Power distribution network state estimation, regulation and control method considering multi-source data fusion

Abstract

The invention discloses a power distribution network state estimation, regulation and control method considering multi-source data fusion. The method comprises the following steps: considering the measurement precision and time scale of multi-source data and the synchronism of state estimation calculation, respectively preprocessing day-ahead and intra-day measurement data, and carrying out classification processing on the measurement data according to different characteristics of a power supply; day-ahead and intra-day data of distributed power supply output and corresponding irradiance, air temperature and wind speed are used as a total data set for training, and then a multi-layer output single-hidden-layer prediction nonlinear model considering system state variables is established; defining an upper limit value and a lower limit value of a state variable, rewriting a multi-layer output single hidden layer prediction nonlinear model into a linear model, and expressing the linear model in a matrix form; constructing an uncertain set to describe a fluctuation interval of a system state, and rewriting the fluctuation interval into a form consistent with an active-reactive coordination optimization model of the power distribution network; and constructing an active-reactive coordinated optimization relaxation model of the power distribution network based on the branch power flow, and optimizing each index of the system.

Description

Technical Field

[0001] The present invention relates to a distribution network state prediction and control method considering multi-source data fusion. Background Art

[0002] With the sustainable development of society, building a new power system with clean energy as the main body is the future development trend. The power industry must actively change the way of energy production and consumption, optimize the energy structure, and increase the development of renewable energy. When a high proportion of distributed power sources such as photovoltaics and wind power are connected to the distribution network, due to their inherent intermittent and volatile characteristics, it increases the difficulty of output prediction and optimal scheduling.

[0003] Specifically, the high proportion of distributed power supply access will lead to the following problems: 1) The output of distributed power supply is easily affected by weather and human factors, and the large fluctuation increases the difficulty of prediction; 2) Insufficient measurement level and insufficient measurement accuracy lead to missing historical data or increased errors; 3) Traditional distribution network models have slow solution speed and poor ability to handle high penetration scenarios; 3) It is difficult to summarize a probability model from massive data, and it is necessary to analyze and model based on a large amount of historical sampling data. However, due to the insufficient measurement level of my country's distribution network, there are often problems such as missing data and insufficient measurement accuracy. The true value of the parameters is uncertain, and it is difficult to directly establish an accurate probability model. Therefore, it is necessary to design a new method to meet actual needs. Summary of the invention

[0004] The present invention takes into account the uncertain factors in the distribution network under high penetration conditions, classifies the quantity measurements, and constructs an uncertainty set based on the original deterministic optimization model of the distribution network, so that the scheduling scheme has higher adaptability.

[0005] The present invention is implemented by the following technical means: a distribution network state prediction and control method considering multi-source data fusion, comprising the following steps:

[0006] Step 1: Considering the measurement accuracy, time scale and synchronization of state estimation calculation of multi-source data, pre-process the measured data of the day before and classify the measured data according to the different characteristics of the power supply;

[0007] Step 2: Use the day-ahead and intraday data of distributed power output and the corresponding natural statistical data of irradiance, temperature and wind speed as the training data set, and then establish a multi-layer output single hidden layer prediction nonlinear model considering the system state variables;

[0008] Step 3: Define the upper and lower limits of the state variables, rewrite the single hidden layer prediction nonlinear model of multi-layer output into a linear model, and express it in the form of a matrix;

[0009] Step 4: Based on the above prediction model, obtain the upper and lower bounds of the prediction results, construct an uncertainty set to describe the fluctuation range of the system state, and rewrite it into a form consistent with the active-reactive coordinated optimization model of the distribution network;

[0010] Step 5: Construct a distribution network active-reactive coordinated optimization relaxation model based on branch power flows, define the boundaries of control variables and state variables, and then optimize the system indicators.

[0011] In step 1, the measured value information is classified and processed according to the different characteristics of the power supply. The directly measured data needs to be processed. The basis for classification is the continuity of the measured data. The power generation system without inertia is more affected by natural factors such as weather, and the data gap is more serious in extreme cases. When the power generation system has inertia, the requirements for measurement accuracy can be reduced. At the same time, data with a larger sampling interval can be appropriately selected during data processing. The power generation system includes: non-inertia units represented by photovoltaic units, and units with inertia represented by wind turbines.

