attention is used to identify load responses at critical moments;

[0027] Frequency attention is used to focus on important frequency components;

[0028] Spatial attention is used to highlight key electrical parameters;

[0029] Among them, adversarial training enhances the classifier, introducing a Generative Adversarial Network (GAN) framework, specifically including:

[0030] The generator G is used to generate load samples that are difficult to distinguish;

[0031] Discriminator D is used to distinguish between actual load and generated load;

[0032] Classifier C is used to train on adversarial examples to improve generalization ability;

[0033] Among them, the dynamic weight fusion strategy is used for dynamic weight allocation based on uncertainty estimation and real-time adjustment of the contribution weight of each sub-network.

[0034] Among them, the online incremental learning mechanism uses the experience replay mechanism to store historical samples.

[0035] Furthermore, this invention also proposes a non-intrusive load identification and simulation modeling analysis system for power systems, comprising:

[0036] The preprocessing unit is used to collect key node data of the power distribution network in the power system through a non-intrusive measurement device, preprocess the key node data, extract and classify load characteristics of the preprocessed key node data, and obtain response characteristics.

[0037] The modeling unit is used to model static loads and dynamic loads based on the response characteristics and to identify dynamic loads. Based on the modeling and identification results, a load composition identification matrix is ​​established to distinguish different types of loads.

[0038] The simulation verification unit is used to establish simulation models including static and dynamic loads after intelligent classification of loads using the Multi-Scale Spatiotemporal Graph Attention Network (MSTG-Attention) for different types of loads, and to perform simulation verification using the simulation models.

[0039] Optional, key node data, including: electrical quantity data.

[0040] Optionally, the key node data may be preprocessed, including:

[0041] After denoising, filtering, and normalizing the key node data, the voltage amplitude and frequency disturbance information are extracted as load response excitation signals.

[0042] Optionally, load characteristics can be extracted from the preprocessed key node data, including: extracting the load response characteristics under voltage and frequency disturbances.

[0043] Optional static load modeling includes:

[0044] Based on the ZIP model framework, three static load components are identified: constant impedance Z, constant current I, and constant power P.

[0045] The proportion coefficients of each component are estimated using the least squares method to establish a static load mathematical model.

[0046] Optionally, a dynamic load model can be established for power electronic loads.

[0047] Optionally, a load composition identification matrix can be established to distinguish different types of loads, including:

[0048] Distinguish between traditional inductive loads such as motors, power electronic loads such as frequency converters and charging piles, and distributed power loads;

[0049] Calculate the proportion of each type of load in the total load;

[0050] Construct a load component database.

[0051] Optional, the multi-scale spatiotemporal graph attention network MSTG-Attention includes:

[0052] Adaptive multi-head attention mechanism, adversarial training to enhance classifier, dynamic weight fusion strategy and online incremental learning mechanism;

[0053] The adaptive multi-head attention mechanism includes a three-layer attention structure, specifically:

[0054] Time attention is used to identify load responses at critical moments;

[0055] Frequency attention is used to focus on important frequency components;

[0056] Spatial attention is used to highlight key electrical parameters;

[0057] Among them, adversarial training enhances the classifier, introducing a Generative Adversarial Network (GAN) framework, specifically including:

[0058] The generator G is used to generate load samples that are difficult to distinguish;

[0059] Discriminator D is used to distinguish between actual load and generated load;

[0060] Classifier C is used to train on adversarial examples to improve generalization ability;

[0061] Among them, the dynamic weight fusion strategy is used for dynamic weight allocation based on uncertainty estimation and real-time adjustment of the contribution weight of each sub-network.

[0062] Among them, the online incremental learning mechanism uses the experience replay mechanism to store historical samples.

[0063] In another aspect, the present invention also provides a computing device, comprising: one or more processors;

[0064] A processor is used to execute one or more programs;

[0065] When the one or more programs are executed by the one or more processors, the method described above is implemented.

[0066] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method described above.

