Distribution network equivalent modeling and calling method considering distributed photovoltaic uncertainty

Abstract

The invention discloses a power distribution network equivalent modeling and model calling method considering distributed photovoltaic uncertainty. The method comprises the following steps: firstly, collecting a photovoltaic output historical prediction value and an actual value in a power distribution network area, and calculating a prediction error; clustering prediction errors according to prediction time and prediction values, and calculating probability density distribution of each clustering result; establishing an equivalent model of the power distribution network to determine parameters to be identified, and then collecting electrical measurement values of the power distribution network to identify the parameters of the equivalent model; and finally, calling the equivalent model to predict distributed photovoltaic output in the power distribution network.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and more specifically, relates to a distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaics. Background Art

[0002] Power system analysis is the basis of power system operation and control. However, the topology of the distribution network is complex and large in scale. It is inefficient and time-consuming to directly model and simulate the distribution network in detail. Therefore, it is necessary to establish an equivalent model of the distribution network to reduce the complexity of power system analysis and calculation. In addition, the high penetration photovoltaic characteristics have changed the characteristics of the distribution network. The uncertainty of photovoltaic power generation makes the distribution network more random and the impact on the power system is more complex. Therefore, the analysis of the power system cannot ignore the uncertainty of photovoltaic power generation. In summary, it is necessary to invent a distribution network probabilistic equivalent modeling method that can accurately characterize the output characteristics of the distribution network and the uncertainty of distributed photovoltaic power generation.

[0003] The distribution network equivalent model is divided into two categories: deterministic equivalent model and probabilistic equivalent model. The deterministic equivalent model equates the distribution network to a load or a load in parallel with a photovoltaic module. The former is difficult to describe the actual power characteristics of photovoltaic power generation, while the latter distinguishes between photovoltaic modules and loads and can describe photovoltaic power characteristics, but can usually only establish a single-point expected value of the distribution network state under a certain working state, ignoring the uncertainty of photovoltaic power generation, resulting in distribution network equivalent errors caused by the uncertainty of photovoltaic power generation.

[0004] The probability equivalence of the distribution network constructs the relationship between environmental factors and photovoltaic output in a probabilistic way, and then establishes a probability equivalence model of the distribution network, which provides more abundant and accurate uncertainty information, is conducive to reducing the equivalence error caused by photovoltaic uncertainty, and provides key support for power system analysis and operation control. However, there are many factors that affect the uncertainty of photovoltaic output, such as distributed photovoltaic off-grid, the relationship between environmental factors and power fluctuates near the theoretical curve, and the uncertainty caused by geographical location. It is difficult to consider all the factors that cause uncertainty. In the existing distribution network equivalent modeling process, only environmental factors are used to measure the uncertainty of photovoltaic power generation, which has low accuracy and affects the accuracy of the equivalent model. In addition, the parameter identification of the distribution network probability model often only considers the consistency of the model flow before and after the equivalent, ignoring the consistency of sensitivity, which affects the accuracy of the equivalent model. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a distribution network equivalent modeling and model calling method taking into account the uncertainty of distributed photovoltaics, describe the photovoltaic uncertainty through the distribution of photovoltaic output prediction errors, and then establish an equivalent model of the distribution network connected to distributed photovoltaics.

[0006] To achieve the above-mentioned purpose of the invention, the present invention provides a distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaics, characterized in that it includes the following steps:

[0007] (1) Collect historical photovoltaic data;

[0008] The actual value and corresponding predicted value of photovoltaic active power of distributed photovoltaic power stations at each historical moment are collected over a period of time, which is recorded as photovoltaic data set P pv ;

[0009]

[0010] Among them, P pv_act is a data set consisting of the actual value of photovoltaic active power, P pv_pre The data set consists of the predicted values ​​of photovoltaic active power, P pv_act_t is the actual value of photovoltaic active power at time t, P pv_pre_t is the predicted value of photovoltaic active power at time t, and N is the total number of collected moments;

[0011] (2) Clustering of photovoltaic active power prediction error distribution;

[0012] (2.1), in the photovoltaic dataset P pv In the above example, all data points with zero PV active power prediction values ​​are deleted, and then the PV active power prediction error P at the remaining moments is calculated. pv_err_t =P pv_pre_t -P pv_act_t , thereby establishing the photovoltaic active power prediction error data set P pv_err ;

[0013] P pv_err =[P pv_err_1 ,P pv_err_2 ,…,P pv_err_t ,…,P pv_err_M ]

[0014] Among them, M represents the number of samples after eliminating invalid data;