[0012] For non-inertia units represented by photovoltaic units: the sampling interval is required to be small and the accuracy is required to be high. The predicted value at time t is expressed as:

[0013]

[0014] Where M is the number of measurement sequences before and after time t; T is the sampling period of the measurement; x t+kT is the measured value at time t+kT.

[0015] For wind turbines, which have inertia, the sampling interval and measurement error are allowed to be slightly larger, and the predicted value at time t is expressed as:

[0016]

[0017] In the formula, x t-1 is the measured value of the sampling point before time t; is the sampling time before and after time t; It is the measured value at the sampling moment after time t and the predicted value at the previous moment.

[0018] In step 2, a multi-output single hidden layer nonlinear model considering the system state variables is established:

[0019] 1) Assume a set of original data samples {x i ,y i ,v i}, where x i =[x i1 ,x i2 ,...,x ir ]T is the input vector of the model, including the historical data of distributed power sources, including irradiance, temperature and wind speed, y i =[y i1 ,y i2 ,...,y ir ] T Represents the predicted target vector, including the output range of the power supply at the next moment, i∈N DG is the number of power supplies, r is the dimension of historical data; in order to enhance the association between the output variable and the state variable or reflect the coupling relationship between the two, the bias of the hidden layer node is set to the state variable v i Multiply by the coefficient b j In the form of, the output of a single hidden layer neural network with τ hidden nodes is expressed as:

[0020]

[0021] In the formula, λ j The weight value of the model input vector x; ω j is the target vector y i The weight of j is the bias of the jth hidden layer node; g(·) is the activation function tanh(·) based on the hyperbolic tangent; the goal of neural network learning is to make the actual output f j With target y j The error between them is the smallest, that is, f j =y j .

[0022] 2) In the distribution network state prediction, the branch active power P, reactive power Q, branch current amplitude I, node voltage amplitude U and phase angle θ are all pseudo measurements described in interval form; the selection of interval values ​​is to set a fluctuation of ±α% based on the corresponding reference value; the possible operating state of the estimated system is determined by the interval information of the measurement and a set of nonlinear mapping relationships, and its mathematical relationship can be expressed as:

[0023]

[0024] Where E(·) is the interval set of the state variables to be determined; h(·) is the nonlinear mapping function between the state variables and the model input; and Z(·) is the pseudo-measurement uncertainty set of the input.

[0025] In step 3, the upper and lower limits of the state variables are defined, and the single hidden layer prediction nonlinear model of multi-layer output is rewritten into a linear model as follows:

[0026] Define the upper and lower limits of the state variables:

[0027]

[0028] Substituting the above formula into the multi-output single hidden layer nonlinear model considering the system state variables, we can get:

[0029]

[0030] In the formula, are the upper and lower limits of the state variable. When the upper and lower bounds of the state variable can be expressed, v i It can be regarded as a set of constant variables involved in the calculation.

[0031] The construction of the uncertainty set in step 4 includes the following steps:

[0032] The interval model is used to describe the uncertainty factors in the active distribution network system. The load active and reactive power considering the uncertainty can be expressed as:

[0033]

[0034] In the formula, It is the uncertain set form of distributed generation output; is the predicted value of output; G is the park

[0035] gen

[0036] Game results on the power generation side; represents the deviation between the actual value and the predicted value; ψ1 and ψ2 are coefficients representing the fluctuation range;

[0037] Considering that the calculation form is a matrix expression, the uncertainty set is rewritten as follows:

[0038]

[0039] In step 5, the construction of the active-reactive coordinated optimization relaxation model of the distribution network is constructed based on the branch flow:

[0040]

[0041] In the formula, P = [P ij ], Q = [Q ij ]、p=[p i ], q=[q i ]、R=diag[R ij ] n×n 、X=diag[X ij ] n×n , The first row of the matrix A0 is the transformer node that needs to be removed, so [0 n ,I]A0-I=A, I is the n-dimensional unit matrix, 0 n is a zero vector of n rows.

[0042] Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on branch power flow:

[0043] Voltage V is both a state variable and a target variable. l is a state variable. Its upper and lower bounds are expressed as follows:

[0044]

[0045] Where D X+ , H + Represents the matrix D X , the elements of H that are not less than zero; D X- , H _ represents elements less than zero; P g Represents the power matrix of the distribution network, P g + , P g - Represents the maximum and minimum values ​​of the power output; P L Represents the load matrix of each node in the distribution network; V + =V(p + ),q + =q(p + ), l max , l min Represents the maximum and minimum values ​​of l(p).