[0067] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0068] This invention proposes a non-intrusive load identification and simulation modeling analysis method for power systems, comprising: collecting key node data of the power distribution network through a non-intrusive measurement device, preprocessing the key node data, extracting and classifying load features from the preprocessed key node data to obtain response features; modeling and identifying static and dynamic loads based on the response features; establishing a load composition identification matrix based on the modeling and identification results to distinguish different types of loads; and establishing a simulation model including static and dynamic loads after intelligent load classification using a multi-scale spatiotemporal graph attention network (MSTG-Attention) for different types of loads, and performing simulation verification using the simulation model. This invention solves the problems of insufficient load identification accuracy and limited dynamic modeling capabilities in existing technologies. Attached Figure Description

[0069] Figure 1 This is a flowchart of the method of the present invention;

[0070] Figure 2 This is a flowchart illustrating the implementation of the method of the present invention;

[0071] Figure 3 This is a schematic diagram illustrating the load feature extraction implemented by the method of the present invention;

[0072] Figure 4 This is a diagram of the deep learning network structure implemented by the method of the present invention;

[0073] Figure 5 This is a diagram of the device system architecture for implementing the method of the present invention;

[0074] Figure 6 This is an example diagram showing the load classification results implemented by the method of the present invention;

[0075] Figure 7 This is a simulation verification comparison diagram of the implementation of the method of the present invention. Detailed Implementation

[0076] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0077] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0078] Example 1:

[0079] The invention proposes a non-intrusive load identification and simulation modeling analysis method for power systems, such as... Figure 1 As shown, it includes:

[0080] Step 1: Collect key node data of the power distribution network in the power system through a non-intrusive measurement device, preprocess the key node data, extract and classify load characteristics of the preprocessed key node data, and obtain response characteristics.

[0081] Step 2: Based on the response characteristics, model the static load and dynamic load, and identify the dynamic load. Based on the modeling and identification results, establish a load composition identification matrix to distinguish different types of loads.

[0082] Step 3: For different types of loads, after intelligent classification of the load using the Multi-Scale Spatiotemporal Graph Attention Network (MSTG-Attention), a simulation model including static loads and dynamic loads is established, and the simulation model is used for simulation verification.

[0083] The following provides further explanation of steps 1-3 above:

[0084] The implementation process of steps 1-3 above is as follows: Figure 2-7 As shown, it includes:

[0085] S1 Data Acquisition and Preprocessing:

[0086] Electrical quantity data such as voltage, current, and power at key nodes of the power distribution network are collected using non-intrusive measuring devices.

[0087] The collected data is preprocessed by denoising, filtering, and normalization.

[0088] Extract voltage amplitude and frequency disturbance information as load response excitation signals;

[0089] S2 Load Feature Extraction and Classification:

[0090] Based on waveform data analysis, the response characteristics of the load under voltage and frequency disturbances are extracted;

[0091] Construct a load frequency response characteristic fingerprint database, including:

[0092] Voltage sensitivity parameters (α, β exponents);

[0093] Frequency response time constant;

[0094] Dynamically adjust characteristic parameters;

[0095] Power factor variation pattern;

[0096] Deep learning neural networks are used to intelligently classify load types;

[0097] S3 static load modeling:

[0098] Based on the ZIP model framework, three static load components are identified: constant impedance (Z), constant current (I), and constant power (P).

[0099] The proportion coefficients of each component are estimated using the least squares method;

[0100] Establish a static load mathematical model:

[0101] P=P0[(αz(U / U0)2+αi(U / U0)+αp)]Q=Q0[(βz(U / U0)2+βi(U / U0)+βp)]

[0102] S4 Dynamic Load Identification and Modeling:

[0103] For power electronic loads, a dynamic load model is established:

[0104] Tp(dP / dt)+P=P s [1+Kpf(f-f0)+Kpv(V-V0)]Tq(dQ / dt)+Q=Q s [1+Kqf(f-f0)+Kqv(V-V0)]

[0105] Where Tp and Tq are time constants, Kpf and Kqf are frequency adjustment coefficients, and Kpv and Kqv are voltage adjustment coefficients;

[0106] Model parameters are determined using a parameter identification algorithm;

[0107] S5 Loading Component Analysis and Content Determination:

[0108] Establish a load characterization matrix to distinguish different types of loads:

[0109] Traditional inductive loads (electric motors);

[0110] Power electronic loads (frequency converters, charging piles, etc.);

[0111] Distributed power load;

[0112] Calculate the proportion of each type of load in the total load;

[0113] Construct a database of load components;

[0114] The S6 AI-optimized classification employs an innovative "Multi-Scale Spatiotemporal Graph Attention Network (MSTG-Attention)" for intelligent classification with high load.

