A comprehensive decomposition and identification method for total load regulation in virtual power plants

Abstract

The present invention discloses a comprehensive decomposition and identification method for regulating the total load of a virtual power plant, comprising the following steps: step (1): collecting multidimensional load data in the virtual power plant through multi-type hardware monitoring equipment; step (2): preprocessing the multidimensional load data to obtain a standardized load data set; step (3): decomposing the standardized load data set based on principal component analysis and a dynamic weight allocation algorithm to obtain characteristic load components; step (4): extracting time-frequency domain features of the characteristic load components using wavelet transform to construct a multidimensional load characteristic vector; step (5): combining equipment operating status and user behavior parameters, and identifying the adjustability of the multidimensional load characteristic vector using a dynamic threshold clustering algorithm. The present invention constructs a multidimensional monitoring system, realizes dynamic feature extraction and identification, and performs precise scheduling and closed-loop control.

Description

Technical Field

[0001] The present invention relates to the field of virtual power plants, and in particular to a comprehensive decomposition and identification method for regulating the total load of a virtual power plant. Background Art

[0002] With the large-scale integration of distributed energy resources, virtual power plants (VPPs), as a key technology for integrating distributed resources, must achieve precise control of diverse loads. The core challenge lies in the complexity of the load structure: VPPs encompass multiple user types, including industrial, commercial, and residential users, each with significantly varying load characteristics. Traditional single-parameter monitoring cannot fully characterize these loads.

[0003] Timeliness of regulation response: The power grid has higher requirements for real-time response on the demand side, but the existing load decomposition algorithm cannot dynamically adapt to changes in equipment status and differences in user behavior.

[0004] Multi-source data fusion requirements: Load regulation potential depends not only on electrical parameters but also on the physical state of the equipment and environmental factors. Traditional data collection and analysis architectures lack the ability to process the spatiotemporal correlation of multi-dimensional data.

[0005] The existing technology has problems such as the limitation of single device monitoring, insufficient communication and synchronization accuracy, insufficient adaptability of static algorithms and single feature extraction dimension. Therefore, a comprehensive decomposition and identification method for total load regulation of virtual power plants is proposed. Summary of the Invention

[0006] The technical problem to be solved by the present invention is: how to solve the problems existing in the existing technology, such as the limitations of single device monitoring, insufficient communication and synchronization accuracy, insufficient adaptability of static algorithms and single feature extraction dimension, and provide a comprehensive decomposition and identification method for regulating the total load of a virtual power plant.

[0007] The present invention solves the above technical problems through the following technical solutions, which include the following steps:

[0008] Step (1): Collect multi-dimensional load data in the virtual power plant through multiple types of hardware monitoring equipment;

[0009] Step (2): Preprocess the multidimensional load data to obtain a standardized load data set;

[0010] Step (3): Decompose the load of the standardized load data set based on principal component analysis and dynamic weight allocation algorithm to obtain characteristic load components;

[0011] Step (4): using wavelet transform to extract the time-frequency domain features of the characteristic load components and construct a multi-dimensional load feature vector;

[0012] Step (5): Combine the equipment operation status and user behavior parameters to identify the adjustability of the multi-dimensional load feature vector through the dynamic threshold clustering algorithm;

[0013] Step (6): Generate and execute the virtual power plant optimization scheduling strategy based on the identification results.

[0014] Furthermore, the multiple types of hardware monitoring devices in step (1) include:

[0015] Smart meter arrays deployed at the user end and at the incoming line end of preset key equipment are used to collect voltage, current, power and power factor parameters;

[0016] The transformer group installed on the low-voltage side of the transformer and the main feeder circuit is used to measure the line current and voltage;

[0017] Temperature sensors placed on or inside preset key equipment, as well as environmental sensors covering different environmental areas, are used to collect equipment operating temperature, ambient temperature, humidity, and light intensity parameters;

[0018] The multi-dimensional load data is synchronously aggregated according to a unified timestamp through a data acquisition terminal to form a spatiotemporal correlation data set containing electrical parameters, equipment status parameters, and environmental parameters;

[0019] The data acquisition terminal has a hardware self-check function, which monitors the communication status of the equipment and the data acquisition accuracy by periodically sending a verification signal.