[0046] Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on the branch flow: The model sets three indicators, which are expressed as:

[0047]

[0048] In the formula, G ij is the conductance of branch ij; U i,t and U j,t is the voltage amplitude of nodes i and j at time t; N represents the number of common nodes; U i,t and U i,ref Refers to the voltage level of node i and its reference standard; N g represents the number of DGs; Refers to the output power of the DG of node i at time t.

[0049] Normalize the above objectives to obtain a comprehensive objective function:

[0050]

[0051] Where f1 * and They are the network loss and voltage deviation before optimization; is the predicted output of DG; α, β, γ are the proportions of each target, determined by the hierarchical analysis method, α+β+γ=1; let the scales of f1, f2, f3 in the hierarchical analysis be 3, 1, 5 respectively, and the calculated weight coefficients α, β, γ are 0.258, 0.637, 0.105 respectively.

[0052] Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on branch power flow:

[0053] The state variable can be selected as l or V. Considering that the relaxation model is adopted and the voltage optimization target is set in the model, l is set as the state variable and expressed as:

[0054]

[0055] The derivation of the boundary value can be expressed as:

[0056]

[0057] In the formula, J ij+ , J ij- For J ij The matrix composed of non-negative elements and the matrix composed of negative elements in δ ij+ , δ ij - is δ ij The matrix consists of non-negative and negative elements in . However, according to equations (15) and (16), we know that l ij,min Negative values ​​are possible, but Therefore, when the actual simulation is running, ij The lower bound of should be max{0, l ij,min}; In addition, since the Hessian matrix H ij Two eigenvalues ​​of are strictly positive and one is zero, so it is positive semidefinite, and therefore the bound of l is convex.

[0058] Through learning and processing of historical data prediction and real-time measurement data, interval state estimation is carried out based on the characteristics of various elements within the distribution network, and an active-reactive coordinated optimization model is established. According to the characteristics of the model, the relaxation technology combined with the KKT condition is adopted, and the solution is performed through the CPLEX solving package in Matlab.

[0059] The beneficial effects of the present invention after adopting the above technical solution are:

[0060] 1. Considering the measurement accuracy, time scale and synchronization of state estimation calculation of multi-source data, the measured data of the day before and within the day are pre-processed and classified according to the different characteristics of the power supply, which greatly improves the availability of data;

[0061] 2. The single hidden layer nonlinear model with multi-layer output considering the state variables enhances the correlation between the output variables and the state variables or reflects the coupling relationship between the two, greatly improving the accuracy and practicality of the prediction model;

[0062] 3. Define the upper and lower limits of the state variables, and rewrite the single hidden layer prediction nonlinear model with multi-layer output into a linear model, which improves the solution efficiency and adaptability of the model;

[0063] 4. By constructing an uncertain set to describe the fluctuation range of input variables and state variables in the active-reactive coordinated optimization model, the deterministic optimization model is rewritten as an uncertain optimization model. After optimization, a more applicable scheduling scheme can be obtained; at the same time, the non-convex model of the network is improved, and the relaxation boundary of the branch current is defined based on the rated operating point to ensure the flow constraint while improving the solution efficiency of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0065] Figure 1 The figure shows a workflow diagram of a distribution network state estimation and control method considering multi-source data fusion.

[0066] Figure 2 Shown is a flow chart for distribution network state estimation and control modeling considering multi-source data fusion.

[0067] Figure 3 Shown is a schematic diagram of the structure of the output range prediction model for wind and solar power generation systems.

[0068] Figure 4 Shown is a schematic diagram of the output range of a wind and solar power generation system.

[0069] Figure 5 The figure shows a schematic diagram of the voltage fluctuation range of the system nodes.

[0070] Figure 6 Shown is the equivalent wiring diagram of the radial distribution network. DETAILED DESCRIPTION

[0071] The following will be combined with the attached embodiment of the present invention Figure 1-6 , clearly and completely describe the technical solutions in the embodiments of the present invention.

