Distributed photovoltaic grid-connected power distribution network risk assessment method

Abstract

The invention discloses a distributed photovoltaic grid-connected power distribution network risk assessment method, which overcomes the problem of low power distribution network risk assessment accuracy after distributed photovoltaic grid connection in the prior art, and comprises the following steps: analyzing distributed photovoltaic output characteristics, carrying out time sequence acquisition on power distribution network load data under different illumination intensities, and carrying out power distribution network risk assessment on the power distribution network load data under different illumination intensities; processing the collected load data; decomposing load data by using singular spectrum analysis, and establishing a BP neural network to predict a load after distributed photovoltaic grid connection; and taking the load prediction result as input, establishing a distributed photovoltaic grid-connected power grid influence quantitative model, and performing risk assessment on the distributed photovoltaic grid-connected power distribution network. The photovoltaic output uncertainty is fully considered, the load prediction precision is improved, power distribution network risk assessment is performed based on the load prediction result, related risk points can be predicted in advance, and the power grid operation reliability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid risk assessment, and in particular to a distributed photovoltaic grid-connected distribution network risk assessment method. Background Art

[0002] In recent years, with the development of distributed power generation technology, the application of distributed photovoltaics on the user side has made breakthrough progress. A large number of distributed photovoltaics have penetrated into the power system by directly connecting to the distribution network system, and their proportion in power production is increasing. When photovoltaics are connected to the distribution network, occasional output fluctuations will bring shocks to the distribution network, causing the distribution network to be in an uncertain state, and its safety and reliability are threatened; at the same time, the access of distributed photovoltaics has changed the topological structure of the distribution network, and the power flow distribution has become more complex, which has an impact on the voltage, active power and other state quantities of the distribution network that cannot be ignored.

[0003] Therefore, it is necessary to study the impact of large-scale distributed photovoltaics on the operating risks of distribution networks, and propose reasonable and accurate assessment methods to analyze the risks of distribution networks, so as to provide a basis for risk control, and to assess the risks brought by distributed photovoltaic access to distribution networks as accurately as possible, so as to provide a basis for the safe and stable operation of distribution networks. Existing assessment technologies mainly include multi-scenario risk assessment, over-limit risk assessment, subjective and objective mixed assessment methods, etc. However, due to the uncertain output scenarios of large-scale distributed photovoltaics, existing assessment methods cannot effectively realize the risk analysis of distribution networks. Summary of the invention

[0004] The purpose of the present invention is to overcome the problem of low accuracy of distribution network risk assessment after distributed photovoltaic grid-connected in the prior art, and provide a distributed photovoltaic grid-connected distribution network risk assessment method, which fully considers the uncertainty of photovoltaic output, improves the load forecasting accuracy, and performs distribution network risk assessment based on the load forecasting results, which can predict relevant risk points in advance and improve the reliability of grid operation.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: A distributed photovoltaic grid-connected distribution network risk assessment method, characterized in that it includes the following steps: Analyze the output characteristics of distributed photovoltaics, collect the load data of distribution networks under different light intensities in a time series, and process the collected load data; The load data is decomposed by singular spectrum analysis, and a BP neural network is established to predict the load of distributed photovoltaic power generation after grid connection. With the load forecast results as input, a quantitative model of the impact of distributed photovoltaic grid-connected power grid is established, and the risk assessment of the distribution network of distributed photovoltaic grid-connected power grid is carried out.

[0006] After distributed photovoltaic grid connection, the distributed photovoltaic output is unstable, so the error of distributed photovoltaic load prediction is large. The present invention analyzes the photovoltaic output characteristics during load prediction, and uses seven singular spectrum analysis to reduce data dimensions and improve load prediction accuracy. Based on the load prediction results, the distribution network risk assessment is carried out, which can predict relevant risk points in advance and improve the reliability of grid operation.

[0007] Preferably, the load characteristics of the power grid at the current moment are analyzed to generate a load fluctuation signal; if the load fluctuation signal is a low fluctuation signal, the load is predicted; if the load fluctuation signal is a high fluctuation signal, a risk warning signal is immediately output to control the power grid load.

[0008] Preferably, the decomposition of load data by singular spectrum analysis includes: performing singular value decomposition on the load data, deleting the noise sequence, and obtaining a plurality of subsequences; using BP neural network to perform load prediction on each subsequence respectively, superimposing the load prediction results of each sequence, and obtaining a load prediction value.

