Virtual power plant management method based on multi-source data fusion

Abstract

The invention relates to the technical field of power plant management, and discloses a virtual power plant management method based on multi-source data fusion, which is used for solving the problem of equipment response delay among equipment during virtual power plant management. Acquiring equipment response behavior characteristic parameters of each distributed energy equipment according to the initial equipment operation sequence library, evaluating according to the equipment response behavior characteristic parameters to obtain a response lag index, and judging whether each distributed energy equipment has an asynchronous response lag effect or not according to the response lag index, all response lag devices are screened out, delay compensation is carried out on all the response lag devices, an actual device operation sequence library is obtained, a fusion input matrix is obtained according to the actual device operation sequence library, main feature dimensions are extracted based on the fusion input matrix, a state feature vector sequence is constructed, and a time sequence prediction model is obtained through training. And the accuracy of virtual power plant management is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power plant management, more particularly to a virtual power plant management method based on multi-source data fusion. BACKGROUND

[0002] As an important form of aggregation and coordinated management of distributed energy systems, virtual power plants have become a key support means in new power systems and integrated energy dispatching systems. They are widely used in photovoltaic power generation, wind power generation, energy storage regulation, electric vehicle charging, controllable load management and other scenarios, involving multiple engineering technology fields such as energy management, power electronics, dispatching optimization, edge computing and communication protocols.

[0003] In the construction and operation of virtual power plants, in order to achieve the whole process management from distributed resource registration, operation data collection, situation assessment, optimization dispatching to control feedback closed loop, it is necessary to rely on multi-source data fusion mechanism to build a unified management platform for heterogeneous devices, heterogeneous protocols and asynchronous communication, and to collaboratively complete tasks such as multi-device state monitoring, dispatchable capacity assessment, fine control strategy generation and execution verification. At the same time, with the in-depth development of concepts such as "multi-energy collaboration", "source-load interaction" and "cloud-edge collaboration", the virtual power plant platform gradually evolves from a "static aggregation platform" to a "dynamic prediction and response platform".

[0004] In recent years, data collection protocols for multiple types of distributed resources, feature standardization methods and dispatching strategy generation engines based on machine learning have developed significantly. In the prior art, the future system state is usually predicted by establishing a unified data access structure, constructing a device operation state vector and using a time series prediction model, and a dispatching instruction set is constructed accordingly. The dispatching platform usually relies on a fused input matrix for unified modeling, combining the data of multiple devices at the same time point as model input to improve the synergy of prediction and control.

[0005] However, the above-mentioned technology at least has the following technical problems:

[0006] In a real operating environment, there are natural differences in the time of distributed devices responding to control instructions, for example, energy storage systems can usually execute quickly after receiving charge and discharge instructions, while the response of the load side may have behavioral inertia or control lag. If the causal response time delay between devices is not considered, but all state data is directly aligned at a unified time step and used for modeling, it may cause the logical association between feature data to be misplaced. Such implicit lag cannot be discovered through conventional data integrity checks, and if it is directly used to construct a fused input matrix, it will destroy the time sequence consistency of the model, mislead state assessment and dispatching path generation, and thus affect the prediction accuracy and control strategy effectiveness. SUMMARY

[0007] In order to overcome the above-mentioned defects of the prior art, the present application provides a virtual power plant management method based on multi-source data fusion to solve the problems existing in the background art.

[0008] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0009] A virtual power plant management method based on multi-source data fusion, comprising the following steps: step 1: obtaining the identification code set of each distributed energy equipment accessed by the virtual power plant, and obtaining the equipment information of each distributed energy equipment, and establishing a unified data access structure according to the equipment information; step 2: presetting a time step, collecting historical equipment operation parameters in a set time period according to the time step based on the data access structure, the equipment operation parameters including photovoltaic output power, energy storage charge state and load response power, constructing a historical structured data table according to the historical equipment operation parameters, and establishing an initial equipment operation sequence library with the identification code of the equipment as the index; step 3: obtaining the equipment response behavior characteristic parameters of each distributed energy equipment according to the initial equipment operation sequence library, the equipment response behavior characteristic parameters including control instruction trigger record, state switching record and equipment operation parameter, evaluating the response lag index according to the equipment response behavior characteristic parameters, determining whether the non-synchronous response lag effect occurs for each distributed energy equipment according to the response lag index, and screening all response lag equipment; step 4: compensating for the delay of all response lag equipment to obtain an actual equipment operation sequence library, and obtaining a fusion input matrix according to the actual equipment operation sequence library; step 5: based on the fusion input matrix, extracting the main feature dimension, constructing a state feature vector sequence, and training a time series prediction model; step 6: obtaining the state feature vectors of the recent time steps, calling the time series prediction model to predict the state feature vector prediction values of the future time steps, and constructing a scheduling response interval set of the virtual power plant according to the state feature vector prediction values; step 7: generating a scheduling instruction set according to the scheduling response interval set and taking the identification code of the equipment as the index.