[0072] The above-mentioned distribution network state estimation and control method considering multi-source data fusion includes the following process:

[0073] Figure 2 This is a flow chart for modeling distribution network state prediction and regulation considering multi-source data fusion, including:

[0074] ① Processing of the measurement data within the previous day;

[0075] ② Establishment of a single hidden layer nonlinear model with multi-layer output considering system state variables;

[0076] ③Rewrite the single hidden layer prediction nonlinear model with multi-layer output into a linear model;

[0077] ④Transform the deterministic optimization model of the distribution network into an uncertain optimization model and improve the non-convex model of the network.

[0078] In step 1, the measured value information is classified and processed according to the different characteristics of the power supply. The directly measured data cannot be directly applied and needs to be processed. First, the measurement accuracy, time scale and synchronization of state estimation calculation of multi-source data are considered, and the measured data of the day before are preprocessed, involving data cleaning (processing missing values, outliers, and duplicate data), data integration (merging different data sources and resolving conflicts and redundancies), data transformation (normalization, standardization, and discretization to adapt to analysis needs), and data reduction (reducing data complexity by reducing dimensions or quantity).

[0079] Then the data is classified and processed. The basis for classification is the continuity of the measured data. The power generation system without inertia is more affected by natural factors such as weather, and the data fault is more serious in extreme cases. When the power generation system has a certain inertia, the requirements for measurement accuracy can be reduced. At the same time, data with a larger sampling interval can be appropriately selected during data processing, including the following methods:

[0080] 1) For non-inertia units represented by photovoltaic units: the sampling interval is required to be small and the accuracy is required to be high. The predicted value at time t can be expressed as:

[0081]

[0082] Where M is the number of measurement sequences before and after time t; T is the sampling period of the measurement; x t+kT is the measured value at time t+kT.

[0083] 2) For wind turbines with a certain inertia: the sampling interval and measurement error are allowed to be slightly larger, and the predicted value at time t can be expressed as:

[0084]

[0085] In the formula, x t-1 is the measured value of the sampling point before time t; is the sampling time before and after time t; It is the measured value at the sampling moment after time t and the predicted value at the previous moment.

[0086] In step 2, the construction of a single hidden layer prediction nonlinear model considering the multi-layer output of the system state variables is as follows:

[0087] Assume that a set of original data samples {x i ,y i ,v i}, where x i =[x i1 ,x i2 ,...,x ir ] T is the input vector of the model, including the historical data of distributed power sources (irradiance, temperature, wind speed, etc.), y i =[y i1 ,y i2 ,...,y ir ] T Represents the predicted target vector, including the output range of the power supply at the next moment, i∈N DG is the number of power sources, and r is the dimension of historical data. It can be seen that the output of a single hidden layer neural network with a hidden node can be expressed as:

[0088]

[0089] In the formula, λ j Input vector x to the model i The weight value of j is the target vector y i The weight of j v i is the bias of the jth hidden layer node; g(·) is the activation function tanh(·) based on hyperbolic tangent. Compared with the traditional sigmoid(·) activation function, tanh(·) converges faster.

[0090] It should be noted that the prediction in step 2 is based on the fitting of historical data. In order to enhance the correlation between the output variable and the state variable or reflect the coupling relationship between the two, the bias of the hidden layer node is set in the form of the state variable multiplied by the coefficient.

[0091] Figure 3 The figure shows a schematic diagram of the structure of the wind and solar power generation system output range prediction model, in which the input information includes historical irradiance parameters, historical environmental parameters, historical temperature parameters, and historical wind speed parameters. The upper and lower bounds of the single hidden layer output are obtained through the middle hidden layer of the model, and the intraday measured data is classified and processed to finally obtain the power output prediction range.

[0092] The goal of neural network learning is to make the actual output fj With target y j The error between them should be as small as possible, that is, the following formula should be established as much as possible:

[0093]

[0094] Even if the input vector x is given i The weight value and bias coefficient b j , the single hidden layer model corresponding to equation (1) and equation (2) is not linear, so these τ equations have no solution.

