Voltage out-of-limit early warning method for power distribution network containing high-proportion distributed photovoltaic

Abstract

The invention discloses a voltage out-of-limit early warning method for a power distribution network containing high-proportion distributed photovoltaic. The method comprises the following steps: collecting and preprocessing node voltage influence factor data of the power distribution network; establishing a power distribution network node voltage time sequence model based on the dynamic Bayesian network based on the preprocessed power distribution network node voltage influence factor data; learning model parameters of the node voltage time sequence model by adopting a maximum posterior probability method; the node voltage is predicted based on the dynamic influence factor prediction curve, and early warning is given out according to the national standard limit value. The method is simple, easy to implement, high in generalization ability and high in accuracy, can adapt to changes of the power grid state and topology, and has great advantages in the aspect of power distribution network node voltage prediction compared with an existing method.

Description

Technical Field

[0001] The present invention relates to the technical field of voltage early warning for distribution networks, and in particular to a voltage over-limit early warning method for distribution networks containing a high proportion of distributed photovoltaics. Background Art

[0002] With the transformation of the global energy structure and the popularization of distributed energy, the distribution network is in a period of rapid development and change. Accurate voltage prediction of distribution network nodes has become the key. The operating status of the distribution network plays a vital role in the safe, stable and efficient operation of the power system, and voltage is an important indicator of the operating status of the power grid. Accurate voltage prediction can optimize the operation and control of the power grid, prevent voltage over-limit, improve the reliability and stability of the power grid, provide important information for planning and investment decisions, and support the effective integration of distributed energy resources. At present, with the popularization of new loads and power generation methods such as electric vehicles and distributed photovoltaics, the operating environment of the distribution network has become more and more complex, and accurate voltage prediction is particularly important.

[0003] Although the voltage prediction of distribution network nodes has been studied to a certain extent, there are still some key issues and challenges. First, existing studies do not fully consider the dynamic key factors that affect voltage, such as grid topology, electric vehicle charging pile power, etc., which may lead to deviations in prediction results. Secondly, most of the existing studies do not effectively deal with the temporal relationship and complex interaction of these factors, resulting in inaccurate predictions. Finally, many existing studies rely on fixed grid models and parameters, which makes it difficult for them to adapt to the dynamic changes of the grid and the access of new energy sources. In addition, some of the required data in the current methods, such as distribution network models, are difficult to obtain, which hinders the implementation of the methods. Therefore, it is urgent to develop a voltage prediction method that comprehensively considers key influencing factors, adapts to grid changes, and is simple and easy to implement.

[0004] Terminology explanation:

[0005] Distributed photovoltaic: Distributed photovoltaic is a photovoltaic power generation method with dispersed layout, local consumption and grid-connected operation. Distributed photovoltaic power generation equipment is usually distributed in places close to users, such as roofs, parking lots, farmland, etc.

[0006] Voltage over-limit: In the power system, voltage over-limit means that the voltage exceeds its specified operating range. Whether it is too high or too low, it may affect the normal operation of the power system, and may even cause equipment damage and affect the power quality. Voltage over-limit includes voltage over-limit and voltage over-limit, which means that the voltage exceeds or falls below the specified upper and lower limits respectively, and lasts for a certain period of time.

[0007] Network topology: In the present invention, it refers to the connection and layout between power sources, loads and power equipment in the distribution network.

[0008] Photovoltaic active output: refers to the active electrical energy actually delivered to the grid by the photovoltaic power generation system, usually in kilowatts or megawatts. This parameter is affected by factors such as light intensity and temperature.

[0009] Ordinary load: in this invention, it refers to the power load in the distribution network other than electric vehicle charging piles, including households, commercial and industrial users, etc.

[0010] Prior distribution: refers to the assumption about the probability distribution of unknown parameters before any data is observed, which is based on the observer's prior knowledge, experience or subjective judgment of the parameters.