[0010] Preferably, the step of establishing a unified data access structure according to the equipment information comprises: extracting the running field supported by each type of equipment according to the equipment information of each distributed energy equipment, identifying the numerical type and physical unit corresponding to each running field, and constructing a field type mapping table; setting a standard sampling frequency specification for each distributed energy equipment, and uniformly specifying the sampling frequency and time granularity of each type of running field; constructing a field index structure for each type of distributed energy equipment, and associating and binding the field structure with the identification code of the equipment to form a "equipment identification code-field template" mapping table; setting a unified data exception processing rule, and specifying the processing method of missing data; encapsulating the field type mapping table, the sampling frequency specification, the field index structure and the data exception processing rule into a unified data access structure.

[0011] Preferably, the initial device operation sequence library acquisition step is: identifying the device operation parameters of the device according to the field type mapping table in the unified data access structure and the sampling frequency specification; collecting the device operation parameters at a set time step for each distributed energy device to form a time-sequentially arranged device operation parameter sequence; arranging the device operation parameter sequence of each type of device into a historical structured data table; binding the structured data table corresponding to each distributed energy device with an identification code to construct an initial operation sequence library indexed by the device identification code.

[0012] Preferably, the response lag index acquisition step is: setting a detection time period, obtaining the control instruction trigger record of each distributed energy device within the detection time period, and evaluating the control trigger frequency fluctuation coefficient according to the control instruction trigger record of each distributed energy device; obtaining the state switching record of each distributed energy device within the detection time period, and evaluating the state switching lag diffusion coefficient according to the state switching record; obtaining the device operation parameters of each distributed energy device within the detection time period, and evaluating the index change asynchronous offset coefficient according to the device operation parameters; normalizing the control trigger frequency fluctuation coefficient, the state switching lag diffusion coefficient, and the index change asynchronous offset coefficient, and calculating the response lag index according to the normalized control trigger frequency fluctuation coefficient, the state switching lag diffusion coefficient, and the index change asynchronous offset coefficient, and the specific acquisition steps are: ; wherein, is the response lag index, is the normalized control trigger frequency fluctuation coefficient, is the normalized state switching lag diffusion coefficient, is the normalized index change asynchronous offset coefficient, , , is the weight coefficient of the normalized control trigger frequency fluctuation coefficient, the weight coefficient of the normalized state switching lag diffusion coefficient, and the weight coefficient of the normalized index change asynchronous offset coefficient.

[0013] Preferably, the control trigger frequency fluctuation coefficient obtaining step is: obtaining control instruction trigger records of each distributed energy equipment in a detection period, the trigger record containing the identification code of the equipment, the instruction issue timestamp and the instruction type information; taking a preset time step as a unit, counting the number of control instructions received by each equipment in each time step to form a control trigger frequency sequence of each equipment in the detection period; calculating the average trigger frequency of the control trigger frequency sequence of each equipment in the detection period to obtain the average control trigger frequency of each equipment in the current period; calculating the standard deviation of the control trigger frequency sequence of each equipment to obtain the control trigger frequency standard deviation of each equipment in the current period; dividing the control trigger frequency standard deviation of each equipment in the current period by the average control trigger frequency to obtain the control trigger frequency fluctuation coefficient.

[0014] Preferably, the state switching lag diffusion coefficient obtaining step is: obtaining state switching records of each distributed energy equipment in a detection period, the state switching record containing the identification code of the equipment, the switching start time, the switching completion time and the state information before and after the switching; for each state switching record, calculating the difference between the switching completion time and the switching start time to obtain the lag time of a single state switching; averaging all state switching lag times of each equipment in the detection period to obtain the average state switching lag time of the equipment; obtaining the maximum and minimum values of the average lag time of all equipment in the detection period, subtracting the minimum value from the maximum value of the average lag time between equipment, and then dividing by the maximum value of the average lag time between equipment to obtain the state switching lag diffusion coefficient.

[0015] Preferably, the index change asynchronous offset coefficient obtaining step is: extracting equipment operation parameters of each distributed energy equipment from an initial equipment operation sequence library in a detection period; for each equipment operation parameter of each equipment, calculating the change amount of adjacent time steps as the operation change amount; for each equipment and each time step, calculating the average difference of the operation change amount of the equipment and other equipment in the same time step; for each equipment, constructing a difference degree sequence according to the average difference of the operation change amount of the equipment and other equipment in each time step, calculating the mean of the difference degree sequence and the standard deviation of the difference degree sequence; for each equipment, performing ratio calculation on the standard deviation of the difference degree sequence and the mean of the difference degree sequence to obtain the index change asynchronous offset coefficient.

[0016] Preferably, the step of determining whether each distributed energy source device has a non-synchronous response lag effect according to the response lag index is: comparing the response lag index of each distributed energy source device with the lag threshold value, if the response lag index is greater than or equal to the lag threshold value, it is determined that the device has a non-synchronous response lag effect, and the device is recorded as a response lag device; if the response lag index is less than the lag threshold value, it is determined that the device does not have a non-synchronous response lag effect, and all distributed energy source devices are traversed to obtain all response lag devices.