[0020] Furthermore, the pre-processing in step (2) includes:

[0021] Data cleaning: Statistical analysis and wavelet denoising algorithm are used to remove abnormal data points, and the mean μ of the data sequence is calculated. x And standard deviation σ identifies the mutation point, and performs wavelet transform on the noisy signal, through the threshold function Reconstruct the signal after processing the wavelet coefficients;

[0022] The threshold function is , where is the wavelet coefficient after wavelet transformation, is a sign function used to preserve the sign of the wavelet coefficients, is the adaptive threshold, which is used to distinguish the critical value of the effective signal coefficient from the noise coefficient;

[0023] Data normalization: The normalization method is used to convert data of different dimensions into dimensionless relative values. The specific formula is:

[0024] Maximum and minimum normalization: , where is the original data point, and is the minimum and maximum value of the data sequence;

[0025] Z-score normalization: .

[0026] Furthermore, the specific process of load decomposition in step (3) includes:

[0027] Perform principal component analysis on the preprocessed load data set and calculate the eigenvalue λi and eigenvector ei of the covariance matrix C;

[0028] The calculation process of the variance matrix C is:

[0029] ;

[0030] in, is the number of data samples, The original data vector of the i-th sample;

[0031] Extracting cumulative variance contribution rate The first k principal components of form the principal component matrix ;

[0032] in, Represents the total number of features of the original load data, that is, the covariance matrix The dimension corresponds to the number of all variables in the data set, such as voltage, current, power, temperature and other multi-dimensional monitoring parameters. is the number of principal components extracted;

[0033] Based on the equipment type and historical adjustment data, the principal component weight ωi is dynamically adjusted so that the principal component after weight distribution satisfies , and obtain the orthogonal characteristic load components .

[0034] Furthermore, the wavelet transform feature extraction process in step (4) includes:

[0035] For each characteristic load component Perform J-layer discrete wavelet transform to obtain detail coefficients wj(t) and approximate coefficients aj(t) of different frequency bands;

[0036] Calculate the wavelet energy value of each frequency band , construct a multidimensional feature vector containing time domain fluctuation characteristics and frequency domain energy distribution ;

[0037] is the frequency band energy value of the j-th layer wavelet transform, reflecting the energy distribution of the signal in this frequency band;

[0038] is the detail coefficient of the j-th layer wavelet transform, describing the high-frequency component of the signal;

[0039] T is the number of sampling points of a single sample and the length of the time series;

[0040] is the total number of wavelet transform layers.

[0041] Furthermore, the adjustability identification process in step (5) includes:

[0042] Build in load change rates , equipment adjustment margin and user response levels

[0043] The evaluation index system includes is the real-time power, is the historical average power, and Adjust the upper and lower limits of the equipment power. is the rated power;

[0044] The load change rate Reflect real-time power and historical average power Relative change of equipment adjustment margin Reflects the power adjustment range of the equipment, that is, the upper and lower limits of the equipment power adjustment and the rated power The proportional relationship between the user response level Comprehensive assessment of the user's historical behavioral characteristics in demand response;

[0045] The K-means algorithm is used to jointly cluster the feature vector E and the evaluation index, and the optimal number of clusters K is determined by the elbow rule, and the initial threshold corresponding to the maximum distance between classes is calculated. , and based on the user response level Dynamic correction threshold is obtained : ,in is the adjustment coefficient, which takes +0.2 when the user response level is high, 0 when it is medium, and -0.2 when it is low;

[0046] when When it is determined to be an adjustable load, is the corresponding indicator weight coefficient.

[0047] Furthermore, the generation of the optimized scheduling strategy in step (6) includes:

[0048] Establish an optimization model with the goal of minimizing scheduling costs:

[0049] ;

[0050] Considering the distributed energy generation cost and the adjustable load regulation cost, the power regulation range is set as a constraint condition. The constraint condition is:

[0051] ;

[0052] ;

[0053] in, For the The distributed energy generation cost at the moment, For the The adjustable load adjustment cost at each moment, To regulate power for power generation, is the load regulation power, N is the number of time points in the scheduling cycle, and is the preset maximum and minimum thresholds;

[0054] The optimization model is solved through intelligent optimization algorithms to generate the optimal scheduling plan and send it to each device through the communication network. The communication network supports hybrid wired and wireless transmission, and the data acquisition terminal has local caching and breakpoint resumption functions.

[0055] Furthermore, the user response level η is evaluated by a fuzzy logic algorithm, which takes the user's historical frequency of participation in demand response, adjustment amplitude and response time as input variables, and maps them into three adjustable levels of high, medium and low through a membership function, providing a basis for user-side adjustment potential for adjustability identification of multi-dimensional load characteristic vectors.