[0011] Posterior probability: refers to the probability obtained by updating the probability distribution of unknown parameters after observing the data. It combines the information of the prior distribution and the observed data and is calculated according to Bayes' theorem. Summary of the invention

[0012] The purpose of the present invention is to solve the deficiencies of the above-mentioned prior art, thereby providing a distribution network voltage over-limit warning method containing a high proportion of distributed photovoltaics. The method is simple, easy to implement, has strong generalization ability, high accuracy, and can adapt to changes in grid status and topology. Compared with existing methods, it has great advantages in distribution network node voltage prediction.

[0013] The technical solution adopted by the present invention is as follows: a voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaics, comprising the following steps:

[0014] Data collection and preprocessing of factors affecting voltage at distribution network nodes;

[0015] Based on the pre-processed data of factors affecting the voltage of distribution network nodes, a voltage time series model of distribution network nodes based on dynamic Bayesian network is established;

[0016] The maximum a posteriori probability method is used to learn the model parameters of the node voltage timing model;

[0017] The node voltage is predicted based on the dynamic influencing factor prediction curve and an early warning is issued according to the national standard limit.

[0018] Data collection and preprocessing of factors affecting voltage at distribution network nodes, specifically:

[0019] 1) Data collection of factors affecting voltage at distribution network nodes

[0020] The opening and closing conditions of each connecting switch in the distribution network form different topological structures. Assuming that the distribution network has R topologies, M m (m=1,…,R) represents the above R topological structures;

[0021] Collect the topological structure S(t) of the distribution network at time t, S(t) = M m ;

[0022] The photovoltaic active power output power P of access node j at time t pv,j (t);

[0023] The load power P of access node j at time t load,j (t), unit is kVA;

[0024] The charging pile power P connected to node j at time t charge,j (t), unit is kW;

[0025] And the voltage per unit value of node j at time t is V(t,j), in units of pu;

[0026] 2) Data preprocessing

[0027] Normalize the data collected in step 1) to ensure that the data is in the range of [0,1] and remove the influence of different dimensions to facilitate the subsequent model construction and parameter learning; let X r , r = 1 to 5 represent the topology, photovoltaic output, load power, charging pile power and voltage per unit value, respectively, x r , r = 1 to 5 represent the data of the above factors at time t, and then normalize them according to the following formula:

[0028]

[0029] The time interval t for collecting data on factors affecting voltage at distribution network nodes is 15 minutes, and the duration is no less than 3 months.

[0030] A voltage time series model of distribution network nodes based on dynamic Bayesian network is established, specifically:

[0031] Assuming that in a distribution network with a high proportion of distributed photovoltaics, the impact of each influencing factor on the node voltage is linear, and an error term is introduced to capture the randomness of the impact, the node voltage timing model is:

[0032]

[0033] Where w1 represents the influence of network topology on node voltage, w 2,k ,w 3,k and w 4,k Indicates the influence of the photovoltaic output, load power and charging pile power of node k on the voltage of node j. K2, K3 and K4 are the node numbers of the distributed photovoltaic, common load and charging pile connected to the network, respectively. ε j(t) is the error term of the node j voltage, which is usually assumed to be normally distributed with a mean of 0;

[0034] θ={w1,w 2,k ,w 3,k ,w 4,k} is the model parameter set, D={S(t),P pv,k (t),P load,k (t),P charge,k (t),V j (t)} is the set of observed data of influencing factors.

[0035] The maximum a posteriori probability method is used to learn the model parameters of the node voltage timing model, specifically:

[0036] 1) Construct likelihood function

[0037] For each node j, the node voltage V at time t j The probability density function of (t) is expressed as:

[0038]

[0039] Where N represents normal distribution, σ 2 is the variance of the error term, also the parameter to be estimated;

[0040] 2) Define the prior distribution

[0041] Use uniform distribution as the prior distribution of the model parameters to be estimated, and set the weights w1,w 2,k ,w 3,k ,w 4,k The value range is [-1,1], then set w1,w 2,k ,w 3,k ,w 4,k The prior distribution of is U(-1,1), where U represents uniform distribution, which is expressed by the following formula:

[0042]

[0043]

