Data-driven predictive control method and system

Abstract

The invention discloses a data-driven prediction control method and system. The method comprises the following steps: acquiring first historical data of a target power distribution system; establishing a first prediction model and a second prediction model according to the first historical data; establishing a third prediction model, wherein the input of the third prediction model is the output of the first prediction model and the second prediction model; and establishing a first predictive control strategy based on the third predictive model, and performing data-driven predictive control according to the first predictive control strategy. According to the method, the wind power output and the power distribution network response are finely controlled, so that the transient stability of the power distribution network can be judged in advance, the response speed and the regulation and control precision of the power grid are remarkably improved, emergency control measures are conveniently implemented, and high efficiency and reliability of power grid operation are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of high-proportion renewable energy distribution system technology, and in particular to a data-driven predictive control method and system. Background Art

[0002] As a renewable and clean energy, wind power generation plays a vital role in optimizing my country's energy structure and achieving the "dual carbon" goals. However, the randomness and uncontrollability of wind energy will significantly affect the output stability of wind turbines, which in turn poses a challenge to the reliable operation of the power system. In order to effectively respond to these challenges, academia and industry have proposed accurate predictions of wind speed and wind power. However, with the increase in the proportion of renewable energy, traditional predictive control methods have gradually become inadequate, posing new challenges to the control of wind power systems.

[0003] At present, the predictive control technology of wind power systems mainly includes physical model-based methods and data-driven methods. The physical model method relies on a deep understanding of wind power mechanical and meteorological conditions. Although it has theoretical rigor, it has high requirements for data and models and is difficult to respond to rapid changes in the environment in real time. In contrast, data-driven methods, especially machine learning-based technologies, can directly learn the changing patterns of wind speed and power from historical data, have better adaptability and flexibility, but may lack sufficient explanatory power and rely on a large amount of high-quality data.

[0004] To overcome these limitations, some scholars have proposed a predictive control scheme using a two-layer long short-term memory (LSTM) network. This method uses a two-layer LSTM model. The first layer focuses on learning and predicting future wind power output from historical wind speed and wind power data, and the second layer predicts grid demand and response strategies based on the output of the first layer, thereby achieving more accurate control of wind power output. Although this method improves the prediction accuracy and control response speed, the current two-layer LSTM predictive control method still has some defects: (1) data input often relies on a single data source or a limited data source, and cannot fully utilize the various data available in the distribution network; (2) relying on relatively simple mathematical models and control logic, it lacks sufficient response speed and adjustment flexibility when dealing with rapidly changing grid conditions; (3) in the face of unforeseen changes in grid load, the current prediction method cannot effectively predict and regulate, thereby affecting the stable operation of the entire grid. In general, although the predictive control technology of wind power systems based on two-layer LSTM provides a promising solution, these technologies still need further research in order to adapt to rapidly changing environments and provide stable and reliable power output. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides a data-driven predictive control method and system, which can solve the problems mentioned in the background technology.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides a data-driven predictive control method, comprising: acquiring first historical data of a target power distribution system; establishing a first prediction model and a second prediction model based on the first historical data; establishing a third prediction model, wherein the input of the third prediction model is the output of the first prediction model and the second prediction model; establishing a first predictive control strategy based on the third prediction model, and performing data-driven predictive control according to the first predictive control strategy.

[0009] As a preferred solution of the data-driven predictive control method of the present invention, wherein: the establishing of the first prediction model and the second prediction model according to the first historical data comprises:

[0010] The first prediction model is any model whose input is some data in the first historical data and whose output is wind power or related parameters of wind power that can be directly or indirectly obtained;

[0011] The second prediction model is any model whose input is some data in the first historical data and whose output is load power or related parameters of load power that can be directly or indirectly obtained.

[0012] As a preferred solution of the data-driven predictive control method of the present invention, the establishment of the third prediction model includes:

[0013] The third prediction model is an arbitrary model whose input is the output of the first prediction model and the second prediction model, and whose output is the first performance indicator of the target power distribution system;

[0014] The first performance indicator includes at least one of voltage deviation, power loss, frequency stability index or other performance indicators.

[0015] As a preferred solution of the data-driven predictive control method of the present invention, the first predictive control strategy includes:

[0016] Establishing a first objective function of a first performance indicator output by the third prediction model;

[0017] Solving the first objective function to obtain optimal control parameters;

[0018] The inverter output in the target power distribution system is adjusted according to the optimal control parameters.

[0019] As a preferred solution of the data-driven predictive control method described in the present invention, the inverter output at least includes the power and attack angle of the inverter.

[0020] As a preferred solution of the data-driven predictive control method of the present invention, wherein: the establishing of the first prediction model and the second prediction model according to the first historical data comprises:

[0021] performing a first preprocessing on the first historical data;

[0022] Performing a first classification on the first historical data after the first preprocessing to obtain first classification data and second classification data;

[0023] Establishing a first prediction model according to the first classification data;

[0024] A second prediction model is established based on the second classification data.

