A multi-energy complementary system double-layer energy optimization scheduling method

Abstract

The application discloses a kind of multi-energy complementary system double-layer energy optimization scheduling method, comprising: based on Nataf conversion obtains the correlation coefficient matrix of each variable in normal space and obtains sample matrix group;With sample matrix group as input variable, 2PEM evaluation and series expansion fitting are carried out, and probability density function is obtained;Quadratic programming model is constructed, and quadratic programming model is solved based on improved efficient set method;Based on risk aversion strategy, construct robust optimization model, avoid the influence of wind light load uncertainty on energy storage configuration result;MPC rolling optimization model is constructed to reduce deviation;Based on MPC rolling optimization model and robust optimization model, construct double-layer energy optimization scheduling framework;Based on double-layer energy optimization scheduling framework, correct day-ahead scheduling plan.The application adopts MPC to carry out rolling optimization, and according to actual situation, deviation is corrected on the basis of day-ahead IGDT robust scheduling plan, and the shortcomings of IGDT open-loop control are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of energy optimization scheduling, and particularly relates to a double-layer energy optimization scheduling method for a multi-energy complementary system. BACKGROUND

[0002] In recent years, with the development of society and the continuous improvement of people's living standards, people's demand for various energy is increasing rapidly, which leads to a rapid increase in global electricity consumption, and thus increasingly prominent problems. On the one hand, traditional fossil energy such as coal is running out due to the increasing consumption; on the other hand, the use of a large amount of fossil energy has made environmental problems increasingly prominent. Therefore, developing clean energy and improving the energy consumption structure have become an important method for the world to solve the problems of energy shortage and environmental pollution in recent years. In order to explore a sustainable development path, researchers focus on clean energy such as wind power, photovoltaic and tidal energy, and use new energy to replace traditional fossil energy. At present, clean energy such as wind power and photovoltaic has been rapidly developed in the world because of its renewability and non-pollution. With the increase of various new energy capacities of the multi-energy complementary system, not only the energy imbalance problem caused by the introduction of multiple uncertain factors needs to be considered, but also the influence of the output correlation between uncertain factors on the system needs to be considered.

[0003] The organic coordination and complementation of various heterogeneous energies in the system can realize energy cascade utilization, and by integrating internal resources, the primary energy utilization rate can be improved. However, due to a large number of uncertain interference factors such as renewable energy represented by wind energy and solar energy, combined with the uncertainty of load demand growth, the uncertainty in the system is increasingly prominent. Therefore, the source-load prediction error and uncertainty make the system operation more complex, which brings great challenges to the optimization scheduling of various heterogeneous energies. SUMMARY

[0004] The purpose of the present application is to provide a double-layer energy optimization scheduling method for a multi-energy complementary system to solve the problems existing in the prior art.

[0005] To achieve the above purpose, the present application provides a double-layer energy optimization scheduling method for a multi-energy complementary system, comprising:

[0006] obtaining a correlation coefficient matrix of each variable in a normal space based on Nataf transformation;

[0007] obtaining a sample matrix group based on the correlation coefficient matrix;

[0008] performing 2PEM evaluation and series expansion fitting on the sample matrix group as input variables to obtain a probability density function;

[0009] A quadratic programming model is constructed, and the model is solved based on the improved effective set method.

[0010] A robust optimization model is constructed based on a risk avoidance strategy, and the impact of wind and solar load uncertainties on energy storage configuration results is avoided based on the robust optimization model.

[0011] Construct an MPC rolling optimization model, and reduce the deviation based on the MPC rolling optimization model;

[0012] A two-layer energy optimization scheduling framework is constructed based on the MPC rolling optimization model and the robust optimization model.

[0013] Energy optimization scheduling is performed based on the aforementioned two-layer energy optimization scheduling framework.

[0014] Optionally, the process of obtaining the correlation coefficient matrix of each variable in the normal space based on the Nataf transform includes:

[0015] Independent sequences are generated in the normal space, and the independent sequences are transformed into correlated sequences based on correlation control. The correlated sequences are then mapped to the non-normal space based on the Nataf transform.

[0016] Optionally, the process of obtaining the sample matrix group based on the correlation coefficient matrix includes:

[0017] The correlation coefficient matrix is ​​decomposed into a sample matrix based on SVD decomposition, and a sample matrix group is constructed based on the sample matrix that is the same as the rank correlation matrix of the desired correlation coefficient matrix.

[0018] Optionally, the process of using the sample matrix set as input variables for 2PEM evaluation and series expansion fitting includes:

[0019] Based on the correlation coefficient matrix of the input variables, calculate the linear correlation matrix of the standard normal distribution random variables after Nataf transformation, and perform SVD decomposition on the linear correlation matrix of the standard normal distribution random variables to obtain the unitary matrix and the singular matrix.