[0095] In the distribution network state prediction, the branch active power P, reactive power Q, branch current amplitude I, node voltage amplitude U and phase angle θ are all pseudo measurements described in interval form; the interval value is selected by setting a fluctuation of ±α% based on the corresponding reference value. In summary, the possible operating state of the estimated system can be determined by the interval information of the measurement and a set of nonlinear mapping relationships, and its mathematical relationship can be expressed as:

[0096]

[0097] Where E(·) is the interval set of the state variables to be determined; h(·) is the nonlinear mapping function between the state variables and the model input; and Z(·) is the pseudo-measurement uncertainty set of the input.

[0098] In step 3, the single hidden layer prediction nonlinear model of multi-layer output is rewritten into a linear model.

[0099] Define the upper and lower limits of the state variables:

[0100]

[0101] Substituting formula (5) into formula (3), we can obtain:

[0102]

[0103] In the formula, are the upper and lower limits of the state variable. When the upper and lower bounds of the state variable can be expressed, v i It can be regarded as a set of constant variables involved in the calculation.

[0104] In step 4, the construction of the uncertainty set includes the following steps:

[0105] The output of distributed power sources often presents the characteristics of randomness and volatility, especially when they are connected at a high ratio, prediction becomes particularly difficult. Most current studies use probability models, fuzzy models and interval models to describe this uncertainty. From a mathematical and physical point of view, the applicability of probability models is higher than that of interval models. Problems such as missing data, insufficient measurement accuracy and uncertain true values ​​of parameters make it almost impossible to summarize a probability model with strong applicability. Fundamentally, this is because uncertain factors do not have random characteristics, that is, although a large amount of historical sampling data has provided many results, there will still be unknown results or operating states that have not been considered. This is an uncertain factor rather than a random factor, so its probability distribution is always difficult to determine.

[0106] The interval model is somewhat conservative, but compared with the probability model, it does not need to obtain the prior distribution state of the network parameters, but only needs to obtain the upper and lower bounds of the state variables, which makes it more adaptable in practical applications.

[0107] The load active and reactive power considering uncertainty can be expressed as:

[0108]

[0109] In the formula, It is the uncertain set form of distributed generation output; is the predicted output value of distributed generation; represents the deviation between the actual value and the predicted value, where ψ1 and ψ2 are coefficients that characterize the fluctuation range.

[0110] Considering that all operations on variables are in matrix form, the above uncertainty set is rewritten into matrix form:

[0111]

[0112] In the formula, X is the uncertain parameter set of the input variables, that is, the uncertain set of active and reactive power of distributed generation composed of multiple groups of scenarios; are the upper and lower limits of fluctuation; Z′ is the uncertain parameter set of state variables; are the upper and lower limits of the state variable.

[0113] Based on the system state model established above, combined with the measurement data of the previous day, the model can be solved to obtain the following Figure 4 (a) is a schematic diagram of the output range of a photovoltaic power generation system, and (b) is a schematic diagram of the output range of a wind power generation system.

[0114] Figure 5 The figure shows a schematic diagram of the voltage fluctuation range of the system nodes.

[0115] In step 5, based on Figure 6From the single-phase equivalent circuit diagram of the distribution network shown in the figure, it can be seen that the relationship between node voltage and power is:

[0116]

[0117] &

[0118] Where U j is the voltage vector of node i; Z ij =R ij+ JXj ij is the impedance of branch ij; P ij and Q ij is the active and reactive power flowing on feeder ij. Based on the above distribution line equivalent model, it can be seen that U j , I ij is the state variable, and the input variable is P ij , Q ij The state variable v i It's not about x i This results in the single hidden layer model in step 1 being non-convex, which increases the difficulty of solving and prolongs the prediction time.

[0119] Due to the existence of power flow constraints, the coordinated optimization model between multiple time sections is essentially a type of mixed integer nonlinear programming problem. Relaxation techniques represented by second-order cone relaxation can transform the non-convex original problem into a convex problem, but its expression is still nonlinear. Therefore, the present invention rewrites the power balance equation of the distribution network into the following form based on the convex inner approximation method:

[0120]

[0121] Where, P = [P ij ], Q = [Q ij ]、p=[P g,i ], q=[Q g,i ]、R=diag[R ij ] n×n 、X=diag[X ij ] n×n , The first row of the matrix A0 is the transformer node that needs to be removed, so [0 n ,I]A0-I=A, I is the n-dimensional unit matrix, 0 n is a zero vector of n rows.