[0009] Preferably, the quantification model of the impact of distributed photovoltaic grid-connected power grid is: the distribution network risk is equal to the product of the first load factor, the second load factor and the severity of the consequence, the first load factor is the integral of the load forecast result over the time interval T, the second load factor is the product of factors affecting the load, and the severity of the consequence is the product of the importance of each device in the distribution network and the degree of equipment loss.

[0010] Preferably, step S3 includes using a random forest algorithm to establish a quantification model of the impact of distributed photovoltaic grid-connected power grids and perform a risk assessment of voltage over-limit in the distribution network: establishing a data set, the data set including historical load data; selecting a base learner, determining the number of decision trees to be constructed, and training a random forest model of voltage over-limit risk in the distribution network for different time periods in the future; and evaluating the obtained random forest model.

[0011] As a preference, four indicators, namely, daily maximum load forecast difference, daily load forecast average difference, daily load forecast root mean square error, and daily minimum load forecast difference, are selected, and the weights of the four indicators are determined as BP neural network evaluation indicators.

[0012] Preferably, the current moment and the predicted moment are determined, and the load fluctuation at the same moment and the same light intensity in the historical data is used as the reference value. If the difference between the load prediction value and the load fluctuation at the current moment and the reference value is greater than the preset threshold, the current distribution network is directly determined to be high risk. If the difference between the load prediction value and the load fluctuation at the current moment and the reference value is less than the preset threshold, the current distribution network is low risk, and the subsequent steps are continued.

[0013] Preferably, the data set samples are subjected to unbalanced processing and standardized processing, wherein the unbalanced processing includes under-sampling the normal voltage data and over-sampling the low voltage over-limit and high voltage over-limit voltage data.

[0014] Preferably, the risk of voltage exceeding the limit is suppressed by controlling the reactive power of distributed photovoltaic grid-connected power.

[0015] Preferably, the time series acquisition includes placing data collected at the same time together, dividing the collected data according to date, and arranging and encoding them in chronological order.

[0016] Preferably, the importance of the variables of each random forest model is calculated to determine the influence of each variable on the voltage over-limit of the distribution network.

[0017] Therefore, the present invention has the following beneficial effects: 1. When predicting the load of the distribution network after the distributed photovoltaic grid is connected, the present invention first analyzes the distributed photovoltaic output characteristics, collects load data based on the analysis results, classifies the load data collected at the same time, and uses the singular spectrum to reduce the data dimension. In this way, a BP neural network model is constructed to predict the load of the distribution network, thereby improving the accuracy and effectiveness of load prediction.

[0018] 2. Based on the load forecast results, the present invention evaluates the risk of the distribution network after the distributed photovoltaic grid is connected, which can predict the relevant risk points in advance, improve the management efficiency of the distribution network, and improve the reliability of the distribution network operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a flowchart of the overall steps of the distributed photovoltaic grid-connected distribution network risk assessment method in Example 1.

[0020] Figure 2 This is a flowchart of the overall steps of the distributed photovoltaic grid-connected distribution network risk assessment method in Example 2. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments: Embodiment 1: like Figure 1In the illustrated embodiment, a distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting can be seen, and its operation process is as follows: Step 1, analyzing the distributed photovoltaic output characteristics, performing time series acquisition of distribution network load data under different light intensities, and processing the collected load data; Step 2, using singular spectrum analysis to decompose the load data, establish a BP neural network, and predict the load after the distributed photovoltaic grid is connected; Step 3, based on the load forecast results, establish a distributed photovoltaic grid-connected power grid impact quantification model, and conduct a distributed photovoltaic grid-connected distribution network risk assessment.

[0022] The distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting provided in this embodiment analyzes the photovoltaic output characteristics during load forecasting, and uses seven singular spectrum analysis to reduce data dimensions and improve load forecasting accuracy. Based on the load forecast results, the distribution network risk assessment is carried out to predict relevant risk points in advance and improve the reliability of grid operation.

[0023] The technical scheme and technical effects of the present invention will be further described below through specific examples. The following examples are for explanation of the present invention and the present invention is not limited to the following examples.

[0024] Step 1: Analyze the output characteristics of distributed photovoltaics, collect the load data of the distribution network under different light intensities in a time series, and process the collected load data.

[0025] The output characteristics of distributed photovoltaics are related to sunlight. The higher the sunlight, the higher the output of the photovoltaic power generation system, and vice versa.