[0020] Calculate the sampled values ​​and weights in the normally independent space and the standard normal space, respectively;

[0021] The sampled values ​​are used to form an evaluation matrix, and the sample matrix in the standard normal space is transformed into a sample matrix in the input variable space based on the inverse Nataf transform.

[0022] The power deficit was evaluated using the 2PEM-Cornish method, and the original moments of each order were obtained and fitted to obtain the probability density function.

[0023] Optionally, the robust optimization model is:

[0024]

[0025] Wherein, f is the objective function; X is the decision variable matrix; d is the uncertainty variable matrix; h and g are the equality constraint and inequality constraint respectively, σ is the deviation factor, that is, the deviation degree between the expected target value and the optimal solution of the deterministic model, the greater σ is, the greater f c is, the greater the risk aversion degree is, and the stronger the robustness is.

[0026] Optionally, the process of constructing the MPC rolling optimization model comprises:

[0027] Constructing a prediction model;

[0028] Optimizing the prediction model based on the objective function;

[0029] Feedback correction is performed on the performance deviation of the objective function.

[0030] Optionally, the prediction model is:

[0031]

[0032] Wherein: X(t) represents the system state quantity of the controlled object; Y(t) represents the system output quantity; Δu(t) represents the control quantity or the related input quantity of the object; ω(t) represents the system disturbance quantity; A, B and C are the system matrix, the input matrix and the output matrix respectively.

[0033] Optionally, the objective function is:

[0034]

[0035] Wherein, f obj is the objective function of rolling optimization; Y p (t+i|t) and Y r (t+i|t) are the prediction value and the reference value at t+i time at t time respectively; w v and w u are the system output deviation weight coefficient and the control quantity weight coefficient.

[0036] Optionally, the feedback correction method is:

[0037]

[0038] Wherein, e k is the system control error; λ k is the error coefficient matrix; Y R is the actual output of the system at t time.

[0039] The technical effect of the present application is:

[0040] In order to ensure the economic operation of the multi-energy complementary system RIES under the influence of uncertain factors, the application establishes a double-layer multi-heterogeneous energy optimization scheduling model based on IGDT-MPC. The two-point method and Cornish-Fisher series expansion are used to convert the multi-element uncertainty into a single power shortage uncertainty, effectively solving the problem that the IGDT model cannot accurately process multiple uncertain factors at the same time, and aiming at the scene that the two-point method cannot be directly applied to the input variables with correlation, the application introduces Nataf transformation and SVD decomposition technology, so that the applicability of the 2PEM-Cornish method is more extensive. The MPC is used for rolling optimization in the day, and the deviation is corrected according to the actual situation in the day on the basis of the IGDT robust scheduling plan in the day, which improves the shortcomings of the IGDT open-loop control, so that the scheduling plan is more suitable for the actual operation state. And the advantages of the MPC day rolling optimization model are played, the control quantity obtained at the last control time is taken as the initial value at the next control time, which greatly improves the calculation efficiency of the effective set method and reduces the calculation cost. Therefore, the IGDT-MPC double-layer optimization scheduling model considering variable correlation provided by the application provides a new idea for the IGDT to process the scene with correlated input variables. BRIEF DESCRIPTION OF DRAWINGS

[0041] The drawings constituting a part of the application are used to provide further understanding of the application, the illustrative embodiments of the application and the description thereof are used to explain the application, and do not constitute improper limitation on the application. In the drawings:

[0042] Figure 1 It is the correlation processing algorithm flowchart in the embodiment of the application;

[0043] Figure 2 It is the improved effective set algorithm flowchart in the embodiment of the application;

[0044] Figure 3 It is the double-layer energy optimization scheduling framework structure diagram in the embodiment of the application;

[0045] Figure 4 It is the power shortage expectation value comparison diagram under different wind-light correlation degrees in the embodiment of the application;

[0046] Figure 5 It is the power shortage standard deviation value comparison diagram under different wind-light correlation degrees in the embodiment of the application;

[0047] Figure 6 It is the power shortage probability distribution curve result comparison diagram of using 2PEM-Cornish-Fisher fitting and using Monte-Carlo fitting in the embodiment of the application;

[0048] Figure 7This is a comparison diagram of the cumulative probability distribution of power deficit between the method of the present invention and the MCSM method in this embodiment of the invention;

[0049] Figure 8 This is a power supply optimization scheduling diagram in an embodiment of the present invention;

[0050] Figure 9 This is a gas supply optimization scheduling diagram in an embodiment of the present invention;