[0122] Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on branch power flow:

[0123] The directly predicted quantity in the present invention is the output of the distributed power source. State variables such as node voltage and branch current belong to indirect prediction quantities and do not satisfy the equation shown in formula (8). Therefore, the state variables can only be determined by calculating the upper and lower limits to determine their fluctuation range. The state variable can be selected as l or V. Considering that a relaxation model is adopted and the weight of the voltage index in the model is relatively high, l is set as the state variable and expressed as:

[0124]

[0125] The derivation of the boundary value can be expressed as:

[0126]

[0127]

[0128] In the formula, J ij+ , J ij- For J ij The matrix composed of non-negative elements and the matrix composed of negative elements in δ ij+ ,δ ij- is δ ij is a matrix composed of non-negative and negative elements in. However, according to formula (3.58), l ij,min Negative values ​​are possible, but Therefore, when the actual simulation is running, ij The lower bound of should be max{0, l ij,min}; In addition, since the Hessian matrix H ij Two eigenvalues ​​of are strictly positive and one is zero, so it is positive semidefinite, and therefore the bound of l is convex.

[0129] After defining the boundaries of the branch current, the upper and lower bounds of the system input variables, state variables, and output variables can be expressed as follows:

[0130]

[0131] Where D X+ , H + Represents the matrix D X , the elements of H that are not less than zero; D X- , H _ represents elements less than zero; P g Represents the power matrix of the distribution network, P g + , P g - Represents the maximum and minimum values ​​of the power output; P L Represents the load matrix of each node in the distribution network; V +=V(p + ),q + =q(p + ), l max , l min Represents the maximum and minimum values ​​of l(p). p =2C T RC, M q =2C T XC, D X =(IA) -1 AX, C=(IA) -1 , D R =(IA) -1 AR、H=C T (2(RD R +XD X )+Z 2 ). In this model, P, Q, and V are linearly related, but this relationship depends on the accurate representation of the branch current boundaries.

[0132] The model sets three indicators:

[0133] 1) Minimum network loss:

[0134]

[0135] 2) Minimum voltage deviation:

[0136]

[0137] 3) Distributed power generation has the largest output:

[0138]

[0139] In the formula, in the formula: G ij is the conductance of branch ij; U i,t and U j,t is the voltage amplitude of nodes i and j at time t; N represents the number of common nodes; U i,t and U i,ref Refers to the voltage level of node i and its reference standard; N g represents the number of DGs; Refers to the output power of the DG of node i at time t.

[0140] Normalize the above objectives to obtain a comprehensive objective function:

[0141]

[0142] Where f1 * and They are the network loss and voltage deviation before optimization; is the predicted output of DG; α, β, γ are the proportions of each target, determined by the hierarchical analysis method, α+β+γ=1; let the scales of f1, f2, f3 in the hierarchical analysis be 3, 1, 5 respectively, and the calculated weight coefficients α, β, γ are 0.258, 0.637, 0.105 respectively.

[0143] The present invention performs learning processing on historical data prediction and real-time measurement data, performs interval state estimation based on the characteristics of various elements within the distribution network and establishes an active-reactive coordinated optimization model, adopts a relaxation technique combined with KKT conditions according to the characteristics of the model, and solves it through the CPLEX solving package in Matlab.

[0144] Table 1 compares the indicators of the traditional distribution network optimization method (method 1) and the uncertainty distribution network optimization method (method 2) considering system state estimation in this paper. The calculation speed of this method is comparable to that of the linear optimal power flow (OPF) and has a higher solution efficiency; the utilization rate of new energy is not significantly different from that of the traditional deterministic optimization method.

[0145] The power loss of the system was reduced by 8.767%; the average voltage deviation was reduced by 19.37%.

[0146] Table 1 Comparison of various indicators of the two methods

[0147] method Voltage deviation average value Power loss / (kW·h) DG Utilization Method 1 0.6723 3350.2 31.53％ Method 2 0.5421 3056.5 28.41％

[0148] The results show that the method proposed in this invention can accurately predict the output range of distributed power sources and the system voltage variation range, and obtain better distribution network optimization effect.