[0115] S6.1 Multi-scale spatiotemporal feature extraction;

[0116] Construct multi-scale time windows: {1 period, 10 periods, 100 periods, 1000 periods};

[0117] Different feature extraction strategies are used for each scale:

[0118] Short-scale (1-10 cycles): Captures transient response characteristics; Medium-scale (10-100 cycles): Captures steady-state regulation characteristics; Long-scale (100-1000 cycles): Captures load mode characteristics.

[0119] Time-frequency domain features at various scales are extracted using wavelet transform;

[0120] S6.2 Graph Neural Network Load Correlation Modeling:

[0121] The load nodes are constructed as a graph structure, and the node characteristics include:

[0122] Voltage sensitivity parameter vector;

[0123] Frequency response eigenvector;

[0124] Power change mode vector;

[0125] Edge weights are calculated based on the correlation between loads:

[0126] W(i,j)=exp(-||F_i-F_j||2 / 2σ 2 )×Correlation(P_i,P_j)

[0127] Graph Convolutional Networks (GCNs) are used to learn the correlation characteristics between loads;

[0128] S6.3 Adaptive Multi-Head Attention Mechanism:

[0129] Design a three-layer attention structure:

[0130] Time attention: Identifying load responses at critical moments;

[0131] Frequency attention: Focusing on important frequency components;

[0132] Spatial attention: Highlighting key electrical parameters;

[0133] Attention weight adaptive adjustment formula:

[0134] α_t=softmax(W_a×tanh(W_t×H_t+b_t))β_f=softmax(W_b×tanh(W_f×F_f+b_f))γ_s=softmax(W_c×tanh(W_s×S_s+b_s))

[0135] S6.4 Adversarial Training Enhances Classifier:

[0136] Introducing the Generative Adversarial Network (GAN) framework:

[0137] Generator G: Generates load samples that are difficult to distinguish;

[0138] Discriminator D: Distinguishes between actual load and generated load;

[0139] Classifier C: Train on adversarial examples to improve generalization ability;

[0140] Ternary loss function:

[0141] L_total = L_classify + λ1L_adversarial + λ2L_regularization S6.5 Dynamic Weight Fusion Strategy:

[0142] Dynamic weight allocation based on uncertainty estimation:

[0143] w_i = exp(-U_i) / Σexp(-U_j)

[0144] Where U_i is the uncertainty measure of the i-th classifier;

[0145] Adjust the contribution weight of each sub-network in real time;

[0146] S6.6 Online Incremental Learning Mechanism:

[0147] Historical samples are stored using an experience replay mechanism;

[0148] Strategies to prevent catastrophic amnesia:

[0149] Elastic Weight Consolidation (EWC);

[0150] Gradient projection memory (GEM);

[0151] A fast adaptive learning algorithm for new load types;

[0152] S6.7 Load Type Sub-Classification:

[0153] Traditional load segmentation:

[0154] Induction motor loads (constant torque type, variable torque type);

[0155] Resistive loads (heaters, incandescent lamps);

[0156] Capacitive load (compensating capacitor);

[0157] Power electronic load segmentation:

[0158] PWM rectifier load (frequency converter, charging pile);

[0159] Switching power supply loads (LEDs, electronic devices);

[0160] Grid-connected inverter loads (photovoltaics, energy storage);

[0161] Intelligent load segmentation:

[0162] Demand response load:

[0163] Virtual power plant load aggregation;

[0164] Algorithm implementation framework of this invention:

[0165] class MSTGattentionClassifier:

[0166] def__init__(self):

[0167] self.multi_scale_extractor=MultiScaleExtractor()

[0168] self.graph_network=LoadGraphGCN()

[0169] self.attention_module=AdaptiveMultiHeadAttention()

[0170] self.adversarial_trainer=AdversarialTrainer()

[0171] self.uncertainty_estimator=UncertaintyEstimator()

[0172] def forward(self, load_data):

[0173] #Multi-scale feature extraction

[0174] multi_features=self.multi_scale_extractor(load_data)

[0175] #Graph Network Association Modeling

[0176] graph_features=self.graph_network(multi_features)

[0177] #Adaptive Attention

[0178] attended_features=self.attention_module(graph_features)

[0179] #Classification Prediction

[0180] predictions=self.classifier(attended_features)

[0181] #Uncertainty Assessment

[0182] uncertainty=self.uncertainty_estimator(predictions)

[0183] return predictions,uncertainty

[0184] S7 simulation modeling verification:

[0185] Establish a comprehensive simulation model that includes both static and dynamic loads;

[0186] Simulation verification was performed on the PSCAD / EMTDC or PowerFactory platform;

[0187] The accuracy of the model was verified by comparing it with actual measured data.