[0044] 3) Calculate and optimize the posterior probability

[0045] According to Bayes' theorem, the posterior probability of the model parameters is expressed as the product of the likelihood function and the prior distribution, that is, p(θ|D)=p(D|θ)p(θ), p(θ|D)=p(D|θ)p(θ) is a function of the model parameter θ. In order to maximize the posterior probability p(θ|D), equation (6) needs to be optimized:

[0046] max p(θ|D)=max p(D|θ)p(θ) (6)

[0047] Taking the logarithm of the above formula (6), it is transformed into the following form:

[0048] max log(p(θ|D))=max[log(p(D|θ))+log(p(θ))] (7)

[0049] Where log(p(θ|D)) is called the logarithmic posterior probability function, and the model parameters when its value is maximum are expressed as:

[0050]

[0051] Using Newton's method, equation (8) is iterated according to equation (9) until the model parameters converge, that is, ||θ (i+1) -θ (i) ||<0.01, the obtained parameter value θ is the parameter value that maximizes the logarithmic posterior probability, that is, w1,w 2,k (k∈K2),w 3,k (k∈K3),w 4,k (k∈K4);||·|| uses the Euclidean norm, that is, the vector θ (i+1) -θ (i) The square root of the sum of the squares of the elements of ;

[0052] θ (i+1) =θ (i) -H -1 (θ (i) )g(θ (i) ) (9)

[0053] In the formula, H -1 (θ (i) ) represents the inverse of the Hessian matrix of the model parameter θ at the i-th iteration. The Hessian matrix is ​​a matrix composed of the second-order partial derivatives of the logarithmic posterior probability function. The calculation expression of the b-th row and c-th column element of the Hessian matrix is:

[0054]

[0055] In the formula, is the second-order partial derivative of the b-th and c-th elements in the model parameter vector;

[0056] g(θ (i) ) is the gradient of the logarithmic posterior probability function at the i-th iteration, that is, the gradient vector composed of the first-order partial derivatives of the logarithmic posterior probability function. The expression of the b-th element in the gradient vector is:

[0057]

[0058] In the formula, is the parameter θ b The partial derivative of .

[0059] Based on the prediction curve of dynamic influencing factors, the node voltage is predicted and an early warning is issued according to the national standard limit value, specifically:

[0060] 1) Collect forecast data

[0061] Collect the distributed photovoltaic active output prediction curve P of each node pv_pre (t), load prediction curve of each node P load_pre (t) and the power prediction curve P of each node charging pile charge_pre (t), and normalized according to formula (1);

[0062] 2) Predict node voltage

[0063] If the operation status of each switch in the future medium- and short-term distribution network can be obtained, that is, S(t) = M1, M2, ..., M R , then combined with P pv_pre (t), P load_pre (t) and P charge_pre (t), substitute the node voltage timing model after the model parameter learning to calculate the node voltage prediction curve V at each time t pre (t);

[0064] If the future distribution network structure state cannot be obtained, the R types of distribution network topologies are substituted into the node voltage time series model after model parameter learning to estimate the R types of node voltage prediction curves V pre (t);

[0065] The error term ε(t) in formula (2) is set to obey a normal distribution with a mean of 0 and a variance of 0.01, that is, ε(t)~N(0,0.01);

[0066] 3) Voltage over-limit warning

[0067] According to the national standard, if the node predicted voltage exceeds or falls below the specified threshold for two consecutive times t, a voltage over-limit warning is issued, that is:

[0068]

[0069] The beneficial effects of the present invention are:

[0070] 1) The present invention proposes for the first time a voltage over-limit warning method for distribution network nodes containing a high proportion of distributed photovoltaics. In view of the defects of existing methods such as the large number of required information, difficulty in implementation, and inability to adapt to dynamic topology networks, the present invention comprehensively considers key dynamic influencing factors such as power grid topology, photovoltaic power generation output, load power, and charging pile power, and realizes accurate characterization of voltage changes in distribution network nodes.