[0025] As a preferred solution of the data-driven predictive control method described in the present invention, the first predictive control strategy also includes: establishing a first constraint condition on the first performance indicator output by the third prediction model.

[0026] In a second aspect, the present invention provides a data-driven predictive control system, comprising:

[0027] A data acquisition module, used for acquiring first historical data of a target power distribution system;

[0028] A first model building module, used to build a first prediction model and a second prediction model according to the first historical data;

[0029] A second model building module, used to build a third prediction model, the input of the third prediction model is the output of the first prediction model and the second prediction model;

[0030] A control module is used to establish a first prediction control strategy based on the third prediction model and perform data-driven prediction control according to the first prediction control strategy.

[0031] In a third aspect, the present invention provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned method when executing the computer program.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the steps of the method described above when executed by a processor.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention proposes a data-driven predictive control method and system, obtains the first historical data of the target distribution system; establishes a first prediction model and a second prediction model according to the first historical data; establishes a third prediction model, the input of the third prediction model is the output of the first prediction model and the second prediction model; establishes a first prediction control strategy based on the third prediction model, and performs data-driven predictive control according to the first prediction control strategy. By utilizing the powerful time series data processing capability of the double-layer long short-term memory network (LSTM), it is possible to predict future wind turbine power output and load demand based on historical multivariate data, and then accurately predict the transient frequency stability index (FSI). This prediction not only supports online applications, but also provides a stability margin, buys valuable time for emergency control measures of the distribution network, and ensures the efficient and reliable operation of the distribution system. At the same time, the present invention optimizes the inverter output according to the predicted FSI by adjusting the control parameters of the model predictive control (MPC) strategy, such as the voltage reference vector, so that the system can determine the transient stability problem in advance, thereby performing advanced control and effectively improving the frequency stability of the system. The method of the present invention can be directly deployed on power electronic equipment without adding additional communication equipment and sensors to the system, overcoming the shortcomings of high construction cost and poor real-time performance in traditional methods, thereby providing an efficient, economical and easy-to-implement solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0035] Figure 1 A method flow chart of a data-driven predictive control method and system provided by one embodiment of the present invention;

[0036] Figure 2 A data-driven predictive control method control structure diagram of a data-driven predictive control method and system provided by an embodiment of the present invention;

[0037] Figure 3 A structure diagram of an ultra-short-term wind speed prediction model based on LSTM for a data-driven prediction control method and system provided in one embodiment of the present invention;

[0038] Figure 4 A data-driven predictive control method and system provided by an embodiment of the present invention provides a short-term load forecasting model structure diagram based on LSTM;

[0039] Figure 5 A virtual synchronous inverter single machine grid-connected topology diagram of a data-driven predictive control method and system provided by an embodiment of the present invention;

[0040] Figure 6 A frequency response comparison diagram of a data-driven predictive control method and system under a data-driven predictive mode provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0042] Example 1

[0043] Reference Figure 1-Figure 6 , which is the first embodiment of the present invention, provides a data-driven predictive control method and system, including:

[0044] There are some problems and deficiencies in the existing related technologies. For example, in the process of predictive control, due to the lack of in-depth analysis of historical data and accurate construction of models, the accuracy of the prediction results is not high, which cannot meet the needs of high-precision prediction in practical applications. In addition, when dealing with multi-variable, nonlinear, and time-varying complex systems, traditional methods often require complex mathematical modeling and a large amount of computing resources, which not only increases the complexity of the system, but also reduces the real-time performance of predictive control.

[0045] The present application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the data-driven predictive control method.

[0046] Figure 1 A method flow chart of a data-driven predictive control method and system is shown, including:

[0047] S101, obtaining first historical data of a target power distribution system;

[0048] In an optional embodiment, the first historical data of the target power distribution system is for establishing various subsequent models to achieve final data-driven predictive control, so the first historical data is acquired for the subsequent models;

[0049] In an optional embodiment, when it is necessary to establish a wind power model and a load power model, the first historical data may include collecting environmental data such as historical wind speed, wind direction, temperature, air pressure, humidity, etc. near multiple distributed wind turbines and wind turbine power generation data at corresponding moments and operating load power data of the distribution network in the same time segment;

[0050] In an optional embodiment, the first historical data may also be preprocessed to achieve data adaptation to the input requirements of the neural network model.

[0051] It should be noted that obtaining the first historical data of the target power distribution system can provide detailed training samples or parameter data for the establishment of subsequent models, which can ensure the accuracy of the subsequent model establishment and the effectiveness of predictive control.

