Distributed virtual power plant uncertainty optimization method and system based on random process

Abstract

The invention discloses a distributed virtual power plant uncertainty optimization method and system based on a random process, and relates to the technical field of energy management and information. Verifying the stationarity of the historical power generation resource data to obtain a stationarity verification result, and determining a minimum difference order according to the stationarity verification result; analyzing the stability verification result to determine an autoregression term order and a moving average term order; constructing a distributed power generation random process model of each node according to the minimum difference order, the autoregression term order and the moving average term order, and predicting future power generation data based on the distributed power generation random process model; constructing a load prediction data set, training the LSTM neural network based on the load prediction data set to obtain a load prediction model, and predicting future load data based on the load prediction model; and optimizing a load scheduling strategy of the power grid according to the future power generation data and the future load data.

Description

Technical Field

[0001] The present application relates to the field of energy management and information technology, and more specifically, it relates to a distributed virtual power plant uncertainty optimization method and system based on stochastic processes. Background Art

[0002] A distributed virtual power plant is a coordinated management system that aggregates and collaboratively optimizes multiple distributed resources, such as distributed power sources (such as wind power, solar power and other renewable energy sources), energy storage systems, controllable loads (such as smart home appliances, electric vehicle charging stations, etc.) and electric vehicles, through advanced information technology and software systems. It does not have the form of a physical power plant, but through aggregation and communication, it "gathers sand into a tower" of various distributed resources to form a virtual power plant for operation. It can respond quickly according to the needs of the power system, adjust the output or load, and reduce the operating costs of the power system through optimized scheduling and resource integration, promote the consumption and utilization of renewable energy, reduce the consumption of fossil energy and the emission of pollutants, improve the overall reliability and stability of the power system, and reduce failures and accidents caused by imbalance between supply and demand.

[0003] As the intelligent steward of the new power system, the distributed virtual power plant plays an important role in resource integration and optimization, energy scheduling and balance, energy conservation, emission reduction and environmental protection. However, in the actual operation of the distributed virtual power plant, due to the instability and prediction error of distributed power generation, as well as the variability of local load between the distribution network and the virtual power plant, the stability of the distribution network and the virtual power plant during coordinated operation is low. Summary of the invention

[0004] The present application provides a distributed virtual power plant uncertainty optimization method and system based on random processes to solve the problems of instability and prediction errors of distributed power generation, as well as the variability of local loads between distribution networks and virtual power plants.

[0005] In a first aspect of the present application, a distributed virtual power plant uncertainty optimization method based on a random process is provided, the method comprising:

[0006] Obtain historical power generation resource data for each node in the distributed virtual power plant;

[0007] Verify the stability of the historical power generation resource data corresponding to each node, obtain the stability verification result of each node, and determine the minimum difference order of each node according to the stability verification result;

[0008] Analyze the stationarity check results of each node to determine the order of the autoregressive term and the moving average term of each node;

[0009] Construct a distributed generation stochastic process model for each node based on the minimum difference order, autoregressive order, and moving average order of each node, and predict the future power generation data of each node based on the distributed generation stochastic process model;

[0010] Construct a load prediction data set, train the LSTM neural network based on the load prediction data set to obtain a load prediction model, and predict the future load data based on the load prediction model; among them, the load prediction data set includes the historical load resource data, historical load resource characteristic data, and historical load resource influencing factor data of each node;

[0011] Optimize the load dispatching strategy of the power grid according to the future power generation data and future load data.

[0012] In one implementation, the ADF test is used to verify the stationarity of the historical power generation resource data corresponding to each node, and the stationarity verification results of each node are obtained.

[0013] In one implementation, the stationarity verification results of each node include the test statistic t, critical value h, and probability value P of the historical power generation resource data;

[0014] Determine the minimum difference order of each node according to the stationarity verification results, specifically: if t < h and P < 0.05, then judge that the historical power generation resource data is stable historical power generation resource data, if t > h or P > 0.05, then judge that the historical power generation resource data is unstable historical power generation resource data;

[0015] When the historical power generation resource data of a node is stable historical power generation resource data, the minimum difference order of the node is 0;

[0016] When the historical power generation resource data of a node is unstable historical power generation resource data, perform differential processing on the historical power generation resource data from low order to high order in sequence to obtain a sequence after differential of the corresponding order, and use the ADF test method to verify the stationarity of the sequence. If the verification result is stable, use the order of the current sequence as the minimum difference order.