[0188] The technical advantages of this invention are as follows:

[0189] Multi-scale Spatiotemporal Graph Attention Network (MSTG-Attention) has the following innovative advantages compared to traditional methods:

[0190] 1. Multi-dimensional feature fusion:

[0191] Time dimension: Capturing the complete load response process from transient to steady state;

[0192] Spatial dimension: Considering the electrical connections and mutual influences between loads;

[0193] Frequency dimension: Identifying the load characteristics of different frequency components;

[0194] Amplitude dimension: Analyzes the differences in load characteristics at different operating points;

[0195] 2. Adaptive learning ability:

[0196] Attention weights are dynamically adjusted based on load operating status;

[0197] It has the ability to quickly learn new load types;

[0198] It can handle the time-varying and uncertain nature of load characteristics;

[0199] 3. Strong anti-interference ability:

[0200] Adversarial training improves robustness to noise and disturbances;

[0201] Multi-scale feature extraction reduces the impact of local disturbances;

[0202] Graph network structures utilize spatial correlation for noise reduction;

[0203] 4. High classification accuracy:

[0204] Theoretical analysis shows that the classification accuracy can reach over 98%;

[0205] The ability to distinguish between similar load types has been significantly improved;

[0206] Both the false positive rate and the false negative rate were less than 1%;

[0207] Spatiotemporal graph construction algorithm;

[0208] Algorithm: LoadGraphConstruction

[0209] Input: Load node set N = {n1, n2, ..., n} n Time series data T

[0210] Output: Load correlation diagram G = (V, E, W)

[0211] 1. Calculate the node feature matrix:

[0212] F_i = [voltage sensitivity, frequency response, power mode, harmonic characteristics]T

[0213] 2. Calculate edge weights:

[0214] W(i,j)=α·Similarity(F_i,F_j)+β·Correlation(P_i,P_j)+γ·Distance(L_i,L_j)

[0215] 3. Graph topology optimization:

[0216] The k-nearest neighbor algorithm is used to retain the k most relevant edges, reducing computational complexity;

[0217] Adaptive attention mechanism:

[0218] Multi-head attention calculation:

[0219] H_multi=Concat(head1,head2,...,head h )W^O

[0220] Calculations for each head:

[0221] head_i=Attention(QW_i^Q,KW_i^K,VW_i^V)

[0222] Adaptive weight adjustment:

[0223] W_adaptive=W_base×(1+α·Uncertainty+β·Confidence)

[0224] Where α and β are adaptive adjustment parameters that are dynamically adjusted according to the classification confidence level;

[0225] Uncertainty Quantification Mechanism:

[0226] Bayesian neural network layers:

[0227] Each weight parameter w ~ N(μ_w,σ_w) 2 )

[0228] Forecast uncertainty:

[0229] U_epistemic=Var[E[p(y|x,w)|D]]#Cognitive uncertainty

[0230] U_aleatoric=E[Var[p(y|x,w)]|D]# Random uncertainty

[0231] Total uncertainty:

[0232] U_total=U_epistemic+U_aleatoric

[0233] The application effect is as follows:

[0234] Test scenario: A 10kV distribution network in an industrial park, including the following loads:

[0235] Traditional motor load: 30 induction motors (power range 50-500kW)

[0236] Power electronic loads: 15 frequency converters, 8 charging piles, 20 sets of LED lighting

[0237] Distributed power generation: 5 sets of photovoltaic inverters, 3 sets of energy storage systems

[0238] A comparison of traditional methods versus innovative methods is shown below.

[0239]

[0240] The unique capabilities of this invention are as follows:

[0241] Fine-grained classification: able to distinguish different control modes of frequency converters (V / f control, vector control);

[0242] Operating status identification: Identify the operating status of the load, such as light load, full load, and overload;

[0243] Fault warning: Early warning of potential equipment faults based on changes in load characteristics;

[0244] Load forecasting: Predicting future load trends by combining historical data.