[0071] 2) The present invention is based on real-time and accessible distribution network data, which enables convenient implementation and application. It can adapt to changes in grid topology and the access of new energy sources or loads, and provides important support for solving voltage over-limit problems in distribution network operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a flow chart of a voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaics according to an embodiment of the present invention;

[0073] Figure 2 A network topology diagram of a power distribution network provided by an embodiment of the present invention;

[0074] Figure 3 This is a diagram of the voltage over-limit warning effect of a distribution network node provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] like Figure 1 As shown, an embodiment of the present invention provides a voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaics, comprising the following steps:

[0077] Step 1: Data collection and preprocessing of factors affecting voltage at distribution network nodes

[0078] Data Collection:

[0079] 1) Network topology information collection

[0080] The opening and closing conditions of each connecting switch in the distribution network form different topological structures. Assuming that the distribution network has R topologies, M m (m=1,…,R) represents the above R topological structures. The distribution network dispatching department collects the historical topological structure information of the distribution network, and S(t) represents the topological structure of the distribution network at time t, S(t)=Mm The interval of time t is 15 minutes, and the data duration is no less than 3 months.

[0081] 2) Photovoltaic active power output

[0082] The metering department collects the historical active output data of distributed photovoltaics. pv,j (t) represents the photovoltaic active power output power of access node j at time t, in kW, and the interval of time t is 15 minutes, and the duration is not less than 3 months

[0083] 3) Load power

[0084] The marketing center collects historical load data of each node in the distribution network. load,j (t) represents the load power of access node j at time t, in kVA. The interval of time t is 15 minutes and the duration is not less than 3 months.

[0085] 4) Charging pile power

[0086] The new energy vehicle charging pile management department collects the historical charging data of the charging piles. charge,j (t) represents the power of the charging pile connected to node j at time t, in kW. The interval of time t is 15 minutes and the duration is not less than 3 months.

[0087] 5) Node voltage

[0088] The distribution network monitoring system collects historical voltage data of each node. V(t,j) represents the voltage per unit value of node j at time t, with the unit being pu. The interval of time t is 15 minutes, and the duration is not less than 3 months.

[0089] Data preprocessing

[0090] Normalize the collected data to ensure that the data is within the range of [0,1] and remove the influence of different dimensions to facilitate the subsequent model construction and parameter learning. r , r = 1 to 5 represent the set of historical data series of grid topology, photovoltaic output, load power, charging pile power and node voltage, respectively, x r , r = 1 to 5 represent the data of the above factors at time t, and then normalize them according to the following formula:

[0091]

[0092] In this embodiment, the above data series are collected for 3 months, that is, t=1, 2, ..., 96, ..., 2880.

[0093] Step 2: Establish a voltage time series model for distribution network nodes based on dynamic Bayesian network

[0094] Figure 2 The figure shows a topological diagram of a power distribution network with a high proportion of distributed photovoltaics provided by an embodiment of the present invention. The model used in the embodiment of the present invention is a modified IEEE-33 node power distribution system, in which 6 sets of distributed photovoltaic systems, 5 groups of electric vehicle charging piles and 3 tie switches are connected to the standard IEEE-33 node power distribution system, and other parameters of the system adopt the standard IEEE-33 node power distribution system data.

[0095] Assuming that in this distribution network, the impact of each influencing factor on the node voltage is linear, and an error term is introduced to capture the randomness of the impact, the node voltage linear model is as follows:

[0096]

[0097] Where w1 represents the influence of network topology on node voltage, w 2,k ,w 3,k and w 4,k represents the influence of the photovoltaic output, load power and charging pile power of node k on the voltage of node j, K2, K3 and K4 are the node numbers of the distributed photovoltaic, common load and charging pile connected to the network respectively. Figure 2 , K2={9,13,16,24,28,33}, K3={3,4,7,8,10,12,14,15,17,19,20,22,23,25,29,31,32}, K4={11,18,21,27,30}. ε j (t) is the error term of the node j voltage, which is usually assumed to be normally distributed with a mean of 0. Let θ = {w1,w 2,k ,w 3,k ,w 4,k} is the model parameter set, D={S(t),P pv,k (t),P load,k (t),P charge,k (t),V j (t)} is the set of observed data of influencing factors.