[0017] In one implementation, the autocorrelation function is used to analyze the stationarity verification results of each node to obtain the autoregressive order of each node;

[0018] Use the partial autocorrelation function to analyze the stationarity verification results of each node to obtain the moving average order of each node.

[0019] In one implementation, the expression of the autocorrelation function is:

[0020] Among them, k is the lag order, n is the length of the time series, xd t is the current value of the time series, xd t-k is the past value of the time series, rho k is the autocorrelation coefficient, t is the independent variable, and barxd is the mean of the time series;

[0021] is the sum of the products of the current value and the past value of the time series, is the sum of squares of deviations of the time series;

[0022] The expression of the partial autocorrelation function is: in, is the partial autocorrelation coefficient, is the partial autocorrelation coefficient of the k-1 order model, λ k is the reflection coefficient of the current order k.

[0023] In one implementation, a distributed generation random process model of each node is constructed according to the minimum difference order, autoregressive term order and moving average term order of each node, specifically:

[0024] The autocorrelation coefficient of each node is calculated according to the autocorrelation function, and the partial autocorrelation coefficient of each node is calculated according to the partial autocorrelation function;

[0025] Draw the ACF graph based on the autocorrelation coefficient, and draw the PACF graph based on the partial autocorrelation function;

[0026] Determine the initial autoregressive term order and initial moving average term order based on the tailing and truncation characteristics of the ACF and PACF graphs;

[0027] The initial distributed generation random process model is constructed according to the minimum difference order, the initial autoregressive term order and the initial moving average term order;

[0028] The least square method is used to estimate the model parameters of the distributed generation random process model to obtain the estimation results of the model parameters;

[0029] The initial autoregressive term order and the initial moving average term order are adjusted according to the estimation results, and when the model parameters of the distributed generation random process model reach the optimal value, the optimal autoregressive term order and the moving average term order are obtained;

[0030] Based on the minimum difference order, the optimal autoregressive term order and the moving average term order, a distributed generation random process model is constructed.

[0031] In one implementation, the load dispatching strategy of the power grid is optimized according to the future power generation data and the future load data, including:

[0032] Construct multiple typical power generation scenarios based on future power generation data and future load data;

[0033] The Monte Carlo simulation method is used to simulate the operating conditions of distributed virtual power plants in each typical power generation scenario, and the load dispatching strategy of the power grid is optimized based on the operating conditions.

[0034] In one implementation, a load forecasting data set is constructed, specifically:

[0035] Obtain historical load resource data of each node in the distributed virtual power plant;

[0036] According to the periodic change characteristics of historical load resource data, the cycle length is selected to obtain the seasonal cycle length, all historical load resource data under the seasonal cycle length are obtained, and the change characteristics of all historical load resource data are captured by using the STL decomposition method to obtain the historical load resource characteristic data;

[0037] Structural processing is performed on the factor data influencing load changes in the historical load resource data to obtain the historical load resource influencing factor data;

[0038] The historical load resource data, historical load resource characteristic data and historical load resource influencing factor data are combined and divided to obtain the load forecast data set.

[0039] In a second aspect of the present application, a distributed virtual power plant uncertainty optimization system based on a random process is provided, the system comprising:

[0040] A data acquisition module is used to obtain historical power generation resource data of each node in the distributed virtual power plant;

[0041] The data verification module is used to verify the stability of the historical power generation resource data corresponding to each node, obtain the stability verification result of each node, and determine the minimum difference order of each node according to the stability verification result;

[0042] An analysis module is used to analyze the stationarity check results of each node to determine the order of the autoregressive term and the order of the moving average term of each node;

[0043] The power generation prediction module is used to construct a distributed power generation random process model of each node according to the minimum difference order, autoregressive term order and moving average term order of each node, and predict the future power generation data of each node based on the distributed power generation random process model;

[0044] The load forecasting module is used to construct a load forecasting data set, train the LSTM neural network based on the load forecasting data set, obtain a load forecasting model, and predict future load data based on the load forecasting model; wherein the load forecasting data set includes historical load resource data, historical load resource characteristic data, and historical load resource influencing factor data of each node;

[0045] The dispatch optimization module is used to optimize the load dispatch strategy of the power grid based on future power generation data and future load data.

[0046] In one implementation, an ADF test is used to verify the stability of the historical power generation resource data corresponding to each node, and a stability verification result of each node is obtained.