[0051] Figure 10 This is a heating optimization scheduling diagram in an embodiment of the present invention. Detailed Implementation

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0054] Example 1

[0055] like Figures 1-10 As shown, this embodiment provides a two-layer energy optimization scheduling method for a multi-energy complementary system, including:

[0056] Multi-random Factor Power Shortage Probability Model Considering Variable Correlation

[0057] Traditional IGDT models struggle to account for the correlation between variables. This invention combines Nataf transformation and singular value decomposition techniques with 2PEM-Cornish. By utilizing Nataf transformation and inverse transformation, the relationship between multidimensional standard normal space and multidimensional non-normal space is established. Through inverse Nataf transformation, sampling points in independent standard normal space can be transformed into related non-normal variable space, enabling the 2PEM-Cornish method to obtain the digital characteristics of power deficit while taking into account the correlation of input variables.

[0058] Nataf transformation

[0059] There are a large number of non-normal variables with correlation in RIES, and it is difficult to obtain the complete probability distribution of the variables, but it is relatively easy to obtain the marginal distribution and correlation coefficient matrix of each variable. At present, the common method is to first generate independent sequences in the normal space, and then transform them into sequences containing correlation by using correlation control, and finally map them to the non-normal space by transformation. The commonly used transformation methods include polynomial normal transformation and Nataf transformation, and the accuracy of Nataf transformation is higher. Therefore, the distribution information of the input variables of stochastic power flow is reconstructed by using Nataf transformation, and the process is briefly described as follows.

[0060] For m-dimensional input variables X = [X1, X2, X3, …, X m ] T , assuming that the marginal PDF of each random variable X i is f i (x i )(i = 1, …, n), and the corresponding cumulative probability density function is F i (x i )(i = 1, …, n), which can be converted as follows:

[0061]

[0062] The standard normal vector Y = [Y1, Y2, Y3, …, Y m ] T is used to obtain, where Φ(·) and Φ -1 (·) are the cumulative distribution function and the inverse cumulative distribution function of the standard normal variable, respectively.

[0063] According to the Nataf transformation theory and the differential rule of the implicit function, the joint PDF of the random vector X can be expressed as:

[0064]

[0065] In the formula: Φ is the PDF of the standard normal distribution.

[0066]

[0067] is the PDF of the n-dimensional standard normal vector with zero mean, unit variance and correlation coefficient matrix. This distribution is usually called Nataf distribution. Assuming that the correlation coefficient matrix is R X = (R ij ) m╳m , assuming that the corresponding correlation coefficient of the standard normal vector S i is R S , and the relationship between R X and R S is:

[0068]

[0069] where is X i , X j the joint probability density function; μ,σ are the mean and standard deviation of the corresponding variable, respectively; is the correlation coefficient ρ Zij The two-dimensional standard normal distribution probability density function. For equation (4-4) there are a variety of solutions, this paper uses the method of numerical integration to solve.

[0070] SVD decomposition

[0071] The correlation coefficient matrix of the standard normal space is determined, in order to obtain the required correlation coefficient matrix, most methods use Cholesky to decompose the correlation coefficient matrix, but it assumes two conditions: (1) assumes that the randomly arranged sample values are independent (2) assumes that the correlation coefficient matrix is positive definite. But when the number of random variables is very large, there are application scenarios where the correlation coefficient matrix is not positive definite or full rank, its Cholesky decomposition does not exist, but the matrix is guaranteed to be symmetric, and there is SVD decomposition as follows:

[0072]

[0073] where: Q is a lower triangular matrix; D is a diagonal matrix composed of singular values; then the matrix Y can be constructed as follows:

[0074]

[0075] It is easy to prove that the mean of y i is 0, so the correlation coefficient matrix R Y of Y is equal to its corresponding covariance coefficient matrix, and its complete derivation is as follows:

[0076]

[0077] Therefore, the matrix Y is independent.

[0078] When the required correlation coefficient matrix is given, the intermediate correlation coefficient matrix can be solved, and its correlation coefficient matrix is the same as equation (4-5). Then the matrix Z can be constructed as follows:

[0079]

[0080] It is easy to prove that the mean of z i is 0, so the correlation coefficient matrix R Z of Z is equal to its corresponding covariance coefficient matrix, and its complete derivation is as follows:

[0081]

[0082] Then the matrix Z is of the form * In particular, when the correlation coefficient matrix is positive definite, the diagonal matrices D and E are the identity matrix, and the SVD is the Cholesky decomposition. When each row of X is permuted according to the matrix Z, the sample matrix S is obtained, which has the same rank correlation matrix as the matrix Z. Since S and the matrix Z have the same rank correlation matrix, the S correlation matrix is close to R X Finally, each column is formed into a group of samples, which is used as an input variable for 2PEM evaluation and series expansion fitting.