[0052] S102, establishing a first prediction model and a second prediction model according to the first historical data;

[0053] In the embodiment of the present application, establishing the first prediction model and the second prediction model according to the first historical data includes:

[0054] Performing a first preprocessing on the first historical data;

[0055] Performing a first classification on the first historical data after the first preprocessing to obtain first classification data and second classification data;

[0056] Establishing a first prediction model according to the first classification data;

[0057] A second prediction model is established based on the second classification data.

[0058] In an optional embodiment, the first preprocessing may include operations such as interpolation fitting, data cleaning, and data normalization to ensure that the data quality meets the requirements of model training.

[0059] In an optional embodiment, the first classification data and the second classification data may be divided based on the feature difference of the data. For example, the first classification data may include features related to the power output of the wind turbine, while the second classification data may include features related to the load demand. Through such classification, the inherent laws of different data sets can be captured more accurately, thereby improving the performance of the prediction model.

[0060] In an optional embodiment, the first classification operation can obtain the first classification data and the second classification data in different ways. For example, cluster analysis, principal component analysis (PCA) or a classification algorithm based on machine learning can be used to classify the data. Cluster analysis can divide the data into different clusters according to the intrinsic structure of the data, while PCA reveals the main features of the data by reducing the dimension. Machine learning classification algorithms, such as support vector machines (SVM) or random forests, can effectively classify according to the labels or features of the data.

[0061] In the embodiment of the present application, the first preprocessing includes operations such as interpolation fitting, data cleaning, and data normalization. The specific steps are as follows:

[0062] In terms of interpolation fitting, the cubic natural spline interpolation method is applied to achieve fitting sampling of historical data. Specifically, the data is segmented, and the second-order derivatives of the nodes are calculated using the three-bend distance equation to obtain the spline coefficients of each segment of data, and finally the node sorting is completed. The wind speed curve is known to be v=f(t), and the cubic spline function is used to interpolate and construct v=s(ρ). The measured wind speed data is time series data, which is described in segments:

[0063]

[0064] Among them, the function S(t) is a polynomial with a maximum order of 3. i-1 ,t i ] has a second-order derivative in the range. The function S(t) needs to satisfy the connection condition and boundary condition to realize interpolation, that is,

[0065] S(t i -0)=S(t i +0),S'(t i -0)=S'(t i +0),S”(t i -0)=S”(t i +0)(i=1,2,…,n-1)

[0066] S”(t0)=f(t0),S”(t n )=f(t n ) (2)

[0067] Furthermore, in terms of data cleaning, the adjacent data replacement method is used to replace abnormal data in historical data, and the mean interpolation method is used to supplement data where data is missing. The specific expression is:

[0068] x i =x i-1 (i=2,3,…,N)

[0069]

[0070] Among them, x i ,x j ∈X(i,j=1,2,…,N) (N is the length of the time series).

[0071] Furthermore, in order to eliminate the influence of the difference in the range of eigenvalues, prevent overfitting and improve generalization ability, it is necessary to normalize the data. Min-max normalization, that is, deviation normalization, is used to normalize the original data so that the result is mapped to [0,1]. The conversion function is

[0072]

[0073] Where x is the original data, x' is the normalized data, xmax and xmin are the maximum and minimum values ​​of the original data, respectively.

[0074] Furthermore, in order to simplify the data structure and extract important information, principal component analysis is performed on the preprocessed multivariate data, that is, by calculating the correlation matrix of the data matrix of multiple input variables, calculating the cumulative variance contribution according to the eigenvalues ​​of the correlation matrix, and then determining the principal components according to the eigenvectors of the correlation matrix.

[0075] Assuming that the environmental data has m measurement indicators and n sets of measurement data, construct an n×m matrix, that is

[0076]

[0077] Among them, x represents a single measurement indicator, and X represents a matrix composed of multiple groups of multivariate measurement indicators.

[0078] Furthermore, the standard matrix Z is generated by the standardization process of formula (5), that is,

[0079]

[0080] In the formula, and S j Represents the indicator variable x j The mean and variance of .

[0081] Furthermore, the correlation matrix R is obtained by formula (6), and its eigenvalues ​​and eigenvectors are calculated, that is,

[0082] R=Z T Z / (n-1) (7)

[0083] Get the eigenvalues ​​λ1,λ2,…,λ of the correlation matrix R m and the corresponding eigenvectors μ1,μ2,…,μ mFinally, the variance contribution rate and cumulative variance contribution rate are calculated by equation (8) and equation (9) respectively to determine the number of principal components. Usually, the first p principal components with cumulative variance contribution rates between 75% and 95% are selected.

[0084]

[0085] In an optional embodiment, the first prediction model and the second prediction model may select specific prediction model types according to the needs of relevant technical personnel, for example, a wind power model, a load power model, a temperature prediction model, etc. may be selected.