[0098] Step 3: Use the maximum a posteriori probability method to learn the model parameters

[0099] The present invention adopts the maximum a posteriori probability method to learn model parameters, and the specific steps are as follows:

[0100] Step 3.1: Construct the likelihood function

[0101] The likelihood function describes the probability of observing actual data given the model parameters. Assuming that each node voltage is observed independently, the likelihood function can be written as the product of the probability density functions of all node voltages. For each node j, the voltage V at time t j The probability density function of (t) can be expressed as:

[0102]

[0103] Where N represents normal distribution, σ 2 is the variance of the error term, which is also the parameter to be estimated.

[0104] Step 3.2: Define the prior distribution

[0105] Prior distribution represents the prior knowledge of model parameters when no data is observed. The present invention adopts uniform distribution as the prior distribution of the parameters to be estimated. According to the operation experience of the distribution network, it is considered that the value range of weight w is [-1, 1], so w1, w 2,k ,w 3,k ,w 4,k The prior distribution of is U(-1,1), where U represents uniform distribution, which is expressed by the following formula:

[0106]

[0107]

[0108] Step 3.3: Calculate and optimize the posterior probability

[0109] According to Bayes' theorem, the posterior probability of the model parameters is expressed as the product of the likelihood function and the prior distribution, that is, p(θ|D)=p(D|θ)p(θ), p(θ|D)=p(D|θ)p(θ) is a function of the model parameter θ. In order to maximize the posterior probability p(θ|D), equation (6) needs to be optimized:

[0110] max p(θ|D)=max p(D|θ)p(θ) (6)

[0111] Taking the logarithm of the above formula (6), it is transformed into the following form

[0112] max log(p(θ|D))=max[log(p(D|θ))+log(p(θ))] (7)

[0113] Where log(p(θ|D)) is called the logarithmic posterior probability function, and the model parameters when its value is maximum are expressed as

[0114]

[0115] Formula (8) is the model parameter expression that needs to be optimized and solved.

[0116] Using Newton's method, equation (8) is iterated according to equation (9) until the model parameters converge, that is, ||θ (i+1) -θ (i) ||<0.01, the obtained parameter value θ is the parameter value that maximizes the logarithmic posterior probability, that is, w1,w 2,k (k∈K2),w 3,k (k∈K3),w 4,k (k∈K4);||·|| uses the Euclidean norm, that is, the vector θ (i+1) -θ (i) The square root of the sum of the squares of the elements of ;

[0117] θ (i+1) =θ (i) -H -1 (θ (i) )g(θ (i) ) (9)

[0118] In the formula, H -1 (θ (i) ) represents the inverse of the Hessian matrix of the parameter θ at the i-th iteration. The Hessian matrix is ​​a matrix composed of the second-order partial derivatives of the logarithmic posterior probability function. The calculation expression of the b-th row and c-th column of the Hessian matrix is:

[0119]

[0120] In the formula, are the second-order partial derivatives of the b-th and c-th elements in the model parameter vector.

[0121] g(θ (i) ) is the gradient of the logarithmic posterior probability function at the i-th iteration, that is, the vector composed of the first-order partial derivatives of the logarithmic posterior probability function. The expression of the b-th element in the gradient vector is:

[0122]

[0123] In the formula, is the parameter θ b The partial derivative of .

[0124] Step 4: Predict the node voltage based on the dynamic influencing factor prediction curve and issue an early warning according to the national standard limit

[0125] Step 4.1: Collect forecast data

[0126] Collect the distributed photovoltaic active output prediction curve P of each node from relevant departments pv_pre(t), load prediction curve of each node P load_pre (t) and the power prediction curve P of each node charging pile charge_pre (t). Perform preprocessing according to the data format and normalization method in step 1.

[0127] Step 4.2: Predict node voltages

[0128] If the operation status of each switch in the future medium- and short-term distribution network can be obtained through the dispatching department, that is, S(t) = M1, M2, …, M R , then combined with P pv_pre (t), P load_pre (t) and P charge_pre (t), and substitute it into the node voltage timing model after model parameter learning to calculate the node voltage prediction curve V at each time t pre (t). If the future distribution network structure state cannot be obtained, the R types of distribution network topologies are substituted into the node voltage time series model after model parameter learning to estimate R types of node voltage prediction curves.