[0056] Furthermore, the installation locations of the hardware monitoring equipment are as follows: the smart meter is installed in the user's meter box or the preset key equipment incoming line end, the mutual inductor group is installed at the preset key position of the low-voltage side of the transformer and the main feeder loop, the temperature sensor is attached to the heating part of the preset key equipment, and the environmental sensors are distributed in different environmental areas to cover the monitoring range.

[0057] Compared with the existing technology, the present invention has the following advantages: the comprehensive decomposition and identification method of the total load of the virtual power plant is used to collect multi-dimensional load data through multiple types of hardware monitoring equipment to form a spatiotemporal correlation data set, which improves the comprehensiveness and reliability of load data monitoring; the preprocessing of data cleaning, normalization and principal component analysis, wavelet transform and other algorithms are used to realize the feature decomposition and spatiotemporal feature extraction of load data, thereby improving the accuracy and efficiency of load analysis; the load adjustability is identified by combining equipment operating status, user behavior parameters and dynamic threshold clustering algorithm, thereby enhancing the dynamic adaptability and identification accuracy of load regulation potential; based on the identification results, an optimized scheduling strategy is generated, with cost minimization as the goal and combined with intelligent algorithm solution, thereby improving the economy and operation efficiency of virtual power plant resource scheduling; the reasonable deployment of hardware equipment and the self-test function of the data acquisition terminal ensure the stability of data acquisition and the robustness of system operation; the user response level evaluation mechanism is introduced, the user-side regulation potential is integrated, and the comprehensive decision-making ability of load regulation is optimized, making the system more worthy of promotion and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is the overall structural diagram of the present invention. DETAILED DESCRIPTION

[0059] The following is a detailed description of an embodiment of the present invention. This embodiment provides a detailed implementation method and a specific operation process based on the technical solution of the present invention, but the protection scope of the present invention is not limited to the following embodiment.

[0060] like Figure 1 As shown, this embodiment provides a technical solution: a comprehensive decomposition and identification method for regulating the total load of a virtual power plant, comprising the following steps:

[0061] Step (1): Collect multi-dimensional load data in the virtual power plant through multiple types of hardware monitoring equipment;

[0062] Step (2): Preprocess the multidimensional load data to obtain a standardized load data set;

[0063] Step (3): Decompose the load of the standardized load data set based on principal component analysis and dynamic weight allocation algorithm to obtain characteristic load components;

[0064] Step (4): using wavelet transform to extract the time-frequency domain features of the characteristic load components and construct a multi-dimensional load feature vector;

[0065] Step (5): Combine the equipment operation status and user behavior parameters to identify the adjustability of the multi-dimensional load feature vector through the dynamic threshold clustering algorithm;

[0066] Step (6): Generate and execute the virtual power plant optimization scheduling strategy based on the identification results.

[0067] The multiple types of hardware monitoring devices in step (1) include:

[0068] Smart meter arrays deployed at the user end and at the incoming line end of pre-set key equipment are used to collect voltage, current, power, and power factor parameters. Pre-set key equipment includes industrial motors, air conditioning units, energy storage devices, and other electrical equipment that have a significant impact on the load regulation of the virtual power plant;

[0069] The transformer group installed on the low-voltage side of the transformer and the main feeder circuit is used to measure the line current and voltage;

[0070] Temperature sensors placed on or inside preset key equipment, as well as environmental sensors covering different environmental areas, are used to collect equipment operating temperature, ambient temperature, humidity, and light intensity parameters;

[0071] The multi-dimensional load data is synchronously aggregated according to a unified timestamp through a data acquisition terminal to form a spatiotemporal correlation data set containing electrical parameters, equipment status parameters, and environmental parameters;

[0072] The data acquisition terminal has a hardware self-check function, which monitors the communication status of the equipment and the data acquisition accuracy by periodically sending a verification signal;

[0073] Through smart meter arrays, mutual inductor groups, temperature sensors and environmental sensors, covering electrical parameters, equipment status and environmental parameters, full-dimensional monitoring of load data in virtual power plants is achieved, providing a rich data foundation for subsequent analysis. Multi-dimensional data is synchronously aggregated through unified timestamps to form a data set containing time and space associations, ensuring the temporal consistency and spatial correlation of different types of data, facilitating subsequent cross-parameter joint analysis and load characteristic modeling. The data acquisition terminal has a hardware self-test function. By periodically verifying the communication status and data acquisition accuracy of the signal monitoring equipment, it can detect equipment anomalies or data distortion in real time, improve the stability and data credibility of the entire monitoring system, and provide a reliable basis for subsequent load decomposition and identification. Various hardware devices are deployed in key locations according to monitoring needs to achieve accurate monitoring of core equipment and key nodes in the virtual power plant, avoid redundancy or blind spots in data acquisition, and improve monitoring efficiency and resource utilization.