[0015] (2.2), Clustering of photovoltaic active power prediction error distribution;

[0016] (2.2.1) Divide the total number of acquisition moments N in step (1) into n small intervals, and then divide the PV active power prediction error dataset P into n small intervals according to the corresponding time. pv_err Divide into n small intervals represents the i-th small interval;

[0017] Calculate the Wasserstein distance between any two cells:

[0018]

[0019] in, yes The cumulative distribution function of α∈(0,1) is Quantile function of , i,j∈[1,n];

[0020] (2.2.2) Based on the distance d between any two cells ij Form a distance matrix that measures the similarity of distributions in different intervals;

[0021]

[0022] (2.2.3) Based on the distance matrix D, the hierarchical clustering algorithm is applied to the photovoltaic active power prediction error data set P pv_err The photovoltaic active power prediction error P pv_err_t Perform grouping to obtain n1 error distribution;

[0023] (2.2.4) Based on the clustering results in step (2.2.3), the photovoltaic active power prediction error P in each type of error distribution is pv_err_t Divide into three categories again, and record the number of categories after classification as n2, n1≤n2≤3n1;

[0024] (3) Use the diffusion kernel density estimation algorithm to estimate the probability density function f of each type of photovoltaic active power prediction error distribution τ (P pv_err_t ), τ=1,2,…,n2;

[0025] (4) Construct an equivalent model of the distribution network;

[0026] Assume that the equivalent model of the distribution network containing the distributed photovoltaic power station includes: an equivalent photovoltaic module, an equivalent load module, an equivalent generator, an equivalent impedance 1 and an equivalent impedance 2; wherein the equivalent generator is connected in series with the equivalent impedance 2 and then connected in parallel with the equivalent photovoltaic module and the equivalent load module, and finally connected to the transmission network through the series equivalent impedance 1;

[0027] (5) Online identification of the parameters of the distribution network equivalent model is performed to identify the model parameters at the current time t;

[0028] (6) Call the distribution network equivalent model;

[0029] (6.1) Calculate the photovoltaic active power sample;

[0030] Collect the predicted photovoltaic active power value P of the distributed photovoltaic power station at time t+1 pv_pre_t+1; According to the current prediction time and prediction value P pv_pre_t+1 From n2 probability density functions f τ (P pv_err_t ) to select a photovoltaic active power prediction error probability density function; then use Latin hypercube sampling to sample the selected probability density function to obtain l photovoltaic active power prediction error samples at time t+1 And according to P pv_eq_t+1 =P pv_pre_t+1 -P pv_err_t+1 Calculate l photovoltaic active power samples at time t+1

[0031] (6.2) Calculate the photovoltaic reactive power sample;

[0032] Substitute l photovoltaic active power samples into the photovoltaic model in the distribution network equivalent model to calculate l photovoltaic reactive power samples

[0033] (6.3) Use the distribution network equivalent model to analyze the steady-state characteristics of the transmission network;

[0034] Input the l photovoltaic active power samples and photovoltaic reactive power samples obtained at time t+1 into the distribution network equivalent model after identification at time t in step (5), and obtain the power injected into the distribution network by the transmission network at time t+1. Boundary node voltage Power of any branch ij in the transmission network Voltage at any node i in the transmission network This allows the steady-state characteristics of the transmission network to be analyzed at time t+1.

[0035] The object of the invention of the present invention is achieved in this way:

[0036] The invention discloses a distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaic. Firstly, the historical predicted values ​​and actual values ​​of photovoltaic output in the distribution network area are collected and the prediction error is calculated; then the prediction error is clustered according to the prediction time and the size of the predicted value, and the probability density distribution of each clustering result is calculated; an equivalent model of the distribution network is established to determine the parameters to be identified, and then the electrical measurement values ​​of the distribution network are collected to identify the parameters of the equivalent model; finally, the equivalent model is called to predict the distributed photovoltaic output in the distribution network.