Claims

1. A distribution network state estimation and control method considering multi-source data fusion, characterized in that: The following steps are involved: Step 1: Considering the measurement accuracy, time scale and synchronization of state estimation calculation of multi-source data, pre-process the measured data of the day before and classify the measured data according to the different characteristics of the power supply; Step 2: Use the day-ahead and intraday data of distributed power output and the corresponding natural statistical data of irradiance, temperature and wind speed as the training data set, and then establish a multi-layer output single hidden layer prediction nonlinear model considering the system state variables; Step 3: Define the upper and lower limits of the state variables, rewrite the single hidden layer prediction nonlinear model of multi-layer output into a linear model, and express it in the form of a matrix; Step 4: Based on the above prediction model, the upper and lower bounds of the prediction results are obtained, and an uncertainty set is constructed to describe the fluctuation range of the system state, and it is rewritten into a form consistent with the active-reactive coordinated optimization model of the distribution network; Step 5: Construct a distribution network active-reactive coordinated optimization relaxation model based on branch power flows, define the boundaries of control variables and state variables, and then optimize the system indicators.

2. According to claim 1, a distribution network state estimation and control method considering multi-source data fusion is characterized by: In step 1, the measured value information is classified and processed according to the different characteristics of the power supply. The directly measured data needs to be processed. The basis for classification is the continuity of the measured data. The power generation system without inertia is more affected by natural factors such as weather, and the data gap is more serious in extreme cases. When the power generation system has inertia, the requirements for measurement accuracy can be reduced. At the same time, data with a larger sampling interval can be appropriately selected during data processing. The power generation system includes: non-inertia units represented by photovoltaic units, and units with inertia represented by wind turbines.

3. According to claim 2, a distribution network state estimation and control method considering multi-source data fusion is characterized in that: For non-inertia units represented by photovoltaic units: the sampling interval is required to be small and the accuracy is required to be high. The predicted value at time t is expressed as: Where M is the number of measurement sequences before and after time t; T is the sampling period of the measurement; x t+kT is the measured value at time t+kT.

4. The method for predicting and controlling the state of a distribution network considering multi-source data fusion according to claim 2 is characterized in that: For wind turbines, which have inertia, the sampling interval and measurement error are allowed to be slightly larger, and the predicted value at time t is expressed as: In the formula, x t-1 is the measured value of the sampling point before time t; is the sampling time before and after time t; It is the measured value at the sampling moment after time t and the predicted value at the previous moment.

5. The method for predicting and controlling the state of a distribution network considering multi-source data fusion according to claim 1 is characterized in that: In step 2, a multi-output single hidden layer nonlinear model considering the system state variables is established: 1) Assume a set of original data samples {x i ,y i ,v i }, where x i =[x i1 ,x i2 ,...,x ir ] T is the input vector of the model, including the historical data of distributed power sources, including irradiance, temperature and wind speed, y i =[y i1 ,y i2 ,...,y ir ] T Represents the predicted target vector, including the output range of the power supply at the next moment, i∈N DG is the number of power supplies, r is the dimension of historical data; in order to enhance the association between the output variable and the state variable or reflect the coupling relationship between the two, the bias of the hidden layer node is set to the state variable v i Multiply by the coefficient b j In the form of, the output of a single hidden layer neural network with τ hidden nodes is expressed as: In the formula, λ j The weight value of the model input vector x; ω j is the target vector y i The weight of j is the bias of the jth hidden layer node; g(·) is the activation function tanh(·) based on the hyperbolic tangent; the goal of neural network learning is to make the actual output f j With target y j The error between them is the smallest, that is, f j =y j . 2) In the distribution network state prediction, the branch active power P, reactive power Q, branch current amplitude I, node voltage amplitude U and phase angle θ are all pseudo-measurements described in interval form; The selection of the interval value is to set a fluctuation of ±α% on the basis of the corresponding reference value; the possible operating state of the estimation system is determined by the interval information of the measurement and a set of nonlinear mapping relationships, and its mathematical relationship can be expressed as: Where E(·) is the interval set of the state variables to be determined; h(·) is the nonlinear mapping function between the state variables and the model input; and Z(·) is the pseudo-measurement uncertainty set of the input.

6. A distribution network state estimation and control method considering multi-source data fusion according to claim 1, characterized in that: In step 3, the upper and lower limits of the state variables are defined, and the single hidden layer prediction nonlinear model of multi-layer output is rewritten into a linear model as follows: Define the upper and lower limits of the state variables: Substituting the above formula into the multi-output single hidden layer nonlinear model considering the system state variables, we can get: In the formula, are the upper and lower limits of the state variable. When the upper and lower bounds of the state variable can be expressed, v i It can be regarded as a set of constant variables involved in the calculation.