[0083] Power deficit probability density calculation with correlation processing

[0084] In the point estimation method, the input variables are assumed to be independent. The present application combines the Nataf transformation and SVD to process the input variable correlation scenario. In order to consider the correlation of the input variables, the basic processing idea is as follows: first, sampling and probability concentration calculation are performed in the independent normal space; second, the Nataf inverse transformation is used to transform the sampling points to the original sample space; and then, the digital characteristics of the output power deficit are evaluated in the original sample space. The specific steps are as follows:

[0085] (1) According to the correlation coefficient matrix R X of the input variables, the correlation coefficient matrix R s of the standard normal distribution variables after Nataf transformation is calculated. s SVD decomposition is performed on R

[0086] (2) Sampling and probability concentration calculation are performed in the standard normal space. The standard sampling value of the standard normal variable is , and the corresponding probability concentration is 1 / 6.

[0087] (3) The values obtained by sampling are combined into an evaluation matrix, and the independent normal variables are transformed to the original sample space by formula (4-6) and the Nataf inverse transformation.

[0088] (4) In the original sample space, the 2PEM-Cornish method is used for evaluation to obtain the origin moments of each order of the power deficit and fit the probability density function.

[0089] Efficient solution algorithm for MPC quadratic programming

[0090] In order to react to the renewable energy and load demand fluctuation in time, reduce the deviation between day-ahead plan and ultra-short term dispatching, the day-ahead economic optimization dispatching plan is corrected by the MPC model with 15 min control time domain and 1 h prediction time domain. The main objective of MPC is to use the latest information to correct the day-ahead dispatching plan. The MPC problem with linear constraints can be expressed as a quadratic programming. From the perspective of calculation cost, different methods need different time to solve the quadratic programming model. For small RIES, the Matlab solver described in Chapter 3 can fully meet the time requirements. When RIES contains multiple wind farms, photovoltaic power plants and distributed energy supply units, the model solving time has strict requirements. This section introduces the improved active set method to solve the quadratic programming model from the perspective of calculation time.

[0091] Quadratic programming model

[0092] In the past two decades, model predictive control has achieved great success in industrial control as an advanced process control method, which is due to its efficient handling of constraints and other reasons. The solution of MPC with constraints needs to use quadratic programming model. Quadratic programming is a special case of nonlinear programming, whose objective function is a quadratic real function and constraints are linear. Since quadratic programming is relatively simple and easy to solve, quadratic programming algorithm has attracted people's attention early and become an important way to solve nonlinear programming. However, depending on the problem, the computational load of the quadratic programming model may be different. This section introduces the general quadratic programming model.

[0093] The MPC problem can be expressed as a quadratic programming with linear inequality constraints:

[0094]

[0095] where H is the Hessian matrix. If the Hessian matrix is positive definite, there is a globally unique optimal solution. If the Hessian matrix is non-positive definite, it is a non-convex quadratic programming, which has multiple stationary points and local minimum points. There are many algorithms for solving quadratic programming problems, such as Lagrange method, Lemke method, interior point method, active set method, ellipsoid algorithm, etc. And there are still many researchers working on this research work. In mathematical programming, since convex quadratic programming has a special role, researchers have always regarded it as an important topic for research.

[0096] Improved active set method

[0097] Active Set Method (ASM) is suitable for inequality constrained quadratic programming problem which cannot be solved by elimination method and Lagrange method. The main idea is that the known point is feasible point, and only the active constraint of the point is considered to minimize the equation, and the new feasible point is obtained, and the above method is repeated.

[0098] Considering quadratic programming problem with inequality constraints

[0099]

[0100] where H is an n x n symmetric matrix, c is an n-dimensional column vector, A is an m x n matrix, the rank of A is m, b is an m-dimensional column vector, and x ∈ R n .

[0101] Due to the presence of inequality, the elimination method and Lagrange method cannot be directly applied as one of the strategies to solve the problem, but the active set method is used to convert it into an equality constrained problem. By using the active set method, in each iteration, the known feasible point is taken as the starting point, the active constraint of the point is taken as the equality constraint, and the remaining constraints are temporarily ignored, and a new better feasible point is obtained, and the above method is repeated until the optimal solution is obtained. The algorithm flow for solving equation (4-10) is shown in Figure 2 .