[0086] In the embodiment of the present application, the first prediction model is a wind power prediction model, and the second prediction model is a load power model;

[0087] In the embodiment of the present application, establishing the first prediction model and the second prediction model according to the first historical data includes:

[0088] The first prediction model is any model whose input is some data in the first historical data and whose output is wind power or related parameters that can directly or indirectly obtain wind power;

[0089] The second prediction model is any model whose input is some data in the first historical data and whose output is load power or relevant parameters that can directly or indirectly obtain load power.

[0090] In an optional embodiment, the establishment of the first prediction model and the second prediction model can be implemented by a variety of techniques, for example, machine learning algorithms such as support vector machines (SVM), neural networks, decision trees or random forests can be used. These models learn the relationship between input and output through training data sets, so that new input data can be accurately predicted. During the training process, model parameters will be adjusted to minimize prediction errors and improve prediction accuracy.

[0091] In an optional embodiment, in order to improve the generalization ability of the prediction model, a cross-validation method can be used to evaluate the performance of the model. Cross-validation divides the data set into multiple subsets, and uses one part as the test set and the rest as the training set in turn, so as to more comprehensively evaluate the performance of the model on different data.

[0092] In an optional embodiment, the establishment of the prediction model also needs to consider the real-time update of data and the online learning ability of the model. As new data is continuously acquired, the model needs to be able to quickly adapt to the new data distribution to maintain the accuracy and reliability of the prediction.

[0093] In an optional embodiment, the relevant parameters of wind power that can be directly or indirectly obtained may include environmental factors such as wind speed, wind direction, temperature, humidity, etc. These parameters are crucial to the wind power prediction model because they directly affect the efficiency and output power of wind power generation. Wind power prediction values ​​can be directly or indirectly obtained through these parameters, so in essence, wind power prediction values ​​are still obtained, and such methods should be within the scope of protection of this application.

[0094] In an optional embodiment, the relevant parameters of load power that can be directly or indirectly obtained may include information such as power consumption, peak and valley periods, user type, weather conditions, etc. These parameters are equally important for the load power prediction model because they affect the load demand and power consumption pattern of the power grid. Through these parameters, the load power prediction value can be obtained indirectly or directly, thereby providing decision support for power grid dispatch and power market. Therefore, it is essentially still the load power prediction value, and such methods should be within the scope of protection of this application.

[0095] In the embodiments of the present application, Figure 2 As shown in the figure, a data-driven predictive control method for a high proportion of renewable energy distribution system includes five parts, among which an ultra-short-term wind speed prediction model based on LSTM is constructed. The input is the meteorological data and power generation data within the T time scale obtained after data preprocessing and principal component analysis, and the output is the predicted wind speed under the same time scale.

[0096] In the embodiment of the present application, the mathematical expression corresponding to the internal structure of the LSTM unit is shown as follows.

[0097] f t =σ(W f x t +U f h t-1 +b f )

[0098] i t =σ(W i x t +U i h t-1 +b i )

[0099] o t =σ(W o x t +U o h t-1 +b o )

[0100] g t =tanh(W g x t +Ug h t-1 +b g )

[0101] c t =f t ⊙c t-1 +i t ⊙g t

[0102] h t =o t ⊙tanh(c t ) (10)

[0103] In the formula, W and U represent weight matrices, b represents bias term, σ represents activation function, and x t Input for the current time, ht -1 is the external state at the previous moment, and ⊙ represents the vector element multiplication operation.

[0104] In an optional embodiment, if Figure 3 This is a flow chart for ultra-short-term wind power generation prediction. The LSTM unit constructed based on the long short-term memory neural network is spliced ​​with the fully connected layer (FC layer) to form a prediction model. The specific steps are: pre-process the collected environmental data of the distributed wind turbine and construct a sample set after processing, substitute the sample set data into the LSTM recurrent neural network for model training, and establish an ultra-short-term wind speed prediction model. In order to improve the prediction accuracy of the network, when predicting the next 15 minutes of data, the actual meteorological data and power generation data at the current moment are added to the original data set and the network is trained again, realizing a cyclic rolling update every 15 minutes.

[0105] In an alternative embodiment, Figure 4 In the LSTM recurrent neural network, the preprocessed training data set is input into the LSTM recurrent neural network and processed through the forget gate, input gate and output gate of the network. In order to avoid overfitting, a random dropout layer is introduced into the LSTM recurrent neural network. However, the data dimension of the random dropout output is different from the desired prediction data dimension. A fully connected layer is introduced for dimensional transformation, and finally the predicted wind speed data is output through an output layer.

[0106] In an optional embodiment, after obtaining the predicted wind speed data, a wind turbine power model is established according to the physical characteristics of the wind turbine. The input of the model is the predicted wind speed within the time scale T, and the output is the corresponding predicted output power of the wind turbine.