[0037] At the same time, the distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaic power generation in the present invention also has the following beneficial effects:

[0038] (1) The distribution network equivalent model established by the present invention uses photovoltaic output prediction error to describe photovoltaic uncertainty, avoiding the consideration of many factors causing photovoltaic uncertainty and effectively reducing the difficulty of describing photovoltaic uncertainty;

[0039] (2) The present invention considers the relationship between photovoltaic output prediction error and prediction time, and between prediction error and prediction value, and proposes a two-layer clustering framework for photovoltaic output prediction error distribution. First, a hierarchical clustering framework based on Wasserstein distance is applied to the relationship between prediction time and prediction error distribution, and then the prediction error is classified according to the size of the prediction value, so as to avoid drowning the error distribution characteristics under different states and improve the accuracy of photovoltaic output prediction error description;

[0040] (3) In the process of establishing the distribution network equivalent model and identifying the model parameters, the present invention takes into account the flow consistency and sensitivity consistency at the same time, which is conducive to improving the model accuracy. In addition, in order to adapt to the changes in the working state of the distribution network, the hybrid gray wolf-particle swarm optimization algorithm is used to perform online identification of the equivalent parameters. Combining the advantages of the two algorithms, the accuracy of the parameter identification results is effectively improved, thereby improving the accuracy of the equivalent model. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaics in the present invention;

[0042] Figure 2 It is a schematic diagram of IEEE33 node distribution network;

[0043] Figure 3 It is a schematic diagram of the equivalent model of the distribution network;

[0044] Figure 4 is the PV output prediction error distribution;

[0045] Figure 5 It is a comparison chart of the probability analysis results of the original distribution network and the equivalent model. DETAILED DESCRIPTION

[0046] The specific implementation of the present invention is described below in conjunction with the accompanying drawings so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

[0047] Example

[0048] In this embodiment, the simulation is carried out on a modified IEEE33-node distribution network, such as Figure 2As shown, node 21 of the IEEE30-node transmission network is connected to node 31 of the IEEE33-node distribution network through a step-down transformer, nodes 40, 44, 46, 48, 49, and 52 of the IEEE33-node distribution network are respectively connected to 0.28MW distributed photovoltaic power generation modules, nodes 54, 57, 61, and 62 are respectively connected to 0.3MW distributed photovoltaic power generation modules, and nodes 38 and 42 are respectively connected to distributed power generation modules with active power of 0.5MW and terminal voltage of 1, and the load in the distribution network is set to a constant impedance, constant current, and constant power model.

[0049] Next we combine Figure 2 A distribution network probabilistic equivalent modeling method considering the uncertainty of distributed photovoltaic output is described in detail. Figure 1 As shown, the following steps are included:

[0050] (1) Collect historical photovoltaic data;

[0051] The actual value and corresponding predicted value of photovoltaic active power of distributed photovoltaic power stations at each historical moment are collected over a period of time, which is recorded as photovoltaic data set P pv ;

[0052]

[0053] Among them, P pv_act is a data set consisting of the actual value of photovoltaic active power, P pv_pre The data set consists of the predicted values ​​of photovoltaic active power, P pv_act_t is the actual value of photovoltaic active power at time t, P pv_pre_t is the predicted value of photovoltaic active power at time t, and N is the total number of collected moments;

[0054] In this embodiment, the photovoltaic data are shown in Table 1:

[0055]

[0056] Table 1

[0057] (2) Clustering of photovoltaic active power prediction error distribution;

[0058] (2.1), in the photovoltaic dataset P pv In the above example, all data points with zero PV active power prediction values ​​are deleted, and then the PV active power prediction error P at the remaining moments is calculated. pv_err_t =P pv_pre_t -P pv_act_t , thereby establishing the photovoltaic active power prediction error data set P pv_err ;

[0059] P pv_err =[P pv_err_1 ,Ppv_err_2 ,…,P pv_err_t ,…,P pv_err_M ]

[0060] Among them, M represents the number of samples after eliminating invalid data;

[0061] (2.2), Clustering of photovoltaic active power prediction error distribution;

[0062] (2.2.1) Based on the 15-min resolution, the total number of acquisition moments N in step (1) is divided into n small intervals, n is 52, and then the PV active power prediction error dataset P is obtained according to the corresponding time. pv_err Divide into n small intervals represents the i-th small interval;

[0063] Calculate the Wasserstein distance between any two cells:

[0064]

[0065] in, yes The cumulative distribution function of α∈(0,1) is Quantile function of , i,j∈[1,n];

[0066] (2.2.2) Based on the distance d between any two cells ij Form a distance matrix that measures the similarity of distributions in different intervals;

[0067]

[0068] (2.2.3) Based on the distance matrix D, the hierarchical clustering algorithm is applied to the photovoltaic active power prediction error data set P pv_err The photovoltaic active power prediction error P pv_err_t After grouping, we get n1 error distribution, where n1 is 4, and the four time periods are 6:00-7:00 and 17:00-19:00, 7:00-9:00, 9:00-13:30, and 13:30-17:00;