[0129] The error term ε(t) in formula (2) is set in the present invention to obey a normal distribution with a mean of 0 and a variance of 0.01, that is, ε(t)~N(0,0.01).

[0130] Step 4.3: Voltage over-limit warning

[0131] If the node predicted voltage exceeds or falls below the specified threshold for two consecutive moments, a voltage over-limit warning is issued, that is:

[0132]

[0133] Comparative analysis to verify the effectiveness of the model

[0134] This embodiment uses three months of historical data to learn model parameters, predicts the 24-hour voltage per unit value curve of node 17 on a certain day in the fourth month, and compares it with the actual voltage monitoring value and over-limit situation on that day to verify the effectiveness of the model.

[0135] Figure 3 The figure shows the effect diagram of voltage prediction and over-limit warning provided by an embodiment of the present invention. The black solid line is the actual voltage curve of the day, and the black dotted line is the predicted voltage curve of the day. It can be seen that the error between the two is small, indicating that the present invention can accurately predict the node voltage curve. The gray squares represent the actual voltage over-limit time periods, which are {10:00~10:15}, {11:00~15:45}, {16:00~16:45} respectively. The overlap rate between the warned voltage over-limit time period and the actual voltage over-limit time period is 95.84%, which shows the effectiveness of the present invention in voltage over-limit warning.

[0136] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaics, characterized in that: The following steps are involved: Data collection and preprocessing of factors affecting voltage at distribution network nodes; Based on the pre-processed data of factors affecting the voltage of distribution network nodes, a voltage time series model of distribution network nodes based on dynamic Bayesian network is established; The maximum a posteriori probability method is used to learn the model parameters of the node voltage timing model; The node voltage is predicted based on the dynamic influencing factor prediction curve and an early warning is issued according to the national standard limit.

2. The voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaic power generation according to claim 1 is characterized in that: Data collection and preprocessing of factors affecting voltage at distribution network nodes, specifically: 1) Data collection of factors affecting voltage at distribution network nodes The opening and closing conditions of each connecting switch in the distribution network form different topological structures. Assuming that the distribution network has R topologies, M m (m=1,…,R) represents the above R topological structures; Collect the topological structure S(t) of the distribution network at time t, S(t) = M m ; The photovoltaic active power output power P of access node j at time t pv,j (t); The load power P of access node j at time t load,j (t), unit is kVA; The charging pile power P connected to node j at time t charge,j (t), unit is kW; And the voltage per unit value of node j at time t is V(t,j), in units of pu; 2) Data preprocessing Normalize the data collected in step 1) to ensure that the data is in the range of [0,1] and remove the influence of different dimensions to facilitate the subsequent model construction and parameter learning; let X r , r = 1 to 5 represent the topology, photovoltaic output, load power, charging pile power and voltage per unit value, respectively, x r , r = 1 to 5 represent the data of the above factors at time t, and then normalize them according to the following formula:

3. The voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaic power generation according to claim 1 is characterized in that: The time interval t for collecting data on factors affecting voltage at distribution network nodes is 15 minutes, and the duration is no less than 3 months.

4. The voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaic power generation according to claim 1 is characterized in that: A voltage time series model of distribution network nodes based on dynamic Bayesian network is established, specifically: Assuming that in a distribution network with a high proportion of distributed photovoltaics, the impact of each influencing factor on the node voltage is linear, and an error term is introduced to capture the randomness of the impact, the node voltage timing model is: Where w1 represents the influence of network topology on node voltage, w 2,k ,w 3,k and w 4,k Indicates the influence of the photovoltaic output, load power and charging pile power of node k on the voltage of node j. K2, K3 and K4 are the node numbers of the distributed photovoltaic, common load and charging pile connected to the network, respectively. ε j (t) is the error term of the node j voltage, which is usually assumed to be normally distributed with a mean of 0; θ={w1,w 2,k ,w 3,k ,w 4,k } is the model parameter set, D={S(t),P pv,k (t),P load,k (t),P charge,k (t),V j (t)} is the observed data set of influencing factors.