[0107] In the embodiment of the present application, the fan output power curve can be approximated as a piecewise function:

[0108]

[0109] In the formula, v r Indicates rated wind speed (m / s); v ci Indicates the cut-in wind speed (m / s); v co Indicates the cut-out wind speed (m / s); P r It represents the rated power of the wind turbine (W), and its calculation formula is:

[0110]

[0111] Where CP is the wind energy utilization coefficient; ρ is the air density (kg / m3); A is the swept area of ​​the wind turbine blade (m2), A = πR2; R is the radius of the rotor rotation surface (m); v w Indicates actual wind speed (m / s).

[0112] In an embodiment of the present application, the input of the LSTM-based short-term load power prediction model is the load power data within the T time scale obtained after data preprocessing and principal component analysis, and the output is the predicted load power under the same time scale. The structure of the LSTM network is similar to that of the wind power prediction model. In order to retain time series information and increase historical data samples, a sliding window method is used to construct a training data set.

[0113] It should be noted that the application of traditional load forecasting models has certain limitations due to the complex and changeable meteorological conditions that will have a certain impact on the load of the power system. Therefore, meteorological factors are introduced to optimize the LSTM model. For daily power load, only the meteorological factors of the day will affect the power load of the day. Therefore, the input of the model is changed to the power load data in the sliding window and the power load data of the day, such as Figure 4 shown.

[0114] S103, establishing a third prediction model, the input of the third prediction model is the output of the first prediction model and the second prediction model;

[0115] In the embodiment of the present application, establishing the third prediction model includes:

[0116] The third prediction model is an arbitrary model whose input is the output of the first prediction model and the second prediction model, and whose output is the first performance indicator of the target power distribution system;

[0117] The first performance indicator includes at least one of voltage deviation, power loss, frequency stability index or other performance indicators.

[0118] In an optional embodiment, the third prediction model can be implemented by constructing an integrated learning model. The integrated learning model can use a variety of different prediction models as base learners, such as support vector machine (SVM), random forest (RF), gradient boosting decision tree (GBDT), etc. These base learners can be trained in parallel, and their prediction results can be input as features into a meta learner to obtain the final prediction results. By integrating the advantages of different models, the accuracy and robustness of the prediction can be improved.

[0119] In an optional embodiment, the meta-learner of the third prediction model can also adopt a linear regression model, the purpose of which is to weight the prediction results of the base learners to obtain the final prediction output. The determination of the weights can be optimized by methods such as cross-validation to ensure the generalization ability of the model in different situations.

[0120] In the embodiment of the present application, a frequency stability index is selected as the first performance indicator;

[0121] In the embodiment of the present application, the specific steps of establishing the third prediction model may be as follows:

[0122] In the power system, frequency stability is a direct reflection of the balance between supply and demand, especially in the distribution network with a high proportion of renewable energy connected to the grid, where frequency stability is particularly critical, because the output power of renewable energy such as wind and solar energy fluctuates greatly, which can easily cause fluctuations in grid frequency.

[0123] Define the output of the second-layer LSTM network. This application uses FSI (Frequency Stability Index) to analyze the frequency response characteristics of the distribution system when facing load changes or power supply fluctuations, and then quantify the system's ability to resist frequency disturbances, providing the necessary voltage trajectory reference for subsequent inverter control, and ensuring the reliability and efficiency of power grid operation.

[0124] In the embodiment of the present application, the transient frequency stability index of the power system may be the sum of the integrals of different frequency bands with different weights, and its mathematical expression is as follows:

[0125]

[0126] in

[0127]

[0128] Among them, K j is the weight coefficient of the frequency band; f[t i ] is t i The frequency of the moment; f N Indicates the rated frequency of the power grid; Δt iis the step size for frequency response calculation; f j is the upper / lower frequency of the band.

[0129] In the embodiment of the present application, the calculation of the transient frequency stability index can be divided into the following steps. First, the range that the frequency response curve can reach is divided into multiple frequency bands, and then different weights are assigned to different frequency bands. Finally, the transient frequency stability index can be obtained by weighted integration. If F ≥ 1, the transient frequency of the system will be unstable, and corresponding control measures should be taken at this time; if F < 1, the transient frequency of the system will be stable.

[0130] In the embodiment of the present application, a frequency stability index calculation formula applicable to a certain area power grid is given below:

[0131]

[0132] It should be noted that if Figure 2 As shown in the figure, the second-layer prediction network is constructed, that is, the wind turbine power generation and load active power at the time scale T predicted by the first-layer LSTM network are used as network input, and the FSI at the same time scale is used as the prediction output to describe the nonlinear relationship between power and frequency stability index.

[0133] S104: Establish a first predictive control strategy based on the third predictive model, and perform data-driven predictive control according to the first predictive control strategy.

[0134] In the embodiment of the present application, the first predictive control strategy includes:

[0135] Establishing a first objective function of a first performance indicator output by a third prediction model;

[0136] Solving the first objective function to obtain optimal control parameters;

[0137] The inverter output in the target power distribution system is adjusted according to the optimal control parameters.