[0017] Preferably, the step of compensating for the delay of all response lag devices to obtain the actual device operation sequence library is: for each response lag device, the response lag index is calculated by the ratio of the lag threshold value to obtain the compensation factor; set the initial compensation duration, and calculate the product of the compensation factor and the initial compensation duration to obtain the actual compensation time; according to the actual compensation time, the running parameters of the response lag device in the initial device operation sequence library are time label shifted and compensated; the running parameters of the response lag device after the delay compensation processing are reorganized and indexed according to the device identification code to generate the actual device operation sequence library.

[0018] The technical effects and advantages of the present application are:

[0019] According to the initial device operation sequence library, the device response behavior characteristic parameters of each distributed energy source device are obtained, the response lag index is evaluated according to the device response behavior characteristic parameters, whether each distributed energy source device has a non-synchronous response lag effect is determined according to the response lag index, all response lag devices are screened out, the delay of all response lag devices is compensated to obtain the actual device operation sequence library, the fusion input matrix is obtained according to the actual device operation sequence library, the main feature dimension is extracted based on the fusion input matrix, the state feature vector sequence is constructed, and the time series prediction model is trained, which effectively improves the accuracy of virtual power plant management. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 A flow chart of a virtual power plant management method based on multi-source data fusion is provided for the embodiments of the present application. DETAILED DESCRIPTION

[0021] The technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application, and the forms of each structure described in the following embodiments are only examples, and the virtual power plant management method based on multi-source data fusion involved in the present application is not limited to each structure described in the following embodiments. All other embodiments obtained by those skilled in the art without making creative efforts belong to the scope of protection of the present application.

[0022] The application provides a virtual power plant management method based on multi-source data fusion, which comprises the following steps as shown in the figure: Figure 1 The application provides a virtual power plant management method based on multi-source data fusion, which comprises the following steps as shown in the figure:

[0023] Step 1: Obtain the identification code set of each distributed energy equipment accessed by the virtual power plant, and obtain the equipment information of each distributed energy equipment, and establish a unified data access structure according to the equipment information;

[0024] The unified data access structure can standardize the field, synchronize the time and manage the structure template of the running data of various distributed energy equipment accessed by the virtual power plant, thereby solving the inconsistency problems of multi-source equipment in data format, time accuracy, sampling frequency and field naming. Through the unified data access structure, the specification collection, stable storage and efficient calling of the equipment running data can be realized, and consistent structure and unified dimension basic data support is provided for subsequent time sequence construction, state feature extraction and scheduling model training, thereby effectively improving the fusion processing capability and scheduling modeling accuracy of the system.

[0025] In this embodiment, it is specifically pointed out that the step of establishing a unified data access structure according to the equipment information is:

[0026] According to the equipment information of each distributed energy equipment, the running fields supported by each type of equipment are extracted, the running fields include voltage value, current value, active power, reactive power and running state field, the numerical type and physical unit corresponding to each running field are identified, and a field type mapping table is constructed;

[0027] A standard sampling frequency specification is set for each distributed energy equipment, and the sampling frequency and time granularity of each type of running field are uniformly specified to ensure the consistency and alignment of multi-device data in the time dimension;

[0028] A field index structure is constructed for each type of distributed energy equipment, and the field structure is associated and bound with the identification code of the equipment to form a “equipment identification code-field template” mapping table, so that the subsequently collected data can be classified, indexed and called according to the equipment number;

[0029] A unified data exception processing rule is set, the data exception processing rule includes illegal value elimination, null value filling and outlier identification, and the processing mode of missing data is specified, such as forward filling or linear interpolation, thereby improving the stability and integrity of data access;

[0030] The field type mapping table, the sampling frequency specification, the field index structure and the data exception processing rule are encapsulated as a unified data access structure, and are stored in categories with the equipment type as the index, for subsequent historical running parameter collection and standardized processing calling.

[0031] Step 2: preset a time step, collect historical device operation parameters in a set time period according to the time step based on the data access structure, the device operation parameters include photovoltaic output power, energy storage state of charge and load response power, construct a historical structured data table according to the historical device operation parameters, and establish an initial device operation sequence library with the identification code of the device as the index;

[0032] It should be specifically pointed out that the time step is the minimum time unit of data collection, and the time step and the set time period can be adjusted according to actual conditions, for example, the time step can be 0.5S or 1S, and the set time period can be 1H or 2H.

[0033] In this embodiment, it should be specifically pointed out that the acquisition step of the initial device operation sequence library is:

[0034] According to the field type mapping table and the sampling frequency specification in the unified data access structure, the device operation parameters of the device are identified, including photovoltaic output power, energy storage state of charge and load response power, to ensure the consistency of the collected fields;

[0035] For each distributed energy device accessed, the device operation parameters are collected according to the set time step to form a device operation parameter sequence arranged in time sequence, ensuring that the operation parameters of each device have a corresponding relationship between time stamp and value;

[0036] The device operation parameter sequence of each type of device is sorted into a historical structured data table, which is divided by field and sorted by time to form a standard time sequence structure, and each row in the data table corresponds to a time step, and each column corresponds to an operation parameter;

[0037] The structured data table corresponding to each distributed energy device is bound with its identification code to construct an initial operation sequence library with the device identification code as the index, realizing the unified management and quick call of the time sequence structure of all distributed energy device operation parameters.