[0074] The pre-processing in step (2) includes:

[0075] Data cleaning: Statistical analysis and wavelet denoising algorithm are used to remove abnormal data points, and the mean μ of the data sequence is calculated. x And standard deviation σ identifies the mutation point, and performs wavelet transform on the noisy signal, through the threshold function Reconstruct the signal after processing the wavelet coefficients;

[0076] The threshold function is , where is the wavelet coefficient after wavelet transformation, is a sign function used to preserve the sign of the wavelet coefficients, is the adaptive threshold, which is used to distinguish the critical value of the effective signal coefficient from the noise coefficient;

[0077] Data normalization: The normalization method is used to convert data of different dimensions into dimensionless relative values. The specific formula is:

[0078] Maximum and minimum normalization: , where is the original data point, and is the minimum and maximum value of the data sequence;

[0079] Z-score normalization: ;

[0080] in It is the standardized data vector after maximum and minimum normalization / Z-score normalization;

[0081] The above process can eliminate abnormal data points, such as sudden changes caused by equipment failure and noise caused by communication interference, to prevent dirty data from misleading the load decomposition and identification results.

[0082] If there is an instantaneous spike in the current data collected by the smart meter, which is not a real load fluctuation, the mean μ of the current data sequence is calculated. x and standard deviation σ, can identify The abnormal points in the range are eliminated to ensure that the subsequent analysis is based on the real load characteristics;

[0083] Data of different dimensions, such as voltage, temperature, and power, cannot be directly compared. After normalization, they are converted into dimensionless values, enabling principal component analysis and clustering algorithms to effectively extract features.

[0084] Maximum and minimum normalization: The power value range of a device is 0 to 100kW. The original value of 60kW is normalized to ;

[0085] Core normalization: If the mean value of a set of voltage data μ x =220V, standard deviation =5V, the voltage 220V at a certain moment is normalized to ;

[0086] Wavelet denoising via adaptive thresholding Distinguish effective signals from noise, remove high-frequency noise while retaining low-frequency useful signals, and avoid the destruction of features by traditional filtering methods.

[0087] The specific process of load decomposition in step (3) includes:

[0088] Perform principal component analysis on the preprocessed load data set and calculate the eigenvalue λi and eigenvector ei of the covariance matrix C;

[0089] The calculation process of the variance matrix C is:

[0090] ;

[0091] in, is the number of data samples, The original data vector of the i-th sample, μ is the mean vector of all samples, which is composed of the mean of the data in each dimension and reflects the central tendency of the sample data in each dimension;

[0092] is a vector The transpose of is used to calculate the covariance between the data of each dimension through matrix multiplication, and finally construct the covariance matrix C that reflects the correlation between the dimensions of the multidimensional load data;

[0093] Extracting cumulative variance contribution rate The first k principal components of form the principal component matrix ;

[0094] in, Represents the total number of features of the original load data, that is, the covariance matrix The dimension corresponds to the number of all variables in the data set, such as voltage, current, power, temperature and other multi-dimensional monitoring parameters. is the number of principal components extracted;

[0095] Dynamically adjust the principal component weight ωi based on the equipment type and historical adjustment data,

[0096] The specific process of dynamically adjusting the principal component weight ωi based on equipment type and historical adjustment data is as follows:

[0097] Preset initial weight matrix based on device type , combined with the root mean square value RMSE of the historical adjustment error to construct the correction coefficient, through the iterative formula: , update the weights, where is the correction factor (range 0.01~0.1), is the root mean square error of the adjustment of principal component i at the tth iteration;

[0098] is the measured characteristic load component, is the predicted value;

[0099] The principal components after weight distribution satisfy , and obtain the orthogonal characteristic load components ;

[0100] Through principal component analysis, multidimensional load data such as voltage, current, power, and temperature are compressed into a few unrelated characteristic load components, eliminating redundant information and reducing the computational complexity of subsequent algorithms.

[0101] If the original data contains 10 load characteristics, the first three principal components can be extracted through PCA, and the cumulative variance contribution rate is ≥90%. The data dimension can be reduced from 10 dimensions to 3 dimensions, significantly reducing the amount of calculation.