[0047] Compared with the prior art, this application has the following beneficial effects:

[0048] The present invention aims at the problem of instability and prediction error of distributed power generation. First, a distributed power generation random process model is constructed to predict future power generation data. Then, for the problem of variability of local load between distribution network and virtual power plant, a load prediction data set including historical load resource data, historical load resource characteristic data and historical load resource influencing factor data is constructed. Then, the load prediction data set is used to train the LSTM neural network to construct a load prediction model. The load prediction model combines historical load resource data and influencing factors (such as weather, holidays, etc.) to predict future loads, analyze the influence degree and sensitivity of different influencing factors on load changes, and thus predict future load data. Then, according to future load data and future power generation data, multiple possible power generation scenarios are generated, multiple possible power generation scenarios are simulated and emulated, and the influence and risk of distributed power generation systems on power grids under different scenarios are evaluated, so as to formulate the load scheduling strategy of power grid, and send the scheduling instructions of load scheduling strategy to each distributed node, so as to realize mutual cooperation and optimized scheduling between nodes in distributed virtual power plant, thereby improving the stability of power generation and power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The drawings described herein are used to provide a further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application. In the drawings:

[0050] Figure 1 A schematic diagram of a flow chart of a distributed virtual power plant uncertainty optimization method based on a random process provided in an embodiment of the present application;

[0051] Figure 2 A principle block diagram of a distributed virtual power plant uncertainty optimization system based on a random process provided in an embodiment of the present application. Detailed implementation manners

[0052] To make the objectives, technical solutions, and advantages of the present application clearer, the present application will be further described in detail below with reference to embodiments and the accompanying drawings. The illustrative embodiments of the present application and their descriptions are only used to explain the present application and do not limit the present application.

[0053] Please refer to Figure 1 , Figure 1 which is a schematic flowchart of a method for optimizing the uncertainty of a distributed virtual power plant based on a stochastic process provided by an embodiment of the present application. As Figure 1 shown, the method includes:

[0054] S101, obtaining historical power generation resource data of each node in the distributed virtual power plant.

[0055] In this embodiment, the virtual power plant is a power coordination management system that realizes the aggregation and coordinated optimization of distributed energy resources (DER) such as distributed power sources (such as solar photovoltaic and wind power generation), energy storage systems, controllable loads, and electric vehicles through advanced information and communication technologies and software systems. It is not a physical power station but an intelligent power grid management platform. The distributed virtual power plant is the distributed characteristics and coordinated optimization of distributed resources.

[0056] Therefore, obtaining historical power generation resource data of each node in the distributed virtual power plant is a conventional technical means well known to those skilled in the art, and no redundant description will be given in this embodiment.

[0057] It should be noted that the historical power generation resource data includes historical power generation amount data, historical light intensity data, historical wind speed data, historical system operation state data, etc. Secondly, the obtained historical power generation resource data can also be subjected to data cleaning processing to remove outliers and missing values in the data.

[0058] S102, verifying the stationarity of the historical power generation resource data corresponding to each node, obtaining the stationarity verification result of each node, and determining the minimum difference order of each node according to the stationarity verification result.

[0059] In this embodiment, the ADF test is used to perform a stability test on the historical power generation resource data to obtain the test statistic t, critical value h, and probability value P of the historical power generation resource data.

[0060] If t < h and P < 0.05, it is determined that the historical power generation resource data is stable historical power generation resource data; if t > h or P > 0.05, it is determined that the historical power generation resource data is non-stable historical power generation resource data;

[0061] When the historical power generation resource data of the node is stable historical power generation resource data, the minimum difference order of the node is 0;

[0062] When the historical power generation resource data of a node is non-stable historical power generation resource data, the historical power generation resource data is differentially processed from low order to high order in turn to obtain a sequence after the corresponding order difference. The ADF test is used to check the stationarity of the sequence. If the check structure is stable, the order of the current sequence is used as the minimum difference order.

[0063] Specifically, the first-order difference algorithm is used to calculate the data difference of the unstable historical power generation resource data, and the sequence after the first-order difference is tested for stationarity using the ADF test. The first-order difference algorithm is shown in the following formula: Δx t =x t -x t-1 ; where x t is the value of the first-order original time series at time point t, Δx t is the value after first-order difference.

[0064] If the sequence after the first-order difference is still non-stationary, the second-order difference is performed. The calculation formula of the second-order difference is as follows: Δ 2 x2 t =x2 t -x2 t-1 ; Among them, x2 t is the value of the second-order original time series at time point t, Δ 2 x2 t is the value after second-order difference.