[0038] Step 3: obtain the device response behavior characteristic parameters of each distributed energy device according to the initial device operation sequence library, the device response behavior characteristic parameters include control instruction trigger record, state switching record and device operation parameter, evaluate the response lag index according to the device response behavior characteristic parameters, determine whether the non-synchronous response lag effect appears for each distributed energy device according to the response lag index, and screen out all response lag devices;

[0039] In this embodiment, it should be specifically pointed out that the response lag index acquisition step is:

[0040] A detection time period is set, and during the detection time period, control instruction trigger records of each distributed energy source device are acquired, and a control trigger frequency fluctuation coefficient is evaluated according to the control instruction trigger records of each distributed energy source device. It needs to be specifically explained that the detection time period can be changed according to actual conditions, for example, the detection time period can be one month or two months;

[0041] During the detection time period, state switching records of each distributed energy source device are acquired, and a state switching lag diffusion coefficient is evaluated according to the state switching records;

[0042] During the detection time period, device operation parameters of each distributed energy source device are acquired, and an index change asynchronous offset coefficient is evaluated according to the device operation parameters;

[0043] The control trigger frequency fluctuation coefficient, the state switching lag diffusion coefficient and the index change asynchronous offset coefficient are normalized, and a response lag index is calculated according to the normalized control trigger frequency fluctuation coefficient, the normalized state switching lag diffusion coefficient and the normalized index change asynchronous offset coefficient. The specific acquisition steps are as follows:

[0044]

[0045] In the formula, is the response lag index, is the normalized control trigger frequency fluctuation coefficient. The greater the frequency fluctuation degree of the control instruction received by each distributed energy source device in unit time, that is, the more unstable the control rhythm or the more obvious the response interval difference between devices, the higher the risk of non-synchronous response of the whole system, thereby causing the response lag index to rise, is the normalized state switching lag diffusion coefficient. When each distributed energy source device responds to the same control instruction, there is a large difference in the completion time of the state switching, that is, part of the devices respond quickly and part of the devices have significant delay, which indicates that the response lag phenomenon between devices has diffusion and inconsistency, thereby causing the response lag index of the whole system to rise, is the normalized index change asynchronous offset coefficient. When multiple distributed energy source devices have significant differences in the change amplitude of the device operation parameters in the same time step, it indicates that the response of the devices to the control instruction or external disturbance is inconsistent and the coordination is poor, which reflects that there is asynchronous offset of the running behavior in the system, thereby causing the response lag index to rise, is the weight coefficient of the normalized control trigger frequency fluctuation coefficient, the weight coefficient of the normalized state switching lag diffusion coefficient and the weight coefficient of the normalized index change asynchronous offset coefficient, ,​​​ , obtained by the analytic hierarchy process, and , , , may be 0.4, 0.2, 0.4, the analytic hierarchy process is a structured multi-criteria decision-making method for quantitative comparison of the relative importance of multiple evaluation factors. Its basic principle is: first, decompose the complex problem into multiple levels such as target layer, criterion layer and index layer, then construct a pair-wise comparison judgment matrix, guide experts to score the importance between any two indexes (such as using 1-9 scale method), form a judgment matrix; then calculate the characteristic vector of the matrix and check the consistency, finally get the weight value of each evaluation factor relative to the overall goal.

[0046] In this embodiment, it needs to be specifically pointed out that the acquisition step of controlling the trigger frequency fluctuation coefficient is:

[0047] In the detection time period, the control instruction trigger record of each distributed energy equipment is obtained, the trigger record contains the identification code of the equipment, the instruction issuing timestamp and the instruction type information, which is used for subsequent calculation of the control trigger number of each equipment;

[0048] In units of preset time steps, the number of control instructions received by each equipment in each time step is counted to form the control trigger frequency sequence of each equipment in the detection time period;

[0049] The average trigger frequency of the control trigger frequency sequence of each equipment in the detection time period is calculated to obtain the average control trigger frequency of each equipment in the current period, and the average value is used to measure the overall level of the equipment control rhythm;

[0050] The standard deviation of the control trigger frequency sequence of each equipment is calculated to obtain the control trigger frequency standard deviation of each equipment in the current period, which is used to measure the fluctuation amplitude of the equipment control trigger frequency, and the larger the standard deviation, the more unstable the trigger frequency is;

[0051] The control trigger frequency standard deviation of each equipment in the current period is divided by the average control trigger frequency to obtain the control trigger frequency fluctuation coefficient, and the larger the coefficient value, the more unstable the control instruction rhythm is, and the greater the possibility of non-synchronous response between equipment.