[0102] In this case, weights are dynamically adjusted based on equipment type and historical regulation data, so that the decomposition results are closer to the actual load characteristics of different equipment.

[0103] For example, for air-conditioning equipment, the load is highly correlated with temperature. Dynamic weight allocation will increase the weight of the temperature-related principal component, highlighting the impact of temperature on load fluctuations.

[0104] For lighting equipment, its load is highly correlated with light intensity, and the weight allocation will focus on the principal components related to light intensity and suppress interference from irrelevant variables.

[0105] The characteristic load components obtained after dynamic weight allocation can correspond to actual physical meanings, such as fundamental frequency load, high-frequency fluctuating load, etc., which facilitates the subsequent analysis of load regulation potential.

[0106] For example, after weight adjustment, a principal component mainly reflects the impact load during the startup of an industrial motor. The weights of the current and power parameters in its eigenvector are significantly higher than those of other parameters, and can be directly used to identify impact loads with low adjustability.

[0107] The wavelet transform feature extraction process in step (4) includes:

[0108] For each characteristic load component Perform J-layer discrete wavelet transform to obtain detail coefficients wj(t) and approximate coefficients aj(t) of different frequency bands;

[0109] Calculate the wavelet energy value of each frequency band , construct a multidimensional feature vector containing time domain fluctuation characteristics and frequency domain energy distribution ;

[0110] is the frequency band energy value of the j-th layer wavelet transform, reflecting the energy distribution of the signal in this frequency band; is the detail coefficient of the j-th layer wavelet transform, describing the high-frequency components of the signal; T is the number of sampling points of a single sample and the length of the time series; is the total number of layers of wavelet transform;

[0111] Joint analysis in time and frequency domains to capture non-stationary load characteristics;

[0112] Wavelet transform can simultaneously analyze the time characteristics and frequency characteristics of load data, and is suitable for feature extraction of non-stationary loads such as electric vehicle charging and renewable energy grid connection in virtual power plants.

[0113] If a short-term peak appears in the load curve during a certain period, such as when the air conditioner is started, after decomposition by wavelet transform, the high-frequency band, such as the first layer detail coefficient Energy value Significant increase allows for quick identification of transient load events.

[0114] Through multi-layer wavelet decomposition (such as J=3), the load signal can be decomposed into features of different time scales, such as high frequency corresponding to minute-level fluctuations and low frequency corresponding to hour-level trends, which makes it easier to distinguish the characteristic differences between adjustable loads and non-adjustable loads.

[0115] For example, the start and stop of industrial motors will be in the medium and high frequency bands, such as the second layer detail coefficient Energy peaks are generated, and residential basic electricity consumption (such as continuous operation of refrigerators) is mainly concentrated in the low frequency band, such as the approximate coefficient , the energy distribution is stable, using the energy value of each frequency band The regulation potential of the two types of loads can be effectively distinguished.

[0116] The multidimensional feature vector E integrates the energy distribution information of different frequency bands. Compared with single time domain or frequency domain features, it can more comprehensively characterize the dynamic characteristics of the load and improve the identification accuracy of subsequent clustering algorithms.

[0117] For example, after wavelet transforming two sets of load data, if the high-frequency energy value of one set is Higher and low frequency energy values By comparing the difference of the multidimensional vector E, we can quickly determine that the former is more likely to be an interruptible load and the latter is a base load.

[0118] The adjustability identification process in step (5) includes:

[0119] Build in load change rates , equipment adjustment margin and user response levels

[0120] The evaluation index system includes is the real-time power, is the historical average power, and Adjust the upper and lower limits of the equipment power. is the rated power;

[0121] The load change rate Reflect real-time power and historical average power Relative change of equipment adjustment margin Reflects the power adjustment range of the equipment, that is, the upper and lower limits of the equipment power adjustment and the rated power The proportional relationship between the user response level Comprehensive assessment of the user's historical behavioral characteristics in demand response;

[0122] The K-means algorithm is used to jointly cluster the feature vector E and the evaluation index, and the optimal number of clusters K is determined by the elbow rule, and the initial threshold corresponding to the maximum distance between classes is calculated. , and based on the user response level Dynamic correction threshold is obtained : ,in is the adjustment coefficient, which takes +0.2 when the user response level is high, 0 when it is medium, and -0.2 when it is low;

[0123] when When it is determined to be an adjustable load, is the corresponding indicator weight coefficient;