[0069] (2.2.4) Based on the clustering results in step (2.2.3), the photovoltaic active power prediction error P in each type of error distribution is pv_err_tDivide into three categories again, and record the number of categories after classification as n2, n2 is 8, n1≤n2≤3n1, which are category 3 in the time periods of 6:00-7:00 and 17:00-19:00, category 2 and category 3 in the time period of 7:00-9:00, category 2 and category 3 in the time period of 9:00-13:30, and category 1, category 2 and category 3 in the time period of 13:30-17:00;

[0070] In this embodiment, according to the grouping result of step (2.2.3), the photovoltaic active power prediction error P in each type of error distribution is extracted. pv_err_t Corresponding prediction error P pv_pre_t , and then enter the following categories:

[0071] Category 1: P pv_pre_t / P pv_max ≥2 / 3;

[0072] Category 2: 1 / 3 ≤ P pv_pre_t / P pv_max <2 / 3;

[0073] Category 3: P pv_pre_t / P pv_max <1 / 3;

[0074] Among them, P pv_max It is the maximum active power output by the distributed photovoltaic power station.

[0075] (3) Use the diffusion kernel density estimation algorithm to estimate the probability density function f of each type of photovoltaic active power prediction error distribution τ (P pv_err_t ), τ=1,2,…,n2;

[0076] (4) Construct an equivalent model of the distribution network;

[0077] In this embodiment, if Figure 3 As shown, the equivalent model of the distribution network containing the distributed photovoltaic power station includes: an equivalent photovoltaic module, an equivalent load module, an equivalent generator, an equivalent impedance 1 and an equivalent impedance 2; wherein the equivalent generator is connected in series with the equivalent impedance 2 and then connected in parallel with the equivalent photovoltaic module and the equivalent load module, and finally connected to the transmission network through the series equivalent impedance 1;

[0078] In this embodiment, the specific models of each module in the distribution network equivalent model are as follows:

[0079] (4.1), the model of the equivalent photovoltaic module is:

[0080] The output active power P of the equivalent photovoltaic module at time t pv_eq_t for:

[0081] Ppv_eq_t =P pv_pre_t -P pv_err_t

[0082] The output reactive power Q of the equivalent photovoltaic module at time t pv_eq_t for:

[0083]

[0084] Among them, cosφ pv_t is the power factor of the photovoltaic system at time t, P th_t is the threshold active power at time t, h t is the slope of the power factor characteristic curve at time t, P th_t With h t is the parameter to be identified. In this embodiment, P th_t =0.5,h t =-0.15;

[0085] (4.2), the model of the equivalent load module is:

[0086] P Leq_t (V d_t )=a p1_t V d_t 2 +a p2_t V d_t +a p3_t

[0087] Q Leq_t (V d_t )=a q1_t V d_t 2 +a q2_t V d_t +a q3_t

[0088] Among them, P Leq_t , Q Leq_t are the active power and reactive power of the equivalent load module at time t, V d_t is the terminal voltage of the equivalent load module at time t, a p1_t ,a p2_t ,a p3_t is the active power load factor of the load model at time t, a q1_t ,a q2_t ,a q3_t is the reactive power load factor of the load model at time t, a p1_t ,a p2_t ,a p3_t , a q1_t ,a q2_t ,a q3_tare all parameters to be identified. In this embodiment, a p1_t ,a p2_t ,a p3_t , and a q1_t ,a q2_t ,a q3_t Changes continue over time;

[0089] (4.3), the model of equivalent impedance 1 and equivalent impedance 2 is:

[0090]

[0091] I B_t +I pv_t +I G_t =I d_t

[0092]

[0093] V d_t =V B_t -I B_t Z eq1_t

[0094] The superscript "*" indicates conjugation. is the total loss of the distribution network equivalent model at time t, S loss_t is the total loss of the original distribution network at time t, I G_t is the current of the generator at time t, I pv_t is the current of the equivalent photovoltaic module at time t, I B_t is the current injected by the transmission network into the distribution network equivalent model at time t, I d_t is the current of the equivalent load module at time t, Z eq1_t , Z eq2_t are the impedance values ​​of equivalent impedance 1 and equivalent impedance 2 at time t, V B_t is the voltage of the boundary node at time t, S pv_eq_t and S Leq_t are the complex powers of the equivalent photovoltaic module and the equivalent load module at time t, Y LL(LB)(LB) is the node admittance matrix of the boundary non-generator nodes, Y LL(LB)(LE) is the node admittance matrix between the boundary non-generator nodes and the non-generator nodes inside the distribution network, Y LL(LE)(LE) is the node admittance matrix of the non-generator nodes in the distribution network, Y LL(LE)(LB) is the node admittance matrix between the non-generator nodes and the boundary nodes in the distribution network, Y BI is the node admittance matrix of the boundary nodes and the internal nodes of the transmission network, Z eq1_t and Z eq2_t is the parameter to be identified;