[0026] Based on the above analysis, this embodiment collects the load data of the distribution network under different illumination intensities and classifies the collected data. Specifically, the data collected at the same time are placed together, and the collected data are divided by date, and arranged and encoded in chronological order, and the load data and the illumination intensity corresponding to the load data are recorded.

[0027] Step 2: Use singular spectrum analysis to decompose the load data, establish a BP neural network, and predict the load after distributed photovoltaic grid connection.

[0028] Analyze the load characteristics of the current power grid and generate a load fluctuation signal; if the load fluctuation signal is a low fluctuation signal, it means that the current distribution network is operating relatively stably and no control is required, then the load is predicted and the risk of the distribution network in the future is assessed; if the load fluctuation signal is a high fluctuation signal, it means that the current distribution network is in an unstable operating state and a risk warning signal needs to be output immediately to control the power grid load.

[0029] The load data is subjected to singular value decomposition, and the noise sequence is deleted to obtain multiple subsequences; the BP neural network is used to perform load forecasting on each subsequence respectively, and the load forecasting results of each sequence are superimposed to obtain the total load forecast value.

[0030] The BP neural network includes three network layers, namely the input layer, the hidden layer (the hidden layer has one or more layers, each layer has several neurons, and a single neuron is composed of three parts: connection weight, summation unit and nonlinear activation function) and the output layer. The input of the BP neural network includes the current moment, the load data at the current moment, the light intensity at the current moment, and the light intensity at the next moment, and the output is the load data at the next moment.

[0031] In other embodiments, the input of the BP neural network may also be the daily load data of the past N days, the daily light intensity, and the light intensity of the predicted day, and the output is the load data of the predicted day.

[0032] Specifically: set the parameters of the BP neural network model, select the tansig function as the activation function, and initialize the weights and thresholds; use the input data to train the BP neural network model to obtain the predicted value, and use the deviation between the predicted value and the actual value to reversely correct the weights and bias of the BP neural network to complete the training of the BP neural network model.

[0033] When training the BP neural network, four indicators are selected, namely the daily maximum load prediction difference, the daily load prediction average difference, the daily load prediction root mean square error, and the daily minimum load prediction difference. The weights of the four indicators are determined as BP neural network evaluation indicators to evaluate the accuracy of BP neural network prediction, thereby determining the optimal BP neural network model.

[0034] Step 3: Based on the load forecast results, a quantitative model of the impact of distributed photovoltaic grid connection on the power grid is established to conduct a distribution network risk assessment of distributed photovoltaic grid connection.

[0035] In this embodiment, the quantification model of the impact of distributed photovoltaic grid-connected power grid is: the risk of the distribution network is equal to the product of the first load factor, the second load factor and the severity of the consequence, the first load factor is the integral of the load forecast result over the time interval T, the second load factor is the product of the factors affecting the load, and the severity of the consequence is the product of the importance of each device in the distribution network and the degree of equipment loss. The importance of the equipment is determined by the cost of the equipment.

[0036] If the calculated future distribution network risk value is greater than the preset threshold, it is necessary to prepare to control the power grid. If the future distribution network risk value is less than the preset threshold, the distribution network should be continuously monitored.

[0037] This embodiment establishes a three-phase load imbalance optimization model for a distributed photovoltaic energy storage distribution network that takes into account voltage over-limit risks, and uses the control of distributed photovoltaic grid-connected reactive power to suppress voltage over-limit risks. Based on the load forecast results, a photovoltaic grid-connected voltage impact quantification model is established, and the control of distributed photovoltaic grid-connected reactive power is considered to suppress voltage over-limit risks, so as to better adapt to the operation changes caused by "multi-source" grid connection in medium and low voltage distribution networks.

[0038] Embodiment 2: like Figure 2 In the illustrated embodiment, a distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting can be seen, and its operation process is as follows: Step 1, analyzing the distributed photovoltaic output characteristics, performing time series acquisition of distribution network load data under different light intensities, and processing the collected load data; Step 2, using singular spectrum analysis to decompose the load data, establish a BP neural network, and predict the load after the distributed photovoltaic grid is connected; Step 3, based on the load forecast results, establish a distributed photovoltaic grid-connected power grid impact quantification model, and conduct a distributed photovoltaic grid-connected distribution network risk assessment.