[0065] Specifically, if the sequence is still non-stationary after the second-order difference, you can continue to try higher-order differences until the sequence becomes stationary, and select the minimum difference order that makes the sequence statistically significantly stationary as the d parameter in the ARIMA model. For example, if the sequence is stationary after the third-order difference, the minimum difference order d parameter that makes the sequence stationary is determined to be 3.

[0066] The ADF test command can be used in Python to load the ADF test function using from statsmodels.tsa.stattools.simport adfuller, and then use adfuller (time series data) to perform the test. In Stata, you can perform the test by entering the dfuller variable name.

[0067] S103, analyzing the stationarity check results of each node to determine the autoregressive term order and the moving average term order of each node.

[0068] Specifically, the autocorrelation function and partial autocorrelation function data analysis are performed on the stationarity test results of each node to obtain the autoregressive term order p parameter and the moving average term order q parameter of each node respectively.

[0069] The expression of the autocorrelation function is:

[0070] Among them, k is the lag order, n is the length of the time series, xd t is the current value of the time series, xd t-k is the past value of the time series, rho k is the autocorrelation coefficient, t is the independent variable, and barxd is the mean of the time series;

[0071] is the sum of the products of the current value and the past value of the time series, is the sum of squares of deviations of the time series;

[0072] The expression of the partial autocorrelation function is: in, is the partial autocorrelation coefficient, is the partial autocorrelation coefficient of the k-1 order model, λ k is the reflection coefficient of the current order k.

[0073] S104, constructing a distributed power generation random process model for each node according to the minimum difference order, autoregressive term order and moving average term order of each node, and predicting future power generation data of each node based on the distributed power generation random process model.

[0074] In this embodiment, it should be noted that the distributed generation random process model is an ARIMA model, and the ARIMA model is more reliable and applicable to data with obvious volatility and time correlation.

[0075] According to the minimum difference order, autoregressive term order and moving average term order of each node, the distributed generation random process model of each node is constructed, specifically: the autocorrelation coefficient of each node is calculated according to the autocorrelation function, and the partial autocorrelation coefficient of each node is calculated according to the partial autocorrelation function; the ACF diagram is drawn according to the autocorrelation coefficient, and the PACF diagram is drawn according to the partial autocorrelation function; the initial autoregressive term order and the initial moving average term order are determined according to the tailing and truncation characteristics of the ACF diagram and the PACF diagram; the initial distributed generation random process model is constructed according to the minimum difference order, the initial autoregressive term order and the initial moving average term order; the model parameters of the distributed generation random process model are estimated by the least squares method to obtain the estimation results of the model parameters; the initial autoregressive term order and the initial moving average term order are adjusted according to the estimation results, and when the model parameters of the distributed generation random process model reach the optimal level, the optimal autoregressive term order and the moving average term order are obtained; based on the minimum difference order, the optimal autoregressive term order and the moving average term order, the distributed generation random process model is constructed.

[0076] Among them, the ACF diagram and PACF diagram are both drawn with the lag order k as the horizontal axis.

[0077] Specifically, calculate the partial autocorrelation coefficient of each node and set the maximum lag order p max ,initialization is the first value of ACF (autocorrelation function), that is initialization Subtract the variance of the time series from Multiply by the variance of the time series, as shown below Calculate using Levinson recursion formula Here, γ0 is the variance of the time series, and γ1 is the first value of the ACF multiplied by the variance.

[0078] The reflection coefficient λ of the current order k is calculated according to the following formula k : Among them, λ k is the reflection coefficient of the current order k, p k For the time series xd t and its k-th lag xd t-k The autocorrelation coefficient between (k = 2, 3...p), φ k-1，j is the partial autocorrelation coefficient of the k-1 order model, p k-j is the corresponding autocorrelation coefficient, ∈ k-1 is the residual variance of the k-1 order model.

[0079] Update the partial autocorrelation coefficient according to the following formula: (j=1,2,...k-1 and ), where φ k，j is the partial autocorrelation coefficient, φ k-1，j is the partial autocorrelation coefficient of the k-1 order model, λ k is the reflection coefficient of the current order k.

[0080] Update the residual variance according to the following formula: in, is the one-step-forward prediction error variance of the k-order AR model, λ k is the reflection coefficient at the current order k, is the one-step-forward prediction error variance of the k-1 order AR model.

[0081] The partial autocorrelation coefficient of each node is calculated As the ordinate, the lag order k as the abscissa, and the graph is the PACF diagram.