[0138] In the embodiment of the present application, the inverter output includes at least the power and attack angle of the inverter.

[0139] In an embodiment of the present application, the first predictive control strategy also includes: establishing a first constraint condition on a first performance indicator output by the third predictive model.

[0140] In the embodiment of the present application, a first prediction control strategy is established based on the third prediction model, and the specific steps of performing data-driven prediction control according to the first prediction control strategy may be as follows:

[0141] Based on the prediction of the frequency stability index in the future T time scale, combined with the model predictive control (MPC) strategy, that is, the first predictive control strategy, the voltage deviation, power loss and other performance indicators (such as FSI) are minimized, so as to accurately adjust the voltage reference vector required for the subsequent control inverter to ensure that the output of the inverter accurately follows the predetermined trajectory. The general steps for implementing this control strategy are as follows:

[0142] First, the changes of FSI are predicted using historical and real-time data, and a dynamic model of the power grid is established, including the influence of the inverter's active power output and the angle of attack on the grid frequency. The discrete state space equation of the inverter is established:

[0143]

[0144] in

[0145]

[0146] Where T s is the sampling time. The changes in frequency, mechanical power, and electromagnetic power can be written as

[0147]

[0148] Secondly, the optimization goal of minimizing FSI is set in MPC, that is, the first objective function. For each control cycle, MPC calculates the optimal control strategy within a period of time by solving a constrained optimization problem based on the current grid state and the predicted value of FSI. The constraints for solving the optimization problem include but are not limited to the upper and lower limits of power output and the range of angle of attack.

[0149] Design cost function, including VSG frequency deviation Δω, active power deviation ΔP m and FSI deviation ΔF, and set the optimization direction to minimize the weighted sum of squares.

[0150]

[0151] Where α, β and γ represent the weights of frequency, power and FSI changes respectively. Therefore, the MPC optimization problem with constraints (i.e., the first constraint condition) can be described as

[0152] J p (Δω(k),ΔP m (k),ΔF(k))(20)

[0153] Finally, the MPC optimization problem is solved, and the actual output of the inverter is adjusted according to the MPC optimization results. Based on the calculation of the optimal active and reactive power, the frequency optimization and voltage dynamic response are achieved in the repeated correction of the inverter reference power, thereby adjusting the operating status of the wind turbine and the overall power supply and demand balance of the power grid, and ensuring the stability of the power grid frequency.

[0154] In summary, the present invention proposes a data-driven predictive control method, which obtains the first historical data of the target distribution system; establishes a first prediction model and a second prediction model based on the first historical data; establishes a third prediction model, and the input of the third prediction model is the output of the first prediction model and the second prediction model; establishes a first predictive control strategy based on the third prediction model, and performs data-driven predictive control according to the first predictive control strategy. By utilizing the powerful time series data processing capability of the double-layer long short-term memory network (LSTM), it is possible to predict future wind turbine power output and load demand based on historical multivariate data, and then accurately predict the transient frequency stability index (FSI). This prediction not only supports online applications, but also provides a stability margin, which buys valuable time for emergency control measures of the distribution network and ensures the efficient and reliable operation of the distribution system. At the same time, the present invention adjusts the control parameters of the model predictive control (MPC) strategy, such as the voltage reference vector, and optimizes the inverter output according to the predicted FSI, so that the system can determine transient stability problems in advance, thereby performing advanced control, and effectively improving the frequency stability of the system. The method of the present invention can be directly deployed on power electronic equipment without adding additional communication equipment and sensors to the system, overcoming the shortcomings of high construction cost and poor real-time performance in traditional methods, thereby providing an efficient, economical and easy-to-implement solution.

[0155] Example 2

[0156] In a preferred embodiment, the single-machine topology diagram of the virtual synchronous inverter under the model predictive control in the data-driven predictive control method of the present invention is as follows: Figure 5 The main power circuit is a three-phase full-bridge inverter, and the midpoint of the bridge arm is connected to the LC filter; u o Represents the voltage at the filter capacitor terminal; i o Represents the output current of the inverter. Establish a complete mathematical model of the topology and establish its simulation model in the time domain simulation software. Set the simulation time to 5s, and the wind power generation system output power drops sharply at 2s. Record the frequency data during the simulation, that is, a total of 5s of data.

[0157] First, for distributed wind turbines, data related to the surrounding environment and wind turbine power generation data are collected, including historical wind speed, wind direction, solar radiation, temperature, air pressure, cumulative precipitation and humidity and other environmental data for at least the past 90 days, as well as wind turbine power generation data at the corresponding time. At the same time, the distribution network load power data collected during the same period ensures the time synchronization of the data, providing an accurate basis for data analysis.

[0158] In the data preprocessing stage, the missing data are first supplemented by linear interpolation to ensure the integrity of the data series. Next, data cleaning is performed and the standard deviation method is used to remove outliers in the data, that is, data points exceeding ±3 times the standard deviation of the mean are removed to eliminate noise and obvious erroneous records.