[0052] In this embodiment, it needs to be specifically pointed out that the acquisition step of the state switching lag diffusion coefficient is:

[0053] In the detection time period, the state switching record of each distributed energy equipment is obtained, the state switching record includes the identification code of the equipment, the switching start time, the switching completion time and the state information before and after the switching, which is used to analyze the time characteristics of the switching process;

[0054] For each state switching record, the difference between the switching completion time and the switching start time is calculated to obtain the lag time of a single state switching, which is used to represent the degree of response delay of the device to the control instruction;

[0055] For each device, the average of all state switching lag times of the device within the detection period is calculated to obtain the average state switching lag time of the device, which is used to measure the overall response level of the device in the period;

[0056] In the detection period, the maximum and minimum values of the average lag times of all devices are obtained respectively, which are used to reflect the distribution difference of the response delay among the devices;

[0057] The maximum value of the average lag times of the devices is subtracted from the minimum value, and then divided by the maximum value of the average lag times of the devices to obtain the state switching lag diffusion coefficient. The larger the value of the state switching lag diffusion coefficient, the more significant the difference in response lag among the devices, and the stronger the asynchrony.

[0058] In this embodiment, it needs to be specifically explained that the acquisition steps of the index change asynchrony offset coefficient are:

[0059] In the detection period, the device operation parameters of each distributed energy device are extracted from the initial device operation sequence library, including photovoltaic output power, energy storage state of charge, and load response power;

[0060] For each device operation parameter, the change amount of adjacent time steps is calculated according to the time step, which is denoted as the operation change amount, and the operation change amount is the current time step operation parameter value minus the last time step operation parameter value, which is used to reflect the dynamic fluctuation of the operation parameter;

[0061] For each device and each time step, the average difference of the operation change amounts of the device and other devices at the same time step is calculated, and the specific acquisition steps are as follows:

[0062] ;

[0063] In the formula, is the average difference of the operation change amounts of the device i and other devices at the time step t, N is the total number of devices, is the operation change amount of the device i at the time step t, is the operation change amount of the device j at the time step t;

[0064] For each device, a difference degree sequence is constructed according to the average difference of the operation change amounts of the device and other devices at each time step, and the mean of the difference degree sequence and the standard deviation of the difference degree sequence are calculated.

[0065] For each device, the standard deviation of the difference sequence is divided by the mean of the difference sequence to obtain the index change asynchronous offset coefficient. The larger the index change asynchronous offset coefficient, the more different the device is from the group index change, and the higher the risk of asynchronous response.

[0066] At the same time step, the running change amount of all devices is counted to calculate the standard deviation, which is used to measure the difference degree of the index change between devices at the time step. The larger the standard deviation, the stronger the asynchrony.

[0067] At the same time step, the running change amount of all devices is counted to calculate the standard deviation, which is used to measure the difference degree of the index change between devices at the time step. The larger the standard deviation, the stronger the asynchrony.

[0068] The standard deviation of the index change amount at each time step is divided by the corresponding mean to obtain the index change asynchronous offset coefficient at the time step. The larger the coefficient value, the more different the running index change between devices, and the higher the possibility of asynchronous response.

[0069] In this embodiment, it needs to be specifically explained that the steps of determining whether each distributed energy device appears non-synchronous response lag effect according to the response lag index and screening all response lag devices are as follows:

[0070] The response lag index of each distributed energy device is compared with the lag threshold value. If the response lag index is greater than or equal to the lag threshold value, it is determined that the device appears non-synchronous response lag effect, and the device is recorded as a response lag device. If the response lag index is less than the lag threshold value, it is determined that the device does not appear non-synchronous response lag effect. All distributed energy devices are traversed to obtain all response lag devices. The lag threshold value is obtained by an adaptive threshold method. The adaptive threshold method is an algorithm method for dynamically determining the determination threshold based on historical data statistical characteristics, which is suitable for automatically adjusting the threshold standard in different scenes to improve the judgment accuracy. The basic principle is: in a set data observation window, the distribution characteristic parameters of the target variable (such as the response lag index) are extracted, for example, the mean, the standard deviation, the maximum value, the quantile, etc., and then a dynamically updated threshold function is constructed combined with the change trend or the fluctuation degree. Unlike the fixed threshold method, the adaptive threshold method can adjust the lag determination threshold in real time according to the historical fluctuation range of the system running state, thereby improving the judgment sensitivity and robustness under complex running conditions.

[0071] Step 4: Delay compensation is performed on all response lag devices to obtain an actual device running sequence library. The execution time of the device running parameters in the actual device running sequence library is time-labeled and aligned, abnormal data segments are removed, and a sliding mean calculation method is used to convert the output values of each device at each time step within a set time period into an equal-dimensional structure to generate a fusion input matrix.

[0072] The moving average calculation method is a commonly used time series smoothing processing technology, which is used to eliminate short-term fluctuations and enhance data trend stability. The basic principle is: in the continuous time series, a fixed length sliding window is set, the value of each time point and the data of several time points before and after it are combined to form a local data set, and the arithmetic mean value of the data set is calculated as the smoothed value of the time point. The window slides on the entire time series to form a continuous mean sequence.