[0124] Integrate equipment operating status and user behavior characteristics to avoid the one-sidedness of a single indicator and ensure that adjustable load identification is consistent with both equipment physical characteristics and user engagement;

[0125] Such as an industrial equipment =0.3, large load fluctuation, =0.6, the adjustment margin is high, but the user response level Low, that is, the historical participation is low. After comprehensive evaluation, it may be determined that it is adjustable but requires careful scheduling;

[0126] Air conditioning system for a commercial user =0.15, =0.4, but If it is high, it may be prioritized as a high adjustable potential load;

[0127] Automatically determine the optimal number of clusters K through the elbow rule to avoid the subjectivity of manually setting parameters; dynamically adjust the threshold based on the user response level , so that the recognition standard can be flexibly adjusted according to the user's cooperation;

[0128] When the user response level High, threshold Reduce by 20%, , expand the adjustable load determination range and make full use of the advantage of high user cooperation;

[0129] like is low, the threshold Increase by 20%, strictly screen adjustable loads, and avoid scheduling execution risks.

[0130] The K-means algorithm automatically groups data based on its features and can effectively identify the inherent correlations among load features. It is suitable for mixed scenarios of multiple types of loads (such as residential, industrial, and commercial) in virtual power plants.

[0131] For example, residential users' water heaters and industrial motors are automatically divided into different clusters, with the former being determined as shiftable loads and the latter as non-adjustable base loads.

[0132] The generation of the optimized scheduling strategy in step (6) includes:

[0133] Establish an optimization model with the goal of minimizing scheduling costs:

[0134] ;

[0135] Considering the distributed energy generation cost and the adjustable load regulation cost, the power regulation range is set as a constraint condition. The constraint condition is:

[0136] ;

[0137] ;

[0138] in, For the The distributed energy generation cost at the moment, For the The adjustable load adjustment cost at each moment, To regulate power for power generation, is the load regulation power, n is the number of time points in the scheduling cycle;

[0139] Refers to the minimum adjustable power (lower limit) of the adjustable load at the nth moment, which is the minimum value of the preset load adjustment power constraint range;

[0140] Refers to the maximum adjustable power (upper limit) of the adjustable load at the nth moment, which is the maximum value of the preset load adjustment power constraint range;

[0141] Refers to the minimum power generation of distributed energy at time n, which is the lower limit of the power generation of distributed energy in the constraint conditions, ensuring that the power generation is not lower than this value to maintain stable operation of the equipment or meet basic power supply needs;

[0142] Refers to the maximum power generation of distributed energy at the nth moment. It is the upper limit of the power generation of distributed energy in the constraint conditions, preventing the power generation from exceeding the rated capacity of the equipment or the acceptance capacity of the grid, avoiding safety issues such as overload;

[0143] The optimization model is solved by intelligent optimization algorithms to generate the optimal scheduling plan and send it to each device through the communication network. The communication network supports hybrid wired and wireless transmission, and the data collection terminal has local caching and breakpoint resume functions.

[0144] By coordinating the cost differences between distributed energy and adjustable loads, low-cost regulation resources are given priority to minimize global costs.

[0145] For example, if the PV power price is 0.3 yuan / kWh (low cost) during a certain period, but the output is insufficient, and the adjustment cost of the interruptible industrial load is 0.2 yuan / kWh (lower than the purchase electricity cost of 0.5 yuan / kWh), the optimization model will prioritize reducing industrial load during this period, reducing the purchase electricity demand and lowering the total cost.

[0146] Mandatory upper and lower limits for power generation and load regulation are set to prevent over-generation of distributed energy or load regulation from causing abnormal grid voltage / frequency, thus ensuring stable operation of the system.

[0147] If the maximum output of a distributed power source is 500kW, then , the optimization model will not schedule its output to exceed this value;

[0148] If the adjustable power range of the air-conditioning load is ±20% of the rated power, that is, , , the model only adjusts the load within this range to avoid equipment damage.

[0149] Intelligent optimization algorithms can quickly handle multi-variable and nonlinear scheduling problems and adapt to dynamic scenarios such as distributed energy fluctuations and user load changes in virtual power plants.

[0150] For example, when a sudden rainstorm causes a sharp drop in photovoltaic output, the particle swarm optimization algorithm can recalculate the optimal scheduling plan within minutes, quickly increase the energy storage discharge power, activate interruptible loads, maintain power balance, and avoid the lag of traditional rule-based scheduling.

[0151] A hybrid wired and wireless communication network is used to issue scheduling plans, and the local cache and breakpoint resume functions of the data acquisition terminal are used to ensure that instructions can still be reliably executed during network fluctuations.