[0102] The active set algorithm starts from a feasible solution, and the given of the initial solution affects the solving speed of the algorithm, so when solving the quadratic programming model in this section, the optimal initial solution is considered to improve the solving efficiency of the model. Usually, our initial solution is set to 0, and since MPC is a series of similar quadratic programming models, the quadratic programming equation solved at each time period is approximate, and the day-ahead scheduling plan involves calculating the control amount of the next 4 control time periods at each time, so a better initial solution can be obtained from the model solution obtained from the previous group. If the initial solution is not in the feasible set, a new initial solution must be obtained by using other methods. Through the example, it can be seen that the modified active set method (MASM) improves the solving speed of the algorithm.

[0103] Double-layer optimization scheduling method

[0104] IGDT overview

[0105] The IGDT method is a non-probabilistic, non-possibilistic (non-fuzzy) uncertainty risk management method. The method has been applied to different decision-making processes and risk management fields. Unlike other uncertainty modeling methods for optimization problems (optimization objective function), the IGDT method aims to maximize / minimize the tolerable range of uncertainty while meeting the predetermined objective, and neither needs the PDF (probability distribution function) of the uncertainty parameter used in most probability methods nor the membership function required in fuzzy methods. Another difference between the IGDT method and other risk management tools is that the IGDT method guarantees that the objective function meets the predetermined objective when the realized prediction error falls within the maximum fluctuation range of the uncertainty variable.

[0106] The IGDT theory includes a risk averse strategy (RAS) and a risk seeker strategy (RSS). The former aims to maximize the avoidance of the impact of uncertainty on the solution result, and a robust model is constructed; the latter aims to seek the maximum benefit that can be obtained from the uncertainty risk, and an opportunity model is constructed. Since the optimization objective of this paper is to avoid the impact of wind and light load uncertainty on the result of energy storage configuration, the IGDT robust model, i.e. the RAS strategy, is selected for modeling.

[0107] The mathematical model of the optimization problem with uncertainty variables can be expressed as:

[0108]

[0109] In the formula, f is the objective function; X is the decision variable matrix; d is the uncertainty variable matrix; h and g are the equality constraints and inequality constraints, respectively. Let U be the set of d values, then the fluctuation of the actual value of the uncertainty variable d around its predicted value can be expressed as:

[0110]

[0111] In the formula, α is the bias coefficient of the uncertainty variable, and α>0.

[0112] Assuming that the optimal solution of the deterministic optimization model is f0 when d takes the predicted value, define f c as the maximum target value that the decision maker can accept after introducing uncertainty, then the robust optimization model defined by the IGDT method is:

[0113]

[0114] In formula (3), σ is the bias factor, i.e. the degree of deviation between the expected target value and the optimal solution of the deterministic model, and the greater σ is, the greater f c is, and the greater the risk aversion degree is, and the stronger the robustness is.

[0115] MPC overview

[0116] Intraday rolling optimization model

[0117] In order to react to renewable energy and load demand fluctuations in time, reduce the deviation between day-ahead planning and ultra-short-term scheduling, add the intraday rolling optimization link, take 15 min as the control time domain and 1 h as the prediction time domain, and establish the MPC rolling optimization model with the minimum deviation from the day-ahead economic optimization scheduling plan as the target. The main target is to use the latest information, calculate through the prediction model, and correct the day-ahead scheduling plan. The core of MPC is composed of the prediction model, rolling optimization and feedback correction.

[0118] Prediction model

[0119] The prediction model is used to describe the controlled object. The state of the controlled object in the future can be predicted through the model, and the output at t+1 time can be predicted according to the state quantity and control quantity of the system at t time. The basic linear state space expression is as follows:

[0120]

[0121] In the formula: X(t) represents the system state quantity of the controlled object; Y(t) represents the system output quantity; Δu(t) represents the control quantity or related input quantity of the object; ω(t) represents the system disturbance quantity; and A, B and C are system matrix, input matrix and output matrix respectively.

[0122] Rolling optimization

[0123] In actual operation, the output needs to be corrected due to the uncertainty of the system. At each sampling time, the optimal control sequence of the finite time domain starting from this time is solved according to the optimization performance index at this time. The target function of the tracking performance is as follows.

[0124]

[0125] In the formula: f obj is the target function of rolling optimization; Y p (t+i|t) and Y r (t+i|t) are the prediction value and reference value at t+i time at t time respectively; w v and w u are the system output deviation weight coefficient and control quantity weight coefficient.

[0126] Feedback correction

[0127] MPC can effectively deal with the uncertainty of the system through feedback correction. At each time, the control amount is compensated according to the performance deviation of the target function, so as to accurately predict the future output of the controlled object, and improve the robustness and control accuracy of the system. The feedback correction formula is as follows:

[0128]

[0129] In the formula: e k is the system control error; λ k is the error coefficient matrix; Y R is the actual output of the system at t time.