[0095] (4.4), the model of the equivalent generator is:

[0096]

[0097] V G_t =V B_t -I B_t Z eq1_t -I G_t Z eq2_t

[0098] Among them, P G_t is the active power of the equivalent generator at time t, V G_t is the terminal voltage of the equivalent generator at time t, P Gi_t is the active power of the ith PV node type generator in the original distribution network at time t, n G is the number of PV node type generators in the original distribution network, n G =2.

[0099] (5) Online identification of the parameters of the distribution network equivalent model is performed to identify the model parameters at the current time t;

[0100] (5.1) Collect real-time operation data of the distribution network;

[0101] Collect the electrical measurement values ​​of the distribution network at the previous m moments before the current moment, m = 6, including the total active power P of the distributed photovoltaic power station at each moment pv_act_k and reactive power Q pv_act_k , the total active power of the load P d_k and reactive power Q d_k , the active power P injected into the distribution network by the transmission network in_k and reactive power Q in_k , PV node type generator total active power P G_k and reactive power Q G_k , where k = 1, 2, …, m;

[0102] (5.2) Calculate the total loss S of the distribution network loss_k ;

[0103] S loss_k =P loss_k +jQ loss_k

[0104] P loss_k =P in_k +P pv_act_k +P G_k -P d_k

[0105] Q loss_k =Qin_k +Q pv_act_k +Q G_k -Q d_k

[0106] (5.3) Construct the objective function and constraints for parameter identification of the distribution network equivalent model;

[0107] (5.3.1) Construct the objective function and constraints of equivalent impedance parameter identification:

[0108] The objective function is:

[0109]

[0110] The constraints are:

[0111]

[0112] (5.3.2) Construct the objective function and constraints of equivalent photovoltaic module parameter identification:

[0113] The objective function is:

[0114]

[0115] The constraints are:

[0116] 0≤P th_t ≤1,h t ≤0

[0117] (5.3.3) Construct the objective function and constraints for equivalent load module parameter identification:

[0118] The objective function is:

[0119]

[0120] V d_k =V B_k -I B_k Z eq1_k

[0121]

[0122] Among them, f Pk (·) and f Qk (·) represent the errors between the measured values ​​of the original distribution network system and the estimated values ​​of the equivalent model;

[0123] (5.4) Parameter identification of distribution network equivalent model;

[0124] Taking the objective function of distribution network equivalent model parameter identification as the goal and combining the constraint conditions of distribution network equivalent model parameter identification, the hybrid grey wolf-particle swarm algorithm is used to solve the objective function so as to identify the parameters to be identified of each equivalent module in the distribution network equivalent model.

[0125] (6) Call the distribution network equivalent model;

[0126] (6.1) Calculate the photovoltaic active power sample;

[0127] Collect the predicted photovoltaic active power value P of the distributed photovoltaic power station at time t+1 pv_pre_t+1 ; According to the current prediction time and prediction value P pv_pre_t+1 From n2 probability density functions f τ (P pv_err_t ) to select a photovoltaic active power prediction error probability density function; for example, the prediction time at 12:00 on July 3, 2011 is 12:00, the prediction value is 0.95MW, and it satisfies P pv_pre_t+1 / P pv_max >2 / 3, so first select the 9:00-13:30 time period in the n1 category according to the prediction time, and then select the photovoltaic active power prediction error probability density function corresponding to the 9:00-13:30 time period in category 1 according to the prediction value. Then apply Latin hypercube sampling to sample the selected probability density function to obtain l samples of photovoltaic active power prediction error at time t+1 And according to P pv_eq_t+1 =P pv_pre_t+1 -P pv_err_t+1 Calculate l photovoltaic active power samples at time t+1 l = 1000;

[0128] (6.2) Calculate the photovoltaic reactive power sample;

[0129] Substitute l photovoltaic active power samples into the photovoltaic model in the distribution network equivalent model to calculate l photovoltaic reactive power samples

[0130] (6.3) Use the distribution network equivalent model to analyze the steady-state characteristics of the transmission network;

[0131] Input the l photovoltaic active power samples and photovoltaic reactive power samples obtained at time t+1 into the distribution network equivalent model after identification at time t in step (5), and obtain the power injected into the distribution network by the transmission network at time t+1. Boundary node voltage Power of any branch ij in the transmission network Voltage at any node i in the transmission network This allows the steady-state characteristics of the transmission network to be analyzed at time t+1.