5. The voltage over-limit warning method for a distribution network containing a high proportion of distributed photovoltaic power generation according to claim 1 is characterized in that: The maximum a posteriori probability method is used to learn the model parameters of the node voltage timing model, specifically: 1) Construct likelihood function For each node j, the node voltage V at time t j The probability density function of (t) is expressed as: Where N represents normal distribution, σ 2 is the variance of the error term, also the parameter to be estimated; 2) Define the prior distribution Use uniform distribution as the prior distribution of the model parameters to be estimated, and set the weights w1,w 2,k ,w 3,k ,w 4,k The value range is [-1,1], then set w1,w 2,k ,w 3,k ,w 4,k The prior distribution of is U(-1,1), where U represents uniform distribution, which is expressed by the following formula: 3) Calculate and optimize the posterior probability According to Bayes' theorem, the posterior probability of the model parameters is expressed as the product of the likelihood function and the prior distribution, that is, p(θ|D)=p(D|θ)p(θ), p(θ|D)=p(D|θ)p(θ) is a function of the model parameter θ. In order to maximize the posterior probability p(θ|D), equation (6) needs to be optimized: max p(θ|D)=max p(D|θ)p(θ) (6) Taking the logarithm of the above formula (6), it is transformed into the following form: max log(p(θ|D))=max[log(p(D|θ))+log(p(θ))] (7) Where log(p(θ|D)) is called the logarithmic posterior probability function, and the model parameters when its value is maximum are expressed as: Using Newton's method, equation (8) is iterated according to equation (9) until the model parameters converge, that is, ||θ (i+1) -θ (i) ||<0.01, the obtained parameter value θ is the parameter value that maximizes the logarithmic posterior probability, that is, w1,w 2,k (k∈K2),w 3,k (k∈K3),w 4,k (k∈K4);||·|| uses the Euclidean norm, that is, the vector θ (i+1) -θ (i) The square root of the sum of the squares of the elements of ; i (i+1) =θ (i) -H -1 (i (i) )g(θ (i) ) (9) In the formula, H -1 (θ (i) ) represents the inverse of the Hessian matrix of the model parameter θ at the i-th iteration. The Hessian matrix is ​​a matrix composed of the second-order partial derivatives of the logarithmic posterior probability function. The calculation expression of the b-th row and c-th column element of the Hessian matrix is: In the formula, is the second-order partial derivative of the b-th and c-th elements in the model parameter vector; g(θ (i) ) is the gradient of the logarithmic posterior probability function at the i-th iteration, that is, the gradient vector composed of the first-order partial derivatives of the logarithmic posterior probability function. The expression of the b-th element in the gradient vector is: In the formula, is the parameter θ b The partial derivative of .

6. The voltage over-limit warning method for a distribution network with a high proportion of distributed photovoltaic power generation according to claim 1 is characterized in that: Based on the prediction curve of dynamic influencing factors, the node voltage is predicted and an early warning is issued according to the national standard limit value, specifically: 1) Collect forecast data Collect the distributed photovoltaic active output prediction curve P of each node pv_pre (t), load prediction curve of each node P load_pre (t) and the power prediction curve P of each node charging pile charge_pre (t), and normalized; 2) Predict node voltage If the operation status of each switch in the future medium- and short-term distribution network can be obtained, that is, S(t) = M1, M2, ..., M R , then combined with P pv_pre (t), P load_pre (t) and P charge_pre (t), and substitute it into the node voltage timing model after model parameter learning to calculate the node voltage prediction curve V at each time t pre (t); If the future distribution network structure state cannot be obtained, the R types of distribution network topologies are substituted into the node voltage time series model after model parameter learning to estimate the R types of node voltage prediction curves V pre (t); The error term ε(t) in formula (2) is set to obey a normal distribution with a mean of 0 and a variance of 0.01, that is, ε(t)~N(0,0.01); 3) Voltage over-limit warning According to the national standard, if the node predicted voltage exceeds or falls below the specified threshold for two consecutive times t, a voltage over-limit warning is issued, that is:

Images