[0082] Secondly, according to the tailing and truncation characteristics of the ACF diagram and the PACF diagram, the autoregressive term order p parameter and the moving average term order q parameter are preliminarily determined. The ARIMA distributed power generation random process model corresponding to each node is constructed using the preliminarily determined p parameter and q parameter. The least squares method is used to estimate the various model parameters of the distributed power generation random process model, and the significance of the model parameter estimation is checked. The p parameter and q parameter are adjusted to find the optimal model parameters, and the optimal autoregressive term order p parameter and moving average term order q parameter can be obtained.

[0083] After constructing the ARIMA distributed generation random process model for each node, the residual autocorrelation diagram is used to check whether the ARIMA distributed generation random process model is well fitted, and the mean square error MSE is used to detect the accuracy of the prediction results. The MSE calculation formula is shown in the following formula: Where n is the number of observations, y t is the actual observation value at the tth time point, is the predicted value of the model at time point t.

[0084] S105, construct a load forecasting data set, train the LSTM neural network based on the load forecasting data set to obtain a load forecasting model, and predict future load data based on the load forecasting model; wherein the load forecasting data set includes historical load resource data, historical load resource characteristic data, and historical load resource influencing factor data of each node.

[0085] In this embodiment, a load forecasting data set is constructed, specifically by: obtaining historical load resource data of each node of the distributed virtual power plant; selecting a cycle length according to the periodic change characteristics of the historical load resource data to obtain a seasonal cycle length, obtaining all historical load resource data under the seasonal cycle length, and using the STL decomposition method to capture the change characteristics of all historical load resource data to obtain historical load resource characteristic data; performing structured processing on the factor data affecting load changes in the historical load resource data to obtain historical load resource influencing factor data; combining and dividing the historical load resource data, the historical load resource characteristic data and the historical load resource influencing factor data to obtain a load forecasting data set.

[0086] The historical load resource data of each node includes historical load demand data, load classification data, meteorological data, economic data, etc. The seasonal cycle length is 4 quarters. The historical load demand data is formatted into a form suitable for STL decomposition, such as pandas's DataFrame or Series, and the time index is correctly set. All historical load demand data are decomposed using the STL decomposition method. The specific steps are as follows:

[0087] Parameter settings include inputting historical load demand data, setting the cycle parameter to 4, setting the length of the seasonal smoother to 7, and setting the length of the trend smoother to 7.

[0088] Decompose the set parameters to extract the long-term load trend component, seasonal load change pattern component, and load residual component.

[0089] Use libraries such as matplotlib or seaborn to visualize the long-term load trend component, seasonal load change pattern component, and load residual component, observe the changing characteristics of the load data, obtain trend characteristic components, seasonal characteristic components, and residual random characteristic components, and thus determine the historical load resource characteristic data.

[0090] Specifically, the statsmodels library in Python can be used to perform the STL decomposition method.

[0091] In this embodiment, other factors affecting load changes in the historical load resource data are partially structured to obtain historical load resource influencing factor data in order to eliminate some of its uncertainties. The implementation method is as follows:

[0092] Collect data on other external factors that affect load changes at each node of the virtual power plant power system, including meteorological data (such as temperature, humidity, wind speed, precipitation, etc.) and economic data (such as regional per capita GDP, gross product growth rate, holiday information, etc.), and pre-process the data on other external factors that affect load changes, including identifying variables that have a significant impact on load changes, identifying and processing outliers through IQR statistical analysis methods, interpolating missing values, and performing normalization to obtain normalized load change external factor data. Perform feature extraction on the normalized load change external factor data, use the Pearson correlation coefficient to evaluate the linear correlation between features and between features and load targets, use the Spearman rank correlation coefficient to evaluate the correlation between features and between features and load targets of non-normal distributions, perform feature selection, and use polynomial combinations to construct features for similar variables and correlated variables to capture nonlinear relationships. Combine the normalized feature data with the newly constructed feature data to obtain historical load resource influencing factor data.

[0093] The historical load resource data, historical load resource characteristic data and historical load resource influencing factor data are combined and divided to construct a load forecasting data set, and the load forecasting data set is specifically divided into a load forecasting data training set, a load forecasting data validation set and a load forecasting data test set. Based on the load forecasting data set, an LSTM load forecasting model is constructed, the number of LSTM neural network layers and the number of neurons in each layer are determined, and training parameters such as learning rate parameters, batch size parameters, and training round parameters are set. The LSTM load forecasting model is trained using the future load data training set, and the learning rate parameters, batch size parameters, and training round parameters are adjusted using the validation set data. After iterative optimization, the load forecasting model is finally obtained.