[0152] When the wireless signal in a certain area is temporarily interrupted, the data collection terminal automatically uses the latest scheduling plan cached locally to control the equipment, and automatically synchronizes the data after the network is restored to avoid scheduling failures caused by communication failures.

[0153] The user response level η is evaluated using a fuzzy logic algorithm, which uses the user's historical frequency of demand response participation, adjustment amplitude, and response time as input variables. The membership function is used to map the user response level into three adjustable levels: high, medium, and low. This provides a basis for user-side adjustment potential for adjustability identification of multi-dimensional load characteristic vectors.

[0154] Integrating the behavioral parameters such as the frequency adjustment amplitude and response time of users' historical participation in demand response, the user response level is evaluated through fuzzy logic algorithm, providing a user-side supplementary user-side adjustment potential basis for the adjustability identification of multi-dimensional load characteristic vectors, improving the comprehensiveness of identification, and using fuzzy logic algorithm to deal with the ambiguity of user behavior. The qualitative description is converted into three adjustable levels of high, medium and low through the membership function, solving the mechanical problem of traditional threshold judgment and making the evaluation results more in line with reality. Based on the user response level, differentiated scheduling of virtual power plants is realized, high-level users are given priority in allocating adjustment tasks, and low-level users are dispatched cautiously, thereby improving the overall adjustment efficiency and execution reliability. An incentive mechanism is formed through the evaluation of users' historical behavior to promote users to actively participate in demand response, enhance the interactivity and coordination between virtual power plants and users, and promote the healthy operation of the system.

[0155] The installation locations of the hardware monitoring equipment are as follows: smart meters are installed in user meter boxes or preset key equipment inlet terminals, such as industrial motor power inlets and air conditioning unit distribution boxes; mutual inductors are installed on the low-voltage side of transformers and preset key locations of main feeder circuits, such as trunk branch nodes and large-capacity load access points; temperature sensors are attached to heat-generating parts of preset key equipment, such as motor windings and air conditioning compressor casings; and environmental sensors are distributed in different environmental areas to cover the monitoring range;

[0156] The preset key locations include but are not limited to the low-voltage side busbar of the transformer, the main feeder loop nodes, the line entry point of the user-side distribution box and the power input terminal of the equipment, and other key nodes that affect power transmission and load monitoring;

[0157] Smart meters, installed in user meter boxes or at the incoming line terminals of key equipment, can directly collect electrical parameters at the user end and at the equipment entrance, ensuring that the data directly reflects the terminal load characteristics and avoiding transmission losses or interference from intermediate links. Transformer groups, deployed on the low-voltage side of transformers and key locations on major feeder circuits, can monitor the current and voltage status of the grid trunk lines in real time, facilitating the understanding of the overall power flow of the virtual power plant and promptly detecting line overloads or abnormal fluctuations. Temperature sensors, located on the heat-generating areas of key equipment, can directly monitor the temperature of core components, providing early warning of overheating failures caused by poor heat dissipation and poor contact, thereby reducing the risk of equipment damage. Environmental sensors, distributed in different environmental areas, can collect multi-dimensional environmental data such as temperature, humidity, and light intensity. This data is used to analyze the impact of environmental factors on load, such as the correlation between temperature changes and air conditioning load, and the relationship between light intensity and photovoltaic output, thereby improving the scientific nature of load forecasting and regulation strategies. Various sensors are deployed in key locations, heat-generating areas, and different environmental areas according to monitoring requirements, ensuring that there are no blind spots in the monitoring of electrical parameters, equipment status, and environmental conditions within the virtual power plant. This creates a temporally and spatially coherent data set, providing comprehensive data support for load decomposition, identification, and scheduling.