[0130] Establishment of double-layer model

[0131] The present application considers the influence of low-frequency components of uncertain factors on day-ahead scheduling of the comprehensive energy system, and the influence of high-frequency components of uncertain variables on day-in scheduling of the system, and further considers the interaction between multiple uncertain factors, and establishes a double-layer energy optimization scheduling framework, and the structure is as shown in Figure 3 .

[0132] The double-layer model proposed in the present application is divided into two parts. The upper layer is a day-ahead optimization scheduling layer. The low-frequency components of uncertain variables bring randomness and uncertainty to the system in day-ahead, IGDT is used for modeling uncertain factors, the randomness of uncertain variables is solved, the correlation between input multi-element random variables is handled by using nataf-svd technology, and the multi-element uncertainty is converted into single uncertainty of power shortage by combining 2pem-cornish-fisher method, and the solving efficiency of IGDT is improved.

[0133] The high-frequency components of uncertain variables exist in day-in, so that the day-in scheduling plan cannot meet the actual operation state in day-in, mpc is used for rolling tracking, and an efficient mpc solving algorithm is used for calculation, so that the model can be applied to more scenarios.

[0134] Simulation analysis

[0135] Example introduction

[0136] In order to verify the effectiveness of the model, in this section, a regional comprehensive energy system is taken as an example for simulation. The simulation platform is a computer with Intel four-core CPU of 2.5GHz and memory of 8.0GB. The parameter configuration of each unit is the same as that in the third chapter and will not be repeated here. This section is divided into two parts, 1) the influence of the correlation of uncertain factors on the operation of RIES is explained, and the effectiveness of the method for processing the correlation of variables is verified, 2) it is explained that the improved effective set method can definitely reduce the calculation overhead of MPC and increase the calculation efficiency of the model.

[0137] Influence of variable correlation on system operating characteristics

[0138] In order to illustrate the influence of input variable correlation on the operation of the system and the accuracy of the method proposed in the present application in processing the correlation, the correlation change between wind and light is analyzed in this section, the influence of the correlation coefficient change on the result accuracy is explored, and the first and second moments of the power shortage of the system under the condition of wind and light correlation coefficient increase are calculated by the method proposed in the present application, Figure 4 Figure 5 The power shortage and wind and light correlation coefficient change relationship curve is given.

[0139] It can be found by comparison that when the correlation increases, the expectation and standard deviation of the power shortage output will also change, but the fluctuation of the expectation is smaller, and the standard deviation changes approximately linearly, and Figure 5 It can be seen that the higher the correlation is, the stronger the synchronization of the output is, that is, there is a situation of simultaneous multiple or simultaneous less, which increases the volatility of the power shortage output, and the probability of low output and high output is greater, which is reflected in the increase of the standard deviation. In summary, it can be concluded that the correlation of the input variable has little influence on the power shortage, but it will significantly increase the volatility of the power shortage.

[0140] Performance evaluation of correlation processing

[0141] In order to evaluate the applicability of the method proposed in the present application in the scene of input variable containing correlation, the wind and light correlation is taken as the test, and the correlation processing method between other variables is similar. Based on the historical data of wind speed, solar irradiance, tidal speed and load demand of a certain region at different times, the present application obtains the wind and light variable data with correlation in each period through data analysis. Independent standard normal distribution variables are obtained by using Monte Carlo sampling technology, and input variable output samples satisfying the correlation condition are obtained by using Nataf inverse transformation, which are then used for MCSM calculation, and the results obtained by MCSM are used as the basis for judging the accuracy of the calculation results of the method proposed in the present application.

[0142] The method proposed in the present application obtains the power shortage of each order moment under the joint action of multiple uncertain variables by using the two-point method after the correlation processing of the input variable, and based on the data of each order moment, the probability density curve of the power shortage is fitted by using Cornish-Fisher series expansion, and the probability density function of the power shortage at each time is obtained, that is, 24 groups of probability density functions corresponding to 24 time points in a day. In time 5, the present method is compared with 5000 times of random simulation Monte Carlo method to test the effectiveness of the model, and the result is shown in Figure 6 .

[0143] From Figure 6It can be seen that the power shortage probability distribution curve fitted by 2PEM-Cornish-Fisher is almost the same as the Monte-Carlo result. Since the correlation between input variables is considered, the power shortage is more than the result in the third chapter when the wind turbine output is large at 5 o'clock because of the negative correlation between wind and photovoltaic. In order to comprehensively verify the effectiveness of the proposed method, two indicators are used to evaluate it from different angles, which are used to evaluate the accuracy of the proposed method in the statistical mathematical characteristics of the output variable and the calculation accuracy of the proposed method in the probability distribution characteristics of the output variable. As shown in the following formula:

[0144]

[0145]

[0146] In the formula: ξ γ is the relative error index and the root mean square of the sum of squares. and are the output variable result values obtained by the proposed method and MCSM, respectively; and are the values of the i-th point of the cumulative probability density function of the output variable obtained by the proposed method and MCSM, respectively. N is the number of points on the cumulative probability density function of the output variable.