[0073] It should be specifically pointed out that the moving average calculation method is used to convert the output values of each device at each time step in the set time period into an equal dimension structure to generate a fusion input matrix. For the prior art, the specific steps are not described in detail.

[0074] In this embodiment, it should be specifically pointed out that the steps for delay compensation of all response lag devices to obtain the actual device operation sequence library are as follows:

[0075] For each response lag device, the response lag index is calculated by ratio with the lag threshold value to obtain a compensation factor, which is used to represent the severity of the current distributed energy device response lag;

[0076] Set the initial compensation duration, multiply the compensation factor by the initial compensation duration to obtain the actual compensation time, which is used to dynamically adjust the response timing of the distributed energy device;

[0077] According to the actual compensation time, the running parameters of the response lag device in the initial device operation sequence library are time-labeled and compensated to correct the position deviation of the running parameters of the response lag device in the time sequence;

[0078] The running parameters of the response lag device after delay compensation are reorganized and indexed according to the device identification code to generate the actual device operation sequence library.

[0079] By calculating the compensation factor by ratio of the response lag index and the lag threshold value, and dynamically adjusting the initial compensation time with the compensation factor, the delay compensation amplitude can be adaptively determined according to the response lag degree under different devices and different running scenarios, so as to accurately correct the time sequence deviation existing in the initial device operation sequence library. The consistency and alignment accuracy of the multi-device data in the actual device operation sequence library in the time dimension can be improved, and a more stable and reliable data basis is provided for subsequent state feature extraction, prediction modeling and scheduling optimization.

[0080] Step 5: Based on the fusion input matrix, the main characteristic dimension is extracted by using the principal component analysis method, the state feature vector sequence is constructed, and the time series prediction model is trained by taking the state feature vector sequence as the input;

[0081] Principal component analysis (PCA) is a common dimension reduction and feature extraction technique, which aims to project the original multi-dimensional data onto a set of new variables (principal components) that are uncorrelated with each other, while preserving as much variance information of the original data as possible. The basic principle is: covariance analysis is performed on the fusion input matrix, and the eigenvectors in the direction with the maximum variance are extracted as the principal components, so as to reduce the dimension while maximizing the main features in the data structure. The principal component analysis method can effectively remove redundant features, reduce the correlation between dimensions, and improve the discriminability of the state feature vector sequence and the model training efficiency.

[0082] In this embodiment, it needs to be specifically pointed out that the construction steps of the time series prediction model are:

[0083] According to the state feature vector sequence, the historical state feature vectors in a set time period are selected as the training input samples, and the state feature vectors corresponding to the future time steps are selected as the training output samples, and a training data set of the time series prediction model is constructed;

[0084] According to the dimension and time step length of the state feature vector sequence, the structure parameters of the time series prediction model are initialized, including the number of input layer nodes, the number of hidden layer units, the number of output layer nodes and the length of time step window, and the model type is determined as a neural network structure with time series prediction ability such as recurrent neural network or long short-term memory network;

[0085] The training data set is input into the time series prediction model, and the forward calculation and error back propagation algorithm is used to iteratively update the model weight parameters. The training process minimizes the error function between the predicted output and the real state feature vector, and optimizes the model weight with the best fitting effect on the current training data set;

[0086] The precision of the trained time series prediction model is verified by using the verification data set which does not participate in the training. If the prediction error is less than the set convergence threshold, the time series prediction model and its optimal weight parameters are saved, and the time series prediction model is obtained.

[0087] The forward calculation and error back propagation algorithm is a commonly used parameter optimization method in neural network model training, which includes two main processes: first, through forward calculation, the input data is sequentially transmitted to each layer of neurons and the output result is calculated; then, by calculating the error between the predicted output and the true value, the error is propagated from the output layer to the input layer by using the chain rule, and the weight and bias parameters of each layer are updated in turn to minimize the error function.

[0088] Step 6: Obtain the state feature vectors of the recent time steps, call the time series prediction model to predict the state feature vector prediction values of the future time steps, and construct the scheduling response interval set of the virtual power plant according to the state feature vector prediction values;

[0089] In this embodiment, it needs to be specifically pointed out that the obtaining step of the scheduling response interval set of the virtual power plant is:

[0090] According to the state feature vector prediction values, the key operating parameters of each distributed energy equipment in each prediction time step are extracted, including the predicted output power of the equipment, the predicted state of charge of the energy storage system, the predicted load value of the adjustable load equipment, and the rated output power upper limit and the minimum operating threshold of various equipment, which are used as the basic input parameters required for scheduling capability evaluation;

[0091] According to the key operating parameters in each prediction time step, the maximum up-regulatable power in the current time step is calculated, wherein: the up-regulatable power of the energy storage system is the remaining charging capacity divided by the allowed charging rate per unit time, the up-regulatable power of the adjustable load is the predicted load value minus the minimum load threshold of the equipment, and the up-regulatable power of the distributed power generation equipment is the difference between the rated power and the predicted output power of the equipment. The up-regulatable powers of the above three types of equipment are summed to obtain the maximum up-regulatable power of the time step;