[0158] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0159] In the description of this specification, the reference terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0160] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A comprehensive decomposition and identification method for regulating the total load of a virtual power plant, characterized in that: The following steps are involved: Step (1): Collect multi-dimensional load data in the virtual power plant through multiple types of hardware monitoring equipment; Step (2): Preprocess the multidimensional load data to obtain a standardized load data set; Step (3): Decompose the load of the standardized load data set based on principal component analysis and dynamic weight allocation algorithm to obtain characteristic load components; Step (4): using wavelet transform to extract the time-frequency domain features of the characteristic load components and construct a multi-dimensional load feature vector; Step (5): Combine the equipment operation status and user behavior parameters to identify the adjustability of the multi-dimensional load feature vector through the dynamic threshold clustering algorithm; The adjustability identification process includes: Build in load change rates , equipment adjustment margin and user response levels Evaluation index system; in is the real-time power, is the historical average power, and Adjust the upper and lower limits of the equipment power. is the rated power; The K-means algorithm is used to jointly cluster the feature vector E and the evaluation index, and the optimal number of clusters K is determined by the elbow rule, and the initial threshold corresponding to the maximum distance between classes is calculated. , and based on the user response level Dynamic correction threshold is obtained ; Step (6): Generate and execute the virtual power plant optimization scheduling strategy based on the identification results, including: Establish an optimization model with the goal of minimizing scheduling costs, consider the distributed energy generation costs and the adjustable load adjustment costs, and set the power adjustment range as a constraint; The user response level η is evaluated through a fuzzy logic algorithm, taking the user's historical frequency of demand response participation, adjustment amplitude, and response time as input variables. It is mapped into three adjustable levels of high, medium, and low through a membership function, providing a basis for user-side adjustment potential for adjustability identification of multi-dimensional load characteristic vectors.

2. A comprehensive decomposition and identification method for regulating total load of a virtual power plant according to claim 1, characterized in that: The multiple types of hardware monitoring devices in step (1) include: Smart meter arrays deployed at the user end and at the incoming line end of preset key equipment are used to collect voltage, current, power and power factor parameters; The transformer group installed on the low-voltage side of the transformer and the main feeder circuit is used to measure the line current and voltage; Temperature sensors placed on or inside preset key equipment, as well as environmental sensors covering different environmental areas, are used to collect equipment operating temperature, ambient temperature, humidity, and light intensity parameters; The multi-dimensional load data is synchronously aggregated according to a unified timestamp through a data acquisition terminal to form a spatiotemporal correlation data set containing electrical parameters, equipment status parameters, and environmental parameters; The data acquisition terminal has a hardware self-check function, which monitors the communication status of the equipment and the data acquisition accuracy by periodically sending a verification signal.

3. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 2, characterized in that: The pre-processing in step (2) includes: Data cleaning: Statistical analysis and wavelet denoising algorithm are used to remove abnormal data points, and the mean μ of the data sequence is calculated. x And standard deviation σ identifies the mutation point, and performs wavelet transform on the noisy signal, through the threshold function Reconstruct the signal after processing the wavelet coefficients; Data normalization: Normalization methods are used to convert data of different dimensions into dimensionless relative values. Normalization methods include maximum and minimum normalization or Z-score normalization.

4. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 3, characterized in that: The specific process of load decomposition in step (3) includes: Perform principal component analysis on the preprocessed load data set and calculate the eigenvalue λi and eigenvector ei of the covariance matrix C; Extract the principal components whose cumulative variance contribution rate meets the preset conditions and form a principal component matrix; The principal component weights are dynamically adjusted based on the equipment type and historical regulation data so that the principal components after weight distribution meet the preset conditions and obtain orthogonal characteristic load components.

5. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 4, characterized in that: The wavelet transform feature extraction process in step (4) includes: Perform multi-layer discrete wavelet transform on each characteristic load component to obtain detail coefficients and approximate coefficients in different frequency bands; The wavelet energy value of each frequency band is calculated to construct a multidimensional feature vector containing the time domain fluctuation characteristics and frequency domain energy distribution.

6. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 5, characterized in that: Load change rate Reflect real-time power and historical average power Relative change of equipment adjustment margin Reflects the power adjustment range of the equipment, that is, the upper and lower limits of the equipment power adjustment and the rated power The proportional relationship between the user response level Comprehensive evaluation through users' historical behavioral characteristics of participating in demand response.

7. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 6, characterized in that: The generation of the optimized scheduling strategy in step (6) further includes: The optimization model is solved through intelligent optimization algorithms to generate the optimal scheduling plan and send it to each device through the communication network. The communication network supports hybrid wired and wireless transmission, and the data acquisition terminal has local caching and breakpoint resumption functions.

8. The method for comprehensive decomposition and identification of total load of a virtual power plant according to claim 2, characterized in that: The installation locations of the hardware monitoring equipment are as follows: the smart meter is installed in the user's meter box or the preset key equipment incoming line terminal, the mutual inductor group is installed on the low-voltage side of the transformer and the preset key position of the main feeder loop, the temperature sensor is attached to the heating part of the preset key equipment, and the environmental sensors are distributed in different environmental areas to cover the monitoring range.

Images