[0094] The test set of load forecast data is input into the load forecasting model to make a comprehensive forecast of future load changes and obtain future load data, which includes future load demand data, future load classification data, future meteorological data, future economic data, etc.

[0095] S106, optimizing the load dispatching strategy of the power grid according to the future power generation data and the future load data.

[0096] In this embodiment, the load dispatching strategy of the power grid is optimized according to the power generation capacity and transmission capacity of the power grid, combined with future load data and future power generation data. The optimization of the power grid dispatching strategy includes analyzing the impact of the future load resource forecast data of the virtual power plant on the load. For example, high temperature in summer may cause a surge in air conditioning load, while cold weather in winter may increase heating load. Before the peak load period is predicted, power saving reminders are sent to users through smart meters, smart home systems, etc., to encourage users to adjust their electricity consumption behavior and reduce the pressure on the power grid. The power generation plan of distributed energy (such as solar energy and wind energy) is adjusted according to the prediction results, and a closer interconnection mechanism is established between regional power grids to achieve surplus and shortage adjustment and mutual assistance according to the prediction results.

[0097] In some embodiments, based on the future power generation data and the future load data, the load dispatch strategy of the power grid is optimized, including: constructing a plurality of typical power generation scenarios based on the future power generation data and the future load data;

[0098] The Monte Carlo simulation method is used to simulate the operating conditions of distributed virtual power plants in each typical power generation scenario, and the load dispatching strategy of the power grid is optimized based on the operating conditions.

[0099] In this embodiment, typical power generation scenarios include peak scenarios, valley scenarios, scenarios with significant seasonal changes, stable scenarios, and emergency response scenarios. In actual applications, the above-mentioned multiple typical scenarios obtain the power generation resource data of the virtual power plant in real time through the power grid detection system. When the power generation resources of the virtual power plant are in the peak scenario, the power grid detection system detects that the load resources have reached a peak, and the power grid detection system starts the energy storage device discharge response mechanism. The energy storage device automatically discharges during the peak load period to supplement the power supply of the power grid and reduce the load peak. The adjustable power unit automatically adjusts the power of the adjustable power unit according to the load demand, such as adjusting the load of the industrial production line, the lighting and air-conditioning system of the commercial building, etc.; when in the valley scenario, the power grid detection system detects that the power load has reached a valley, and the power grid detection system starts the energy storage device charging response mechanism, uses the excess power generation capacity to charge the energy storage device, and stores electric energy for later use. The adjustable power unit is fine-tuned according to actual needs, such as performing equipment maintenance and system upgrades, to optimize energy utilization efficiency.

[0100] Specifically, when the virtual power plant is in a scenario with significant seasonal changes, the energy storage equipment is charged in advance before the peak demand period (such as the peak electricity consumption in summer or winter); during the low demand period (such as spring and autumn or at night), it is discharged or continued to be charged to replenish the energy storage. The load characteristics of the power unit can be adjusted according to seasonal changes, such as increasing the load management of the air-conditioning system in summer and increasing the load management of the heating system in winter.

[0101] Specifically, when the virtual power plant is in a stable scenario and the grid load is relatively stable, the energy storage equipment and the adjustable power unit mainly play a fine-tuning role. According to the slight fluctuations in the grid load, the energy storage equipment makes rapid charging and discharging adjustments to maintain the supply and demand balance of the grid. The adjustable power unit optimizes the load distribution and usage efficiency of the adjustable power unit according to the actual situation.

[0102] Specifically, when the virtual power plant is in an emergency response scenario, an emergency response mechanism is established. When a power grid failure occurs or a sudden large load demand occurs, the response procedure is immediately initiated. The energy storage equipment discharges quickly in an emergency to provide emergency power supply support for the power grid and ensure the power supply of critical loads. The adjustable power unit can quickly adjust the power output or usage mode of the adjustable power unit according to the nature and scope of impact of the emergency.