[0245] Example 2:

[0246] This invention also proposes a non-intrusive load identification and simulation modeling analysis system 200 for power systems, comprising:

[0247] The preprocessing unit 201 is used to collect key node data of the power distribution network in the power system through a non-intrusive measurement device, preprocess the key node data, extract and classify load characteristics of the preprocessed key node data, and obtain response characteristics.

[0248] Modeling unit 202 is used to model static load and dynamic load based on the response characteristics and identify dynamic load. Based on the modeling and identification results, a load composition identification matrix is ​​established to distinguish different types of load.

[0249] The simulation verification unit 203 is used to establish simulation models including static load and dynamic load after intelligent classification of load using the multi-scale spatiotemporal graph attention network MSTG-Attention for different types of load, and to perform simulation verification using the simulation models.

[0250] Key node data includes: electrical quantity data.

[0251] The preprocessing of the key node data includes:

[0252] After denoising, filtering, and normalizing the key node data, the voltage amplitude and frequency disturbance information are extracted as load response excitation signals.

[0253] Among them, load characteristics are extracted from the preprocessed key node data, including: extracting the load response characteristics under voltage and frequency disturbances.

[0254] Static load modeling includes:

[0255] Based on the ZIP model framework, three static load components are identified: constant impedance Z, constant current I, and constant power P.

[0256] The proportion coefficients of each component are estimated using the least squares method to establish a static load mathematical model.

[0257] Specifically, a dynamic load model is established for power electronic loads.

[0258] This includes establishing a load composition identification matrix to distinguish different types of loads, including:

[0259] Distinguish between traditional inductive loads such as motors, power electronic loads such as frequency converters and charging piles, and distributed power loads;

[0260] Calculate the proportion of each type of load in the total load;

[0261] Construct a load component database.

[0262] Among them, the multi-scale spatiotemporal graph attention network MSTG-Attention includes:

[0263] Adaptive multi-head attention mechanism, adversarial training to enhance classifier, dynamic weight fusion strategy and online incremental learning mechanism;

[0264] The adaptive multi-head attention mechanism includes a three-layer attention structure, specifically:

[0265] Time attention is used to identify load responses at critical moments;

[0266] Frequency attention is used to focus on important frequency components;

[0267] Spatial attention is used to highlight key electrical parameters;

[0268] Among them, adversarial training enhances the classifier, introducing a Generative Adversarial Network (GAN) framework, specifically including:

[0269] The generator G is used to generate load samples that are difficult to distinguish;

[0270] Discriminator D is used to distinguish between actual load and generated load;

[0271] Classifier C is used to train on adversarial examples to improve generalization ability;

[0272] Among them, the dynamic weight fusion strategy is used for dynamic weight allocation based on uncertainty estimation and real-time adjustment of the contribution weight of each sub-network.

[0273] Among them, the online incremental learning mechanism uses the experience replay mechanism to store historical samples.

[0274] This invention solves the problems of insufficient load identification accuracy and limited dynamic modeling capability in the prior art.

[0275] Example 3:

[0276] Based on the same inventive concept, this invention also provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement corresponding method flows or corresponding functions, thereby implementing the steps of the methods in the above embodiments.

[0277] Example 4:

[0278] Based on the same inventive concept, this invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the method in the above embodiments.

[0279] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0280] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0281] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0282] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0283] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0284] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A non-intrusive load identification and simulation modeling analysis method for power systems, characterized in that, include: Key node data of the power distribution network in the power system are collected by non-intrusive measurement devices, and the key node data is preprocessed. Load characteristics are extracted and classified from the preprocessed key node data to obtain response characteristics. Based on the response characteristics, static load and dynamic load are modeled and dynamic load is identified. Based on the modeling and identification results, a load composition identification matrix is ​​established to distinguish different types of loads. For different types of loads, a multi-scale spatiotemporal graph attention network (MSTG-Attention) is used for intelligent load classification. Then, a simulation model including static loads and dynamic loads is established, and the simulation model is used for simulation verification.

2. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, The key node data includes: electrical quantity data.

3. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, The preprocessing of the key node data includes: After denoising, filtering, and normalizing the key node data, the voltage amplitude and frequency disturbance information are extracted as load response excitation signals.

4. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, The extraction of load features from the preprocessed key node data includes: extracting the load response features under voltage and frequency disturbances.

5. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, Static load modeling includes: Based on the ZIP model framework, three static load components are identified: constant impedance Z, constant current I, and constant power P. The proportion coefficients of each component are estimated using the least squares method to establish a static load mathematical model.

6. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, A dynamic load model is established for power electronic loads.

7. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, Establish a load composition identification matrix to distinguish different types of loads, including: Distinguish between traditional inductive loads such as motors, power electronic loads such as frequency converters and charging piles, and distributed power loads; Calculate the proportion of each type of load in the total load; Construct a load component database.

8. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 1, characterized in that, The multi-scale spatiotemporal graph attention network MSTG-Attention includes: Adaptive multi-head attention mechanism, adversarial training to enhance classifier, dynamic weight fusion strategy and online incremental learning mechanism; The adaptive multi-head attention mechanism includes a three-layer attention structure, specifically: Time attention is used to identify load responses at critical moments; Frequency attention is used to focus on important frequency components; Spatial attention is used to highlight key electrical parameters; Among them, adversarial training enhances the classifier, introducing a Generative Adversarial Network (GAN) framework, specifically including: The generator G is used to generate load samples that are difficult to distinguish; Discriminator D is used to distinguish between actual load and generated load; Classifier C is used to train on adversarial examples to improve generalization ability; Among them, the dynamic weight fusion strategy is used for dynamic weight allocation based on uncertainty estimation and real-time adjustment of the contribution weight of each sub-network. Among them, the online incremental learning mechanism uses the experience replay mechanism to store historical samples.

9. A non-intrusive load identification and simulation modeling analysis system for power systems, characterized in that, include: The preprocessing unit is used to collect key node data of the power distribution network in the power system through a non-intrusive measurement device, preprocess the key node data, extract and classify load characteristics of the preprocessed key node data, and obtain response characteristics. The modeling unit is used to model static loads and dynamic loads based on the response characteristics and to identify dynamic loads. Based on the modeling and identification results, a load composition identification matrix is ​​established to distinguish different types of loads. The simulation verification unit is used to establish simulation models including static and dynamic loads after intelligent classification of loads using the Multi-Scale Spatiotemporal Graph Attention Network (MSTG-Attention) for different types of loads, and to perform simulation verification using the simulation models.

10. The non-intrusive load identification and simulation modeling analysis system for power systems according to claim 9, characterized in that, The key node data includes: electrical quantity data.

11. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, The preprocessing of the key node data includes: After denoising, filtering, and normalizing the key node data, the voltage amplitude and frequency disturbance information are extracted as load response excitation signals.

12. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, The extraction of load features from the preprocessed key node data includes: extracting the load response features under voltage and frequency disturbances.

13. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, Static load modeling includes: Based on the ZIP model framework, three static load components are identified: constant impedance Z, constant current I, and constant power P. The proportion coefficients of each component are estimated using the least squares method to establish a static load mathematical model.

14. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, A dynamic load model is established for power electronic loads.

15. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, Establish a load composition identification matrix to distinguish different types of loads, including: Distinguish between traditional inductive loads such as motors, power electronic loads such as frequency converters and charging piles, and distributed power loads; Calculate the proportion of each type of load in the total load; Construct a load component database.

16. The non-intrusive load identification and simulation modeling analysis method for power systems according to claim 9, characterized in that, The multi-scale spatiotemporal graph attention network MSTG-Attention includes: Adaptive multi-head attention mechanism, adversarial training to enhance classifier, dynamic weight fusion strategy and online incremental learning mechanism; The adaptive multi-head attention mechanism includes a three-layer attention structure, specifically: Time attention is used to identify load responses at critical moments; Frequency attention is used to focus on important frequency components; Spatial attention is used to highlight key electrical parameters; Among them, adversarial training enhances the classifier, introducing a Generative Adversarial Network (GAN) framework, specifically including: The generator G is used to generate load samples that are difficult to distinguish; Discriminator D is used to distinguish between actual load and generated load; Classifier C is used to train on adversarial examples to improve generalization ability; Among them, the dynamic weight fusion strategy is used for dynamic weight allocation based on uncertainty estimation and real-time adjustment of the contribution weight of each sub-network. Among them, the online incremental learning mechanism uses the experience replay mechanism to store historical samples.

17. A computer device, characterized in that, include: One or more processors; A processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the method described in any one of claims 1-8 is implemented.

18. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the method as described in any one of claims 1-8.

Images