[0159] For data normalization, the MinMaxScaler function is used, which scales each feature to a given minimum and maximum value (usually 0 to 1) and converts it according to the following formula:

[0160]

[0161] In order to simplify the data structure and extract key information, principal component analysis was performed on the preprocessed multivariate data, the cumulative variance contribution was calculated, and the principal components with cumulative contributions exceeding 85% were selected as input data.

[0162] Calculations show that the cumulative contribution of historical wind speed is far greater than environmental data such as wind direction, temperature, air pressure, and humidity, and is the main component of the distributed extension predicted wind speed.

[0163] Secondly, the construction of ultra-short-term wind power prediction model and load power prediction model based on LSTM mainly lies in building LSTM unit, which is composed of forget gate, input gate and output gate. The neurons of LSTM unit in this example are 1.5 times the number of input data each time, that is, assuming that the input time length data is T, and each data consists of l data points, then the number of neurons of LSTM unit is set to 1.5l.

[0164] The model consists of 1 input layer, 5 hidden layers (LSTM layer, Dropout layer and Dense layer) and 1 output layer. The input is the meteorological data and power generation data in the T time scale obtained after data preprocessing and principal component analysis, and the output is the predicted wind speed in the same time scale. The length of the input sequence of the input layer is n, and the input dimension is 24 (data from 0 to 23 hours); the length of the output sequence of the output layer is n, and the output dimension is 24 (data from 0 to 23 hours). The hidden layer consists of 2 LSTM layers, 2 Dropout layers and 1 Dense layer. The number of hidden neurons in the LSTM layer is set to 128; the probability of the Dropout layer is set to 0.5 based on experience to prevent the occurrence of overfitting; the parameter of the Dense layer is set to 24, that is, the final output is the result of 0-23 hours data.

[0165] In order to introduce a certain degree of nonlinearity, it is necessary to set an Activation layer between the hidden layer and the output layer. Here, the Sigmoid function is selected as the activation function of the Activation layer.

[0166] Set the parameters of the LSTM model: the number of training times is 100, the number of batches is 16, the number of hidden neurons is 128, and the loss function is MSE. Divide the preprocessed data into data sets. Since the time series prediction is performed on the data from 0 to 23 hours, it is necessary to divide them in chronological order. After multiple comparative experiments on parameter settings, it is found that when the first 75% of the data set is divided into the training set and the last 25% is divided into the verification set, the MSE value of the model is the smallest. At this time, the model has the best training effect and the highest prediction accuracy.

[0167] After obtaining the predicted wind speed data, a wind turbine power model is established based on the specific physical characteristics of the wind turbine. The input of the model is the predicted wind speed within the time scale T, and the output is the corresponding predicted output power of the wind turbine.

[0168] Again, the construction of the short-term load power forecasting model based on LSTM mainly involves using the load power data in the T time scale after data preprocessing and principal component analysis as input, and the output is the predicted load power in the same time scale. The structure of this LSTM network is similar to that of the wind power forecasting model and will not be repeated here. In order to evaluate the performance of the short-term load power forecasting model, the root mean square error (RMSE) and mean absolute error (MAE) are introduced as loss functions. After multiple rounds of training, until the loss function converges to the minimum, the target is set to less than 5% to ensure that the predicted load power has high accuracy and reliability.

[0169] Next, the range that the frequency response curve can reach is divided into multiple frequency bands, and then different weights are assigned to different frequency bands. Finally, the transient frequency stability index can be obtained by weighted integration using the frequency stability index calculation formula:

[0170]

[0171] The LSTM-based FSI prediction model specifically processes the wind turbine power generation and load power data predicted by steps 2 and 3, and predicts the FSI value for the next hour.

[0172] The model consists of 1 input layer, 5 hidden layers and an output layer. The length of the input sequence of the input layer is n, and the input dimension is 3 (predicted wind power and load power, frequency stability index FS); the length of the output sequence of the output layer is n, and the output dimension is 3 (same as above). The hidden layer consists of 2 LSTM layers, 2 Dropout layers and 1 Dense layer. The number of hidden neurons in the LSTM layer is set to 128; the probability of the Dropout layer is set to 0.1 based on experience to prevent the occurrence of overfitting; the parameter of the Dense layer is set to 2, which is the result of the final output frequency stability index FSI.

[0173] The nonlinear relationship between power and frequency stability index is learned through a deep neural network, and multiple rounds of training are performed until the loss function converges to a minimum, with the target set below 5% to provide a highly accurate FSI evaluation.

[0174] Finally, with minimizing FSI as the optimization goal, the MPC controller is dynamically adjusted according to the latest grid state and FSI prediction results. For each control cycle, MPC calculates the optimal control strategy within a period of time by solving a constrained optimization problem based on the current grid state and the predicted value of FSI.