[0092] Based on the extracted equipment operating parameters in the same time step, the maximum down-regulatable power of the energy storage system, the adjustable load equipment and the distributed power generation equipment is calculated, wherein: the down-regulatable power of the energy storage system is the maximum discharge capacity allowed by the current state of charge, the down-regulatable power of the adjustable load is the difference between the rated load upper limit and the current predicted load of the equipment, and the down-regulatable power of the power generation equipment is the predicted output power minus the minimum operating power value. The down-regulatable powers of the above types of equipment are summed to obtain the maximum down-regulatable power of the time step;

[0093] The maximum down-regulatable power and the maximum up-regulatable power of each prediction time step are taken as the lower limit and the upper limit of power regulation to construct the scheduling response interval of the virtual power plant in the current time step. The scheduling response intervals corresponding to all prediction time steps are arranged in chronological order to form a scheduling response interval set, which is used to describe the scheduling capability boundary of the virtual power plant in the future time periods and is used as a constraint condition input for subsequent scheduling path optimization.

[0094] Step 7: According to the state feature vector prediction values and the scheduling response interval of the future time steps, a target scheduling path is generated by using a dynamic programming method to obtain a scheduling target value set, and a scheduling instruction set is generated with the identification code of the equipment as the index, which includes the target power value, the execution time point, the equipment number and the priority identifier.

[0095] The dynamic programming method is an algorithmic idea for solving multi-stage optimal decision problems. By decomposing the original problem into interrelated sub-problems, the optimal solution of the sub-problems is gradually solved and the results are reused to construct the optimal solution path of the overall problem. Its core principle is "optimal substructure" and "state transition", that is, the optimal decision of each stage depends on the state of the previous stage and its optimal result. When applied to virtual power plant dispatching, the dynamic programming method can consider the state feature vector prediction value and the dispatch response interval in the future multiple time steps, find the path with the lowest overall dispatch cost or the maximum benefit under the conditions of meeting the power limit and equipment capacity constraints, and generate a set of target dispatch values in a continuous time period to realize multi-period joint optimization.

[0096] It should be specifically noted that the dynamic programming method is used to generate the target dispatch path according to the state feature vector prediction value and the dispatch response interval of the future time step, which is prior art, and the specific steps are not described in detail.

[0097] Finally, the above is only the preferred embodiment of the present application and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application should be included in the protection scope of the present application.

[0098] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A virtual power plant management method based on multi-source data fusion, characterized in that, Includes the following steps: Step 1: Obtain the set of identification codes for each distributed energy device connected to the virtual power plant, obtain the device information for each distributed energy device, and establish a unified data access structure based on the device information; Step 2: Preset time step. Based on the data access structure, collect historical equipment operating parameters within the set time period according to the time step. The equipment operating parameters include photovoltaic output power, energy storage charge status and load response power. Construct a historical structured data table based on the historical equipment operating parameters, and establish an initial equipment operating sequence library with the equipment identification code as the index. Step 3: Obtain the device response behavior characteristic parameters of each distributed energy device according to the initial device operation sequence library. The device response behavior characteristic parameters include control command trigger records, state switching records and device operation parameters. Calculate the response lag index based on the device response behavior characteristic parameters. Determine whether each distributed energy device has an asynchronous response lag effect based on the response lag index and screen out all devices with response lag. Step 4: Perform delay compensation on all devices with delayed response to obtain the actual device operation sequence library, and obtain the fusion input matrix based on the actual device operation sequence library; Step 5: Based on the fused input matrix, extract the main feature dimensions, construct the state feature vector sequence, and train the time series prediction model; Step 6: Obtain the state feature vectors of the most recent time steps, call the time series prediction model to predict the predicted values ​​of the state feature vectors of the future time steps, and construct the set of scheduling response intervals of the virtual power plant based on the predicted values ​​of the state feature vectors. Step 7: Generate a scheduling instruction set based on the set of scheduling response intervals and using the device identifier as an index.

2. The virtual power plant management method based on multi-source data fusion according to claim 1, characterized in that: The steps for establishing a unified data access structure based on device information are as follows: Based on the equipment information of each distributed energy device, extract the operating fields supported by each type of device, identify the numerical type and physical unit corresponding to each operating field, and construct a field type mapping table; Standard sampling frequency specifications are set for each distributed energy device, and the sampling frequency and time granularity of various operating fields are uniformly defined; For each type of distributed energy device, a field index structure is constructed, and the field structure is associated and bound with the device's identification code to form a "device identification code - field template" mapping table; Establish unified rules for handling data anomalies and specify how to handle missing data; The field type mapping table, sampling frequency specification, field index structure, and data anomaly handling rules are encapsulated into a unified data access structure.

3. The virtual power plant management method based on multi-source data fusion according to claim 1, characterized in that, The steps for obtaining the initial device runtime sequence library are as follows: Based on the field type mapping table and sampling frequency specification in the unified data access structure, identify the device's operating parameters; For each distributed energy device, the device operating parameters are collected at a set time step to form a sequence of device operating parameters arranged in chronological order; Organize the equipment operating parameter sequences for each type of equipment into a historical structured data table; The structured data table corresponding to each distributed energy device is bound to the identification code to build an initial operation sequence library indexed by the device identification code.