[0132] To verify the accuracy of the present invention, the photovoltaic output prediction error distribution in the distribution network area is first constructed, such as Figure 4 Then, we selected the photovoltaic and load data within 24 hours on a certain day in July to establish the distribution network equivalent model, and compared the probability equivalent model with the original system flow analysis results, as shown in Figure 2. Figure 5 shown.

[0133] Figure 4 The distribution of PV output forecast error is shown. First, the data is clustered according to the PV output forecast time, and four categories are obtained: 6:00-7:00 and 17:00-19:00, 7:00-9:00, 9:00-13:30, and 13:30-17:00. As shown in Figure (a), during the period of 7:00-9:00, the distribution of PV output forecast error is negatively skewed; Figure 4 As shown in (c), during the period of 13:30-17:00, the distribution of photovoltaic output prediction errors is positively skewed. The forecast errors during 6:00-7:00 and 17:00-19:00 are very small. Therefore, the prediction error clustering strategy based on prediction time can effectively distinguish the distributions of different skewnesses and avoid drowning the prediction error distribution characteristics caused by different prediction times. Then, according to the size of the prediction value, the above clustering results are further divided into three types of prediction errors of photovoltaic output prediction value intervals: large, medium, and small. In the time periods of 7:00-9:00 and 13:30-17:00, there are medium photovoltaic output prediction values ​​and small photovoltaic output prediction values. The error distribution in the medium photovoltaic output prediction value interval tends to be positively skewed relative to the small photovoltaic output prediction value interval. During the period of 9:00-13:30 as shown in Figure (b), all three photovoltaic output prediction values ​​exist, and the larger the photovoltaic output prediction value, the greater the probability of positive error. As shown in Figure (d), there are only small PV output prediction values ​​during 6:00-7:00 and 17:00-19:00, and no further classification is required. According to the above analysis, on the basis of the first-level clustering, the distributions with different characteristics are further effectively distinguished through the classification of prediction values.

[0134] Figure 5The comparison chart of the power flow analysis results of the probability equivalent model within 24 hours and the original system is shown. The horizontal axis is the time of the day. The vertical axis is the boundary point voltage diagram of the original system (a), the active power diagram injected into the distribution network by the transmission network (b), the reactive power diagram (c) and the probability results of the equivalent model expressed in the form of confidence intervals. When the photovoltaic output is 0, the distribution of the photovoltaic output prediction error is also 0, so the photovoltaic uncertainty is not considered. At this time, the equivalent model is deterministic. By calculation, the relative error of the boundary bus voltage during this period is less than 0.05%, the relative error of the injected active power is less than 4%, and the relative error of the injected reactive power is less than 3%. The error level of the equivalent model is acceptable. When the photovoltaic output is not equal to 0, the distribution of the photovoltaic output prediction error is used to describe the photovoltaic uncertainty. At this time, the equivalent model is a probabilistic equivalent, and most of the outputs are within the 90% confidence interval, which proves the effectiveness and accuracy of the proposed equivalent model. Compared with the traditional deterministic equivalent model, the probabilistic equivalent not only provides single-point information, but also provides richer probabilistic information.

[0135] Although the above describes the illustrative specific embodiments of the present invention to facilitate those skilled in the art to understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations using the concept of the present invention are protected.