7. A distribution network state estimation and control method considering multi-source data fusion according to claim 1, characterized in that: The construction of the uncertainty set in step 4 includes the following steps: The interval model is used to describe the uncertainty factors in the active distribution network system. The load active and reactive power considering the uncertainty can be expressed as: In the formula, It is the uncertain set form of distributed generation output; is the predicted value of output; G is the game result of the gen power generation side of the park; represents the deviation between the actual value and the predicted value; ψ1 and ψ2 are coefficients representing the fluctuation range; Considering that the calculation form is a matrix expression, the uncertainty set is rewritten as follows:

8. A distribution network state prediction and control method considering multi-source data fusion according to claim 1, characterized in that: In step 5, the construction of the active-reactive coordinated optimization relaxation model of the distribution network is constructed based on the branch flow: Where, P = [P ij ], Q = [Q ij ]、p=[p i ], q=[q i ]、R=diag[R ij ] n×n 、X=diag[X ij ] n×n , The first row of the matrix A0 is the transformer node that needs to be removed, so [0 n ,I]A0-I=A, I is the n-dimensional unit matrix, 0 n is a zero vector of n rows.

9. A distribution network state prediction and control method considering multi-source data fusion according to claim 1, characterized in that: Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on branch power flow: Voltage V is both a state variable and a target variable. l is a state variable. Its upper and lower bounds are expressed as follows: Where D X+ , H + Represents the matrix D X , the elements of H that are not less than zero; D X- , H _ represents elements less than zero; P g Represents the power matrix of the distribution network, P g + , P g - Represents the maximum and minimum values ​​of the power output; P L Represents the load matrix of each node in the distribution network; V + =V(p + ),q + =q(p + ), l max , l min Represents the maximum and minimum values ​​of l(p).

10. The method for predicting and controlling the state of a distribution network considering multi-source data fusion according to claim 1, characterized in that: Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on the branch flow: The model sets three indicators, which are expressed as: In the formula, G ij is the conductance of branch ij; U i,t and U j,t is the voltage amplitude of nodes i and j at time t; N represents the number of common nodes; U i,t and U i,ref Refers to the voltage level of node i and its reference standard; N g represents the number of DGs; Refers to the output power of the DG of node i at time t. Normalize the above objectives to obtain a comprehensive objective function: Where f1 * With f2 * They are the network loss and voltage deviation before optimization; f3 * is the predicted output of DG; α, β, γ are the proportions of each target, determined by the hierarchical analysis method, α+β+γ=1; let the scales of f1, f2, f3 in the hierarchical analysis be 3, 1, 5 respectively, and the calculated weight coefficients α, β, γ are 0.258, 0.637, 0.105 respectively.

11. A distribution network state prediction and control method considering multi-source data fusion according to claim 1, characterized in that: Step 5: Construct the active-reactive coordinated optimization relaxation model of the distribution network based on branch power flow: The state variable can be l or V. Considering that the relaxation model is adopted and the voltage optimization target is set in the model, l is set as the state variable and expressed as: The derivation of the boundary value can be expressed as: In the formula, J ij+ , J ij- For J ij The matrix composed of non-negative elements and the matrix composed of negative elements in δ ij+ ,δ ij- is δ ij The matrix consists of non-negative and negative elements in . However, according to equations (15) and (16), we can know that l ij,min Negative values ​​are possible, but Therefore, when the actual simulation is running, ij The lower bound of should be max{0, l ij,min }; In addition, since the Hessian matrix H ij Two eigenvalues ​​of are strictly positive and one is zero, so it is positive semidefinite, and therefore the bound of l is convex.

12. A distribution network state prediction and control method considering multi-source data fusion according to claim 1, characterized in that: Through learning and processing of historical data prediction and real-time measurement data, interval state estimation is carried out based on the characteristics of various elements within the distribution network, and an active-reactive coordinated optimization model is established. According to the characteristics of the model, the relaxation technology combined with the KKT condition is adopted, and the solution is performed through the CPLEX solving package in Matlab.

Images