4. The virtual power plant management method based on multi-source data fusion according to claim 1, characterized in that, The steps for obtaining the response hysteresis index are as follows: Set a detection period, and within the detection period, acquire the control command trigger records of each distributed energy device. Evaluate the control trigger frequency fluctuation coefficient based on the control command trigger records of each distributed energy device. During the detection period, the state switching records of each distributed energy device are acquired, and the state switching hysteresis diffusion coefficient is evaluated based on the state switching records. During the detection period, the equipment operating parameters of each distributed energy device are acquired, and the asynchronous offset coefficient of the index change is obtained based on the equipment operating parameters. The control trigger frequency fluctuation coefficient, state transition hysteresis diffusion coefficient, and index change asynchronous offset coefficient are normalized. The response hysteresis index is then calculated based on these normalized coefficients. The specific steps for obtaining the index are as follows: ； In the formula, This is expressed as a response lag index. This is expressed as the normalized control trigger frequency fluctuation coefficient. This is expressed as the normalized state transition hysteresis diffusion coefficient. This represents the asynchronous offset coefficient of the index change after normalization. , , These are represented as the weighting coefficients of the control trigger frequency fluctuation coefficient, the state switching hysteresis diffusion coefficient, and the index change asynchronous offset coefficient after normalization.

5. A virtual power plant management method based on multi-source data fusion according to claim 4, characterized in that: The steps for obtaining the control trigger frequency fluctuation coefficient are as follows: During the detection period, acquire the control command trigger records of each distributed energy device. The trigger records include the device identification code, command issuance timestamp, and command type information. Using a preset time step as the unit, the number of control commands received by each device within each time step is counted to form a control trigger frequency sequence of each device within the detection time period; Calculate the average trigger frequency for each device within the detection time period, and obtain the average control trigger frequency for each device in the current time period; The standard deviation of the control trigger frequency sequence for each device is calculated to obtain the standard deviation of the control trigger frequency for each device in the current time period; The control trigger frequency fluctuation coefficient is obtained by dividing the standard deviation of the control trigger frequency of each device in the current time period by the average control trigger frequency.

6. The virtual power plant management method based on multi-source data fusion according to claim 4, characterized in that: The steps for obtaining the state switching hysteresis diffusion coefficient are as follows: During the detection period, the status switching records of each distributed energy device are obtained. The status switching records include the device's identification code, switching start time, switching completion time, and status information before and after the switching. For each state transition record, calculate the difference between the transition completion time and the transition start time to obtain the lag time of a single state transition; The average state switching lag time of each device is obtained by averaging all state switching lag times within the detection time period. During the detection period, the maximum and minimum values ​​of the average lag time of all devices are obtained. The maximum value of the average lag time between devices is subtracted from the minimum value, and then divided by the maximum value of the average lag time between devices to obtain the state switching lag diffusion coefficient.

7. A virtual power plant management method based on multi-source data fusion according to claim 4, characterized in that: The steps for obtaining the asynchronous offset coefficient of the indicator change are as follows: During the detection period, the equipment operation parameters of each distributed energy device are extracted from the initial equipment operation sequence library; For the equipment operating parameters of each device, calculate the change in adjacent time steps according to the time step, and record it as the operating change. For each device and each time step, calculate the average difference in the amount of change in operation between the device and other devices at the same time step; For each device, a difference series is constructed based on the average difference between its operation and that of other devices at each time step, and the mean and standard deviation of the difference series are calculated. For each device, the ratio of the standard deviation of the difference series to the mean of the difference series is calculated to obtain the asynchronous offset coefficient of the index change.

8. The virtual power plant management method based on multi-source data fusion according to claim 1, characterized in that: The step of determining whether each distributed energy device exhibits asynchronous response lag effect based on the response lag index, and screening out all response lag devices, is as follows: The response lag index of each distributed energy device is compared with the lag threshold. If the response lag index is greater than or equal to the lag threshold, the device is determined to have an asynchronous response lag effect and is recorded as a response lag device. If the response lag index is less than the lag threshold, it is determined that the device does not exhibit asynchronous response lag effect. All distributed energy devices are then traversed to obtain all devices with response lag.

9. A virtual power plant management method based on multi-source data fusion according to claim 8, characterized in that: The step of performing delay compensation on all devices with delayed responses to obtain the actual device operation sequence library is as follows: For each device with a lag in response, the ratio of the lag index to the lag threshold is calculated to obtain the compensation factor. Set the initial compensation duration, multiply the compensation factor by the initial compensation duration to obtain the actual compensation time; Based on the actual compensation time, the operating parameters of the lagging devices in the initial equipment operation sequence library are time-labeled and compensated. The delayed response parameters of the equipment are reorganized after delay compensation and indexed according to the equipment identification code to generate an actual equipment operation sequence library.

Images