[0103] Please refer to Figure 2 , Figure 2 A block diagram of a distributed virtual power plant uncertainty optimization system based on a random process provided in an embodiment of the present application, such as Figure 2 As shown, the system includes:

[0104] The data acquisition module 210 is used to acquire the historical power generation resource data of each node in the distributed virtual power plant;

[0105] The data verification module 220 is used to verify the stability of the historical power generation resource data corresponding to each node, obtain the stability verification result of each node, and determine the minimum difference order of each node according to the stability verification result;

[0106] The analysis module 230 is used to analyze the stationarity check results of each node to determine the order of the autoregressive term and the order of the moving average term of each node;

[0107] The power generation prediction module 240 is used to construct a distributed power generation random process model of each node according to the minimum difference order, autoregressive term order and moving average term order of each node, and predict the future power generation data of each node based on the distributed power generation random process model;

[0108] The load forecasting module 250 is used to construct a load forecasting data set, train the LSTM neural network based on the load forecasting data set, obtain a load forecasting model, and predict future load data based on the load forecasting model; wherein the load forecasting data set includes historical load resource data, historical load resource characteristic data, and historical load resource influencing factor data of each node;

[0109] The dispatch optimization module 260 is used to optimize the load dispatch strategy of the power grid according to future power generation data and future load data.

[0110] It can be seen that in the system provided by this embodiment, in order to solve the problem of instability and prediction error of distributed power generation, a distributed power generation random process model is first constructed to predict future power generation data; then, in order to solve the problem of variability of local load between the distribution network and the virtual power plant, a load prediction data set including historical load resource data, historical load resource characteristic data and historical load resource influencing factor data is first constructed, and then the load prediction data set is used to train the LSTM neural network to construct a load prediction model, which combines historical load resource data and influencing factors (such as weather, holidays, etc.) to predict future loads, analyze the degree and sensitivity of different influencing factors on load changes, and thus predict future load data. Then, based on future load data and future power generation data, multiple possible power generation scenarios are generated, multiple possible power generation scenarios are simulated and emulated, and the impact and risk of distributed power generation systems on the power grid under different scenarios are evaluated, so as to formulate a load scheduling strategy for the power grid, and send the scheduling instructions of the load scheduling strategy to each distributed node, so as to achieve mutual cooperation and optimized scheduling between nodes in the distributed virtual power plant, thereby improving the stability of power generation and power supply.

[0111] In some embodiments, an ADF test is used to verify the stability of the historical power generation resource data corresponding to each node, and a stability verification result of each node is obtained.

[0112] A distributed virtual power plant uncertainty optimization system based on a random process in an embodiment of the present application is similar to the above Figure 1 The uncertainty optimization method for a distributed virtual power plant based on a random process shown is a technical solution based on the same inventive concept. Through the detailed description of the uncertainty optimization method for a distributed virtual power plant based on a random process provided in the above embodiment, those skilled in the art can clearly understand the implementation process of each module of the uncertainty optimization system for a distributed virtual power plant based on a random process in this embodiment. Therefore, for the sake of conciseness of the specification, it will not be repeated here.

[0113] The specific implementation methods described above further illustrate the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only the specific implementation method of the present application and is not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A distributed virtual power plant uncertainty optimization method based on stochastic process, characterized in that the method Including: Obtaining historical power generation resource data of each node in a distributed virtual power plant; Verifying the stationarity of the historical power generation resource data corresponding to each node, obtaining the stationarity verification results of each node, and determining the minimum difference order of each node according to the stationarity verification results; Analyzing the stationarity verification results of each node to determine the autoregressive term order and moving average term order of each node; Constructing a distributed power generation stochastic process model for each node based on the minimum difference order, autoregressive term order, and moving average term order of each node, and predicting the future power generation data of each node based on the distributed power generation stochastic process model; Constructing a load prediction data set, training an LSTM neural network based on the load prediction data set to obtain a load prediction model, and predicting future load data based on the load prediction model; wherein, the load prediction data set includes historical load resource data, historical load resource feature data, and historical load resource influencing factor data of each node; Optimizing the load dispatching strategy of the power grid according to the future power generation data and future load data.

2. The method according to claim 1, characterized in that Using the ADF test method to verify the stationarity of the historical power generation resource data corresponding to each node, and obtaining the stationarity verification results of each node.

3. The method according to claim 2, characterized in that The stationarity verification results of each node include the test statistic t, critical value h, and probability value P of the historical power generation resource data; Determining the minimum difference order of each node according to the stationarity verification results, specifically: if t < h and P < 0.05, then judge that the historical power generation resource data is stable historical power generation resource data, if t > h or P > 0.05, then judge that the historical power generation resource data is unstable historical power generation resource data; When the historical power generation resource data of a node is stable historical power generation resource data, the minimum difference order of the node is 0; When the historical power generation resource data of a node is unstable historical power generation resource data, perform differential processing on the historical power generation resource data from low order to high order in sequence to obtain a sequence after differential processing of the corresponding order, and use the ADF test method to verify the stationarity of the sequence. If the verification result is stationary, use the order of the current sequence as the minimum difference order.