[0147] The average value of the relative error index of different output variables and the maximum value are shown in Table 1. The relative error index of the expected value and the standard deviation of the output variable is less than 1%, which shows that the proposed method can accurately obtain the statistical digital characteristics of the output variable when the input variables have correlation.

[0148] Table 1

[0149]

[0150] Table 2 lists the average value and the maximum value of the root mean square of the sum of squares of different output variables, which shows that the proposed method can accurately obtain the probability distribution characteristics of the output variable.

[0151] Table 2

[0152]

[0153] From the above table, it can be seen that the correlation between variables has little effect on the result after being processed by the method proposed in this paper. The proposed method can accurately obtain the probability distribution curve of power shortage when the input variables have correlation.

[0154] In order to illustrate the high efficiency of the method, the method is compared with the Monte Carlo method (the sampling value is increased from 1000 to 10000) in terms of relative error index, and the result of 10000 times Monte Carlo sampling is taken as the benchmark value, and the result is as shown in the following table. Figure 7

[0155] As shown in the above table, the convergence speed of the method is much faster than that of the MCSM, the error is relatively large when the Monte Carlo sampling value is small, and the error gradually decreases with the increase of the sampling number, but the sampling time is exponentially increased with the increase of the sampling number, which can more reflect the high efficiency of the method. Figure 7

[0156] Analysis of calculation overhead of improved active set method

[0157] In order to verify the advantages of the improved active set algorithm of the model clock, the solving speed of the improved active set algorithm is compared with that of the Matlab-quadprog solver and the traditional active set algorithm in the case of different sizes of control objects. In the model, the control objects are 10 control objects such as gas turbine, gas boiler, electric to gas and electric heating, and the calculation overhead is as shown in the following table.

[0158] Table 3

[0159]

[0160] As shown in the above table, the improved ASM effectively improves the solving speed. Since the ASM converts the inequality constraint into an equality constraint and converts the original model into a KKT equation for solving, the solving speed is accelerated since it becomes a numerical method for solving. In the MASM, the continuity of the MPC equation at each time is considered, so that the value at the previous time can be taken as the initial value at this time, the iteration solving time is reduced, and the time consumed by the initial value from 0 is avoided.

[0161] As shown in table 3, because the initial point is reasonably set, the MASM effectively controls the solving time, and the control efficiency is obviously improved compared with other methods. With the increasing richness of regional integrated energy systems, such as roof photovoltaic systems and other distributed energy systems supported by the state, the objects to be controlled by the regional integrated energy system are increasing, which highlights the importance of algorithm overhead.

[0162] Dispatch simulation result analysis

[0163] In the day-ahead scheduling stage, the predicted value of the uncertainty variable is taken in the deterministic model calculation to perform deterministic optimization solving, and the optimal cost f0 of the whole day operation is calculated as 32864 yuan. The corresponding robust cost is f c ​​= f0 x (1 + 5%) = 34507 yuan, the calculation formula (3-23) gets epsilon = 0.87, and the corresponding scheduling result is as follows Figures 8 to 10 As shown in FIG. 1.

[0164] In combination Figure 8 Figure 10 It can be known that, since the correlation of the uncertainty factors is considered, the degree of belief is increased, because the uncertainty factors contained in the RIES have negative correlation, so that the extreme cases in the power shortage probability density curve obtained are less, and therefore the degree of belief can be appropriately amplified, because the given deviation factor is the same as that in the third chapter, the same power shortage is obtained under the same acceptable condition, so the same scheduling result is obtained, but the degree of belief is increased, and the acceptable degree of the scheduling result is increased.

[0165] In order to ensure the economic operation of the RIES under the influence of the uncertainty factors, the application establishes a double-layer heterogeneous energy optimization scheduling model based on IGDT-MPC. The two-point method and Cornish-Fisher series expansion are used to convert the multi-element uncertainty into a single power shortage uncertainty, effectively solving the problem that the IGDT model is difficult to accurately process multiple uncertainty factors at the same time, and aiming at the scene that the two-point method cannot be directly applied to the input variables with correlation, the Nataf transformation and SVD decomposition technology are introduced, so that the applicability of the 2PEM-Cornish method is more extensive. The MPC is used for rolling optimization in the day, and the deviation is corrected according to the actual situation in the day on the basis of the IGDT robust scheduling plan in the day, so as to improve the shortcomings of the IGDT open-loop control, so that the scheduling plan is more suitable for the actual operation state. And the advantages of the MPC day rolling optimization model are played, the control amount obtained at the last control time is taken as the initial value at the next control time, so as to greatly improve the calculation efficiency of the efficient set method and reduce the calculation cost. Therefore, the IGDT-MPC double-layer optimization scheduling model considering the correlation of variables provided by the application provides a new idea for the IGDT to process the scene with correlation of input variables.