[0175] J p (Δω(k),ΔP m (k),ΔF(k))

[0176] Solve the MPC optimization problem and adjust the actual output of the inverter according to the MPC optimization results. Based on the calculation of the optimal active and reactive power, the frequency optimization and voltage dynamic response are achieved in the repeated correction of the inverter reference power, ensuring that the inverter output accurately responds to the grid demand, thereby enhancing the overall stability of the distribution network.

[0177] The parameters of the inverter and the MPC controller are set as shown in the following table:

[0178] Table 1: Main parameters of the inverter.

[0179]

[0180]

[0181] Table 2: MPC controller parameters.

[0182]

[0183] The results are as follows Figure 6 As shown, the system frequency range is roughly between 49.5Hz and 50.1Hz, and the maximum frequency deviation is approximately 0.49Hz, 0.33Hz, and 0.28Hz. In summary, the method proposed in this application can effectively achieve accurate prediction and control of transient stability of power grid frequency, thereby improving the stability of the distribution system under high penetration of renewable energy.

[0184] Example 3

[0185] This embodiment also provides a data-driven predictive control system, including:

[0186] A data acquisition module, used for acquiring first historical data of a target power distribution system;

[0187] A first model building module, used to build a first prediction model and a second prediction model according to the first historical data;

[0188] The second model building module is used to build a third prediction model, the input of the third prediction model is the output of the first prediction model and the second prediction model;

[0189] The control module is used to establish a first prediction control strategy based on the third prediction model and perform data-driven prediction control according to the first prediction control strategy.

[0190] The above-mentioned unit modules may be embedded in or independent of a processor in a computer device in the form of hardware, or may be stored in a memory in a computer device in the form of software, so that the processor can call and execute operations corresponding to the above-mentioned modules.

[0191] This embodiment also provides an electronic device, which includes a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a busbar pressure loss load transfer analysis method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0192] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method proposed in the above embodiment is implemented.

[0193] The storage medium proposed in this embodiment and the method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0194] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0195] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0196] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of complete hardware embodiments, complete software embodiments, or embodiments in combination with software and hardware. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The scheme in the embodiments of the present application can be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0197] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0198] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0199] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1The steps for the functions specified in one or more boxes.

[0200] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0201] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A data-driven predictive control method, characterized in that: include: Acquiring first historical data of a target power distribution system; Establishing a first prediction model and a second prediction model according to the first historical data; Establishing a third prediction model, wherein the input of the third prediction model is the output of the first prediction model and the second prediction model; A first predictive control strategy is established based on the third predictive model, and data-driven predictive control is performed according to the first predictive control strategy.

2. The data-driven predictive control method according to claim 1, characterized in that: The establishing of a first prediction model and a second prediction model according to the first historical data includes: The first prediction model is any model whose input is some data in the first historical data and whose output is wind power or related parameters of wind power that can be directly or indirectly obtained; The second prediction model is any model whose input is some data in the first historical data and whose output is load power or related parameters of load power that can be directly or indirectly obtained.

3. The data-driven predictive control method according to claim 2, characterized in that: The establishing of the third prediction model comprises: The third prediction model is an arbitrary model whose input is the output of the first prediction model and the second prediction model, and whose output is the first performance indicator of the target power distribution system; The first performance indicator includes at least one of voltage deviation, power loss, frequency stability index or other performance indicators.

4. The data-driven predictive control method according to claim 3, characterized in that: The first predictive control strategy includes: Establishing a first objective function of a first performance indicator output by the third prediction model; Solving the first objective function to obtain optimal control parameters; The inverter output in the target power distribution system is adjusted according to the optimal control parameter.

5. The data-driven predictive control method according to claim 4, characterized in that: The inverter output includes at least the power and the angle of attack of the inverter.

6. The data-driven predictive control method according to claim 5, characterized in that: The establishing of a first prediction model and a second prediction model according to the first historical data includes: performing a first preprocessing on the first historical data; Performing a first classification on the first historical data after the first preprocessing to obtain first classification data and second classification data; Establishing a first prediction model according to the first classification data; A second prediction model is established based on the second classification data.

7. The data-driven predictive control method according to claim 6, characterized in that: The first predictive control strategy also includes: establishing a first constraint condition on a first performance indicator output by the third predictive model.

8. A data-driven predictive control system, characterized in that: include: A data acquisition module, used for acquiring first historical data of a target power distribution system; A first model building module, used to build a first prediction model and a second prediction model according to the first historical data; A second model building module, used to build a third prediction model, the input of the third prediction model is the output of the first prediction model and the second prediction model; A control module is used to establish a first prediction control strategy based on the third prediction model and perform data-driven prediction control according to the first prediction control strategy.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the busbar pressure loss load transfer analysis method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the busbar voltage loss load transfer analysis method according to any one of claims 1 to 7.

Images