Claims

1. A distribution network equivalent modeling and model calling method considering the uncertainty of distributed photovoltaics, characterized in that: The following steps are involved: (1) Collect historical photovoltaic data; The actual value and corresponding predicted value of photovoltaic active power of distributed photovoltaic power stations at each historical moment are collected over a period of time, which is recorded as photovoltaic data set P pv ; Among them, P pv_act is a data set consisting of the actual value of photovoltaic active power, P pv_pre The data set consists of the predicted values ​​of photovoltaic active power, P pv_act_t is the actual value of photovoltaic active power at time t, P pv_pre_t is the predicted value of photovoltaic active power at time t, and N is the total number of collected moments; (2) Clustering of photovoltaic active power prediction error distribution; (2.1), in the photovoltaic dataset P pv In the above example, all data points with zero PV active power prediction values ​​are deleted, and then the PV active power prediction error P at the remaining moments is calculated. pv_err_t =P pv_pre_t -P pv_act_t , thereby establishing the photovoltaic active power prediction error data set P pv_err ; P pv_err =[P pv_err_1 ,P pv_err_2 ,…,P pv_err_t ,…,P pv_err_M ] Among them, M represents the number of samples after eliminating invalid data; (2.2), Clustering of photovoltaic active power prediction error distribution; (2.2.1) Divide the total number of acquisition moments N in step (1) into n small intervals, and then divide the PV active power prediction error dataset P into n small intervals according to the corresponding time. pv_err Divide into n small intervals represents the i-th small interval; Calculate the distance between any two cells: in, yes The cumulative distribution function of α∈(0,1) is Quantile function of , i,j∈[1,n]; (2.2.2) Based on the distance d between any two cells ij Form a distance matrix that measures the similarity of distributions in different intervals; (2.2.3) Based on the distance matrix D, the hierarchical clustering algorithm is applied to the photovoltaic active power prediction error data set P pv_err The photovoltaic active power prediction error P pv_err_t Perform grouping to obtain n1 error distribution; (2.2.4) Based on the clustering results in step (2.2.3), the photovoltaic active power prediction error P in each type of error distribution is pv_err_t Divide into three categories again, and record the number of categories after classification as n2, n1≤n2≤3n1; (3) Use the diffusion kernel density estimation algorithm to estimate the probability density function f of each type of photovoltaic active power prediction error distribution τ (P pv_err_t ), τ=1,2,…,n2; (4) Construct an equivalent model of the distribution network; Assume that the equivalent model of the distribution network containing the distributed photovoltaic power station includes: an equivalent photovoltaic module, an equivalent load module, an equivalent generator, an equivalent impedance 1 and an equivalent impedance 2; wherein the equivalent generator is connected in series with the equivalent impedance 2 and then connected in parallel with the equivalent photovoltaic module and the equivalent load module, and finally connected to the transmission network through the series equivalent impedance 1; (5) Online identification of the parameters of the distribution network equivalent model is performed to identify the model parameters at the current time t; (6) Call the distribution network equivalent model; (6.1) Calculate the photovoltaic active power sample; Collect the predicted value of photovoltaic active power P of distributed photovoltaic power station at time t+1 pv_pre_t+1 ; According to the current prediction time and prediction value P pv_pre_t+1 From n2 probability density functions f τ (P pv_err_t ) to select a photovoltaic active power prediction error probability density function; then use Latin hypercube sampling to sample the selected probability density function to obtain l photovoltaic active power prediction error samples at time t+1 And according to P pv_eq_t+1 =P pv_pre_t+1 -P pv_err_t+1 Calculate l photovoltaic active power samples at time t+1 (6.2) Calculate the photovoltaic reactive power sample; Substitute l photovoltaic active power samples into the photovoltaic model in the distribution network equivalent model to calculate l photovoltaic reactive power samples (6.3) Use the distribution network equivalent model to analyze the steady-state characteristics of the transmission network; Input the l photovoltaic active power samples and photovoltaic reactive power samples obtained at time t+1 into the distribution network equivalent model after identification at time t in step (5), and obtain l groups of power injected into the distribution network by the transmission network at time t+1 Boundary node voltage Power of any branch ij in the transmission network Voltage at any node i in the transmission network This allows the steady-state characteristics of the transmission network to be analyzed at time t+1.

2. A distribution network equivalent modeling method considering the uncertainty of distributed photovoltaic according to claim 1, characterized in that: In the step (2.2.4), the photovoltaic active power prediction error P in each type of error distribution is pv_err_t The methods are further divided into three categories: Extract the photovoltaic active power prediction error P in each type of error distribution pv_err_t Corresponding prediction error P pv_pre_t , and then enter the following categories: Category 1: P pv_pre_t / P pv_max ≥2 / 3; Category 2: 1 / 3 ≤ P pv_pre_t / P pv_max <2 / 3; Category 3: P pv_pre_t / P pv_max <1 / 3; Among them, P pv_max It is the maximum active power output by the distributed photovoltaic power station.