4. The method according to claim 1, characterized in that: Analyzing the stationarity verification results of each node using the autocorrelation function to obtain the autoregressive term order of each node; Analyzing the stationarity verification results of each node using the partial autocorrelation function to obtain the moving average term order of each node.

5. The method according to claim 4, characterized in that The expression of the autocorrelation function is: Where k is the lag order, n is the length of the time series, and xd t is the current value of the time series, xd t-k is the past value of the time series, rho k is the autocorrelation coefficient, t is the independent variable, and barxd is the mean of the time series; is the sum of the products of the current value and the past value of the time series, is the sum of squares of deviations of the time series; The expression of the partial autocorrelation function is: in, is the partial autocorrelation coefficient, is the partial autocorrelation coefficient of the k-1 order model, λ k is the reflection coefficient of the current order k.

6. The method according to claim 4, characterized in that Constructing a distributed power generation stochastic process model for each node based on the minimum difference order, autoregressive term order, and moving average term order of each node, specifically: Calculating the autocorrelation coefficient of each node according to the autocorrelation function, and calculating the partial autocorrelation coefficient of each node according to the partial autocorrelation function; Drawing an ACF graph according to the autocorrelation coefficient, and drawing a PACF graph according to the partial autocorrelation function; Determining the initial autoregressive term order and initial moving average term order according to the trailing and truncation characteristics of the ACF graph and PACF graph; Constructing an initial distributed power generation stochastic process model based on the minimum difference order, initial autoregressive term order, and initial moving average term order; The least square method is used to estimate the model parameters of the distributed generation random process model to obtain the estimation results of the model parameters; The initial autoregressive term order and the initial moving average term order are adjusted according to the estimation results, and when the model parameters of the distributed generation random process model reach the optimal value, the optimal autoregressive term order and the moving average term order are obtained; Based on the minimum difference order, the optimal autoregressive term order and the moving average term order, a distributed generation random process model is constructed.

7. The method according to claim 1, characterized in that According to future power generation data and future load data, the load dispatch strategy of the power grid is optimized, including: Construct multiple typical power generation scenarios based on future power generation data and future load data; The Monte Carlo simulation method is used to simulate the operating conditions of distributed virtual power plants in each typical power generation scenario, and the load dispatching strategy of the power grid is optimized based on the operating conditions.

8. The method according to claim 1, characterized in that Construct a load forecasting dataset, specifically: Obtain historical load resource data of each node in the distributed virtual power plant; According to the periodic change characteristics of historical load resource data, the cycle length is selected to obtain the seasonal cycle length, all historical load resource data under the seasonal cycle length are obtained, and the change characteristics of all historical load resource data are captured by using the STL decomposition method to obtain the historical load resource characteristic data; Structural processing is performed on the factor data influencing load changes in the historical load resource data to obtain the historical load resource influencing factor data; The historical load resource data, historical load resource characteristic data and historical load resource influencing factor data are combined and divided to obtain the load forecast data set.

9. A distributed virtual power plant uncertainty optimization system based on random processes, characterized in that the system include: A data acquisition module is used to obtain historical power generation resource data of each node in the distributed virtual power plant; The data verification module is used to verify the stability of the historical power generation resource data corresponding to each node, obtain the stability verification result of each node, and determine the minimum difference order of each node according to the stability verification result; An analysis module is used to analyze the stationarity check results of each node to determine the order of the autoregressive term and the order of the moving average term of each node; The power generation prediction module is used to construct a distributed power generation random process model of each node according to the minimum difference order, autoregressive term order and moving average term order of each node, and predict the future power generation data of each node based on the distributed power generation random process model; The load forecasting module is used to construct a load forecasting data set, train the LSTM neural network based on the load forecasting data set, obtain a load forecasting model, and predict future load data based on the load forecasting model; wherein the load forecasting data set includes historical load resource data, historical load resource characteristic data, and historical load resource influencing factor data of each node; The dispatch optimization module is used to optimize the load dispatch strategy of the power grid based on future power generation data and future load data.

10. The system according to claim 9, characterized in that The ADF test method is used to verify the stability of the historical power generation resource data corresponding to each node, and the stability verification results of each node are obtained.

Images