[0166] The above is only the preferred specific embodiment of the application, but the protection scope of the application is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the application, which should be covered in the protection scope of the application. Therefore, the protection scope of the application should be subject to the protection scope of the claims.

Claims

1. A two-layer energy optimization scheduling method for a multi-energy complementary system, characterized in that, Includes the following steps: The correlation coefficient matrix of each variable in the normal space is obtained based on the Nataf transform; A sample matrix group is obtained based on the correlation coefficient matrix; The sample matrix group is used as input variables for 2PEM evaluation and series expansion fitting to obtain the probability density function; A quadratic programming model is constructed, and the model is solved based on the improved effective set method. A robust optimization model is constructed based on a risk avoidance strategy, and the impact of wind and solar load uncertainties on energy storage configuration results is avoided based on the robust optimization model. Construct an MPC rolling optimization model, and reduce the deviation based on the MPC rolling optimization model; A two-layer energy optimization scheduling framework is constructed based on the MPC rolling optimization model and the robust optimization model. Energy optimization scheduling is performed based on the aforementioned two-layer energy optimization scheduling framework; The process of obtaining the sample matrix group based on the correlation coefficient matrix includes: The correlation coefficient matrix is ​​decomposed into a sample matrix based on SVD decomposition, and a sample matrix group is constructed based on the sample matrix that is the same as the rank correlation matrix of the desired correlation coefficient matrix. The process of using the sample matrix as input variables for 2PEM evaluation and series expansion fitting includes: Based on the correlation coefficient matrix of the input variables, calculate the linear correlation matrix of the standard normal distribution random variables after Nataf transformation, and perform SVD decomposition on the linear correlation matrix of the standard normal distribution random variables to obtain the unitary matrix and the singular matrix. Calculate the sampled values ​​and weights in the normally independent space and the standard normal space, respectively; The sampled values ​​are used to form an evaluation matrix, and the sample matrix in the standard normal space is transformed into a sample matrix in the input variable space based on the inverse Nataf transform. The power deficit was evaluated using the 2PEM-Cornish method, and the original moments of each order were obtained and fitted to obtain the probability density function. The robust optimization model is as follows: Where f is the objective function; X is the decision variable matrix; d is the uncertainty variable matrix; h and g are the equality and inequality constraints, respectively; σ is the deviation factor, i.e., the degree of deviation between the expected target value and the optimal solution of the deterministic model. The larger σ is, the lower f is. c The larger the value, the greater the degree of risk aversion and the stronger the robustness.

2. The two-layer energy optimization scheduling method for multi-energy complementary systems according to claim 1, characterized in that, The process of obtaining the correlation coefficient matrix of each variable in the normal space based on the Nataf transform includes: Independent sequences are generated in the normal space, and the independent sequences are transformed into correlated sequences based on correlation control. The correlated sequences are then mapped to the non-normal space based on the Nataf transform.

3. The two-layer energy optimization scheduling method for multi-energy complementary systems according to claim 1, characterized in that, The process of building an MPC rolling optimization model includes: Build a predictive model; The prediction model is optimized based on the objective function; Feedback correction is performed on the performance deviation of the objective function.

4. The two-layer energy optimization scheduling method for a multi-energy complementary system according to claim 3, characterized in that, The prediction model is as follows: Where: X(t) represents the system state quantity of the controlled object; Y(t) represents the system output quantity; Δu(t) represents the control quantity or related input quantity of the object; ω(t) represents the system disturbance quantity; A, B, and C are the system matrix, input matrix, and output matrix, respectively.

5. The two-layer energy optimization scheduling method for a multi-energy complementary system according to claim 3, characterized in that, The objective function is: Among them, f obj The objective function for rolling optimization; Y p (t+i|t), Y r (t+i|t) represent the predicted value and reference value at time t+i, respectively; w v w u These are the system output deviation weighting coefficient and the control quantity weighting coefficient.

6. The two-layer energy optimization scheduling method for a multi-energy complementary system according to claim 3, characterized in that, The feedback correction method is as follows: Where e k For system control error; λ k Y is the error coefficient matrix; R This is the actual output of the system at time t.

Images