3. A distribution network equivalent modeling method considering the uncertainty of distributed photovoltaic according to claim 1, characterized in that: The equivalent model of the distribution network is specifically: (3.1), the model of the equivalent photovoltaic module is: The output active power P of the equivalent photovoltaic module at time t pv_eq_t for: P pv_eq_t =P pv_pre_t -P pv_err_t The output reactive power Q of the equivalent photovoltaic module at time t pv_eq_t for: Among them, cosφ pv_t is the power factor of the photovoltaic system at time t, P th_t is the threshold active power at time t, h t is the slope of the power factor characteristic curve at time t, P th_t With h t is the parameter to be identified; (3.2), the model of the equivalent load module is: P Leq_t (In d_t )=a p1_t In d_t 2 +a p2_t In d_t +a p3_t Q Leq_t (V d_t )=a q1_t V d_t 2 +a q2_t V d_t +a q3_t Among them, P Leq_t , Q Leq_t are the active power and reactive power of the equivalent load module at time t, V d_t is the terminal voltage of the equivalent load module at time t, a p1_t ,a p2_t ,a p3_t is the active power load factor of the load model at time t, a q1_t ,a q2_t ,a q3_t is the reactive power load factor of the load model at time t, a p1_t ,a p2_t ,a p3_t , a q1_t ,a q2_t ,a q3_t These are parameters to be identified; (3.3), the model of equivalent impedance 1 and equivalent impedance 2 is: I B_t +I pv_t +I G_t =I d_t V d_t =V B_t -I B_t Z eq1_t The superscript "*" indicates conjugation. is the total loss of the distribution network equivalent model at time t, S loss_t is the total loss of the original distribution network at time t, I G_t is the current of the generator at time t, I pv_t is the current of the equivalent photovoltaic module at time t, I B_t is the current injected by the transmission network into the distribution network equivalent model at time t, I d_t is the current of the equivalent load module at time t, Z eq1_t , Z eq2_t are the impedance values ​​of equivalent impedance 1 and equivalent impedance 2 at time t, V B_t is the voltage of the boundary node at time t, S pv_eq_t and S Leq_t are the complex powers of the equivalent photovoltaic module and the equivalent load module at time t, Y LL(LB)(LB) is the node admittance matrix of the boundary non-generator nodes, Y LL(LB)(LE) is the node admittance matrix between the boundary non-generator nodes and the non-generator nodes inside the distribution network, Y LL(LE)(LE) is the node admittance matrix of the non-generator nodes in the distribution network, Y LL(LE)(LB) is the node admittance matrix between the non-generator nodes and the boundary nodes in the distribution network, Y BI is the node admittance matrix of the boundary nodes and the internal nodes of the transmission network, Z eq1_t and Z eq2_t is the parameter to be identified; (3.4), the model of the equivalent generator is: V G_t =V B_t -I B_t Z eq1_t -I G_t Z eq2_t Among them, P G_t is the active power of the equivalent generator at time t, V G_t is the terminal voltage of the equivalent generator at time t, P Gi_t is the active power of the ith PV node type generator in the original distribution network at time t, n G is the number of PV node type generators in the original distribution network.

4. A distribution network equivalent modeling method considering distributed photovoltaic uncertainty according to claim 1, characterized in that: The process of online identification of parameters of the distribution network equivalent model is as follows: (4.1) Collect real-time operation data of the distribution network; Collect the electrical measurement values ​​of the distribution network at the previous m moments before the current moment, including the total active power P of the distributed photovoltaic power station at each moment pv_act_k and reactive power Q pv_act_k , the total active power of the load P d_k and reactive power Q d_k , the active power P injected into the distribution network by the transmission network in_k and reactive power Q in_k , PV node type generator total active power P G_k and reactive power Q G_k , where k = 1, 2, …, m; (4.2) Calculate the total loss S of the distribution network loss_k ; S loss_k =P loss_k +jQ loss_k P loss_k =P in_k +P pv_act_k +P G_k -P d_k Q loss_k =Q in_k +Q pv_act_k +Q G_k -Q d_k (4.3) Construct the objective function and constraints for parameter identification of the distribution network equivalent model; (4.3.1) Construct the objective function and constraints of equivalent impedance parameter identification: The objective function is: The constraints are: (4.3.2) Construct the objective function and constraints of equivalent photovoltaic module parameter identification: The objective function is: The constraints are: 0≤P th_t ≤1,h t ≤0 (4.3.3) Construct the objective function and constraints of equivalent load module parameter identification: The objective function is: V d_k =V B_k -I B_k Z eq1_k Among them, f Pk (·) and f Qk (·) represent the errors between the measured values ​​of the original distribution network system and the estimated values ​​of the equivalent model; (4.4) Parameter identification of distribution network equivalent model; Taking the objective function of distribution network equivalent model parameter identification as the goal and combining the constraint conditions of distribution network equivalent model parameter identification, the hybrid grey wolf-particle swarm algorithm is used to solve the objective function, so as to identify the parameters to be identified of each equivalent module in the distribution network equivalent model.

Images