[0039] This embodiment evaluates the risk of the distribution network after the distributed photovoltaic grid is connected based on the load forecast results, which can predict the relevant risk points in advance, improve the management efficiency of the distribution network, and ensure the reliability of the distribution network during its operation. The reliability of the distribution network is conducive to the conversion of photovoltaic power, and the distributed photovoltaic power source will affect the reliability of the overall system. Therefore, the method provided in this embodiment is used to ensure the reliability of the power supply of the distribution network, and also provide better protection for the production and use of renewable energy.

[0040] The technical scheme and technical effects of the present invention will be further described below through specific examples. The following examples are explanations of the present invention and the present invention is not limited to the following examples.

[0041] Step 1: Analyze the output characteristics of distributed photovoltaics, collect the load data of the distribution network under different light intensities in a time series, and process the collected load data.

[0042] The historical data of the distribution network is collected, and the historical data is the data obtained by time-series collection of distribution network load data under different light intensities.

[0043] The collected load data is preprocessed, including: processing abnormal values ​​and missing values, and normalizing the collected historical load data.

[0044] Step 2: Use singular spectrum analysis to decompose the load data, establish a BP neural network, and predict the load after distributed photovoltaic grid connection.

[0045] Determine the current moment and the predicted moment, and use the load fluctuation at the same moment and the same light intensity in the historical data as the reference value. If the difference between the load prediction value and the load fluctuation of the current moment and the reference value is greater than the preset threshold, the current distribution network is directly determined to be high risk, and a distribution network risk warning signal is output to control the distribution network load. If the difference between the load prediction value and the load fluctuation of the current moment and the reference value is less than the preset threshold, the current distribution network is low risk, and the subsequent steps are continued.

[0046] The specific steps of using singular spectrum analysis to decompose load data and establish a BP neural network to predict the load after distributed photovoltaic grid connection are the same as those in the first embodiment.

[0047] Step 3: Based on the load forecast results, a quantitative model of the impact of distributed photovoltaic grid connection on the power grid is established to conduct a distribution network risk assessment of distributed photovoltaic grid connection.

[0048] This embodiment uses the random forest algorithm to establish a quantified model of the impact of distributed photovoltaic grid-connected power grids to assess the risk of voltage over-limit in the distribution network: 1. Establish a data set, which includes voltage out-of-bounds type, historical load data (predicted load data when using the model for distribution network risk assessment), distribution transformer rated capacity, feeder current carrying capacity, voltage value, etc., among which historical load data includes load mean, maximum load change value, feeder load mean, etc. Perform sample imbalance processing and standardization on the data set. Randomly select 80% of the samples from the data set as the training data set, and the remaining 20% ​​of the samples as the test data set.

[0049] 2. Select the base learner, determine the number of decision trees to be constructed, and train the random forest model for the voltage over-limit risk of the distribution network in different time periods in the future; evaluate the obtained random forest model.

[0050] Among them, the random forest models of distribution network over-limit risks in different time periods in the future include: random forest model of distribution network over-limit risks in the next N hours, random forest model of distribution network over-limit risks in the next day, random forest model of distribution network over-limit risks in the next two days, random forest model of distribution network over-limit risks in the next 3 days, random forest model of distribution network over-limit risks in the next 15 days, etc.

[0051] The input of the random forest model is historical load data (predicted load data when the model is used for distribution network risk assessment), distribution transformer rated capacity, feeder current carrying capacity, voltage value, etc. The historical load data includes load mean, maximum load change value, feeder load mean, and the output is voltage out-of-bounds type.

[0052] Using the resampling method, a group of samples are extracted from the training data set, and then replaced after extraction for m consecutive times to obtain m groups of samples. Based on the m groups of samples, m decision trees are established to obtain m output results.

[0053] The importance of variables in each random forest model is calculated to determine the impact of each variable on the voltage over-limit of the distribution network.

[0054] According to the prediction results of the random forest model (distribution network low voltage exceeding the limit or distribution network high voltage exceeding the limit), formulate the corresponding distribution network control strategy.

[0055] Embodiment three: This embodiment provides a distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting, and brings in a specific application scenario to execute the distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting. Any work must focus on safety protection in the early stage of the work, and the same is true for the operation of the distribution network. The distribution network must focus on safe operation during operation. Safety is the basis for all work. Failure to pay attention to safety will cause adverse effects on the operation of the distribution network. It may cause certain economic losses and even affect the safety of life and property of the staff. In addition, the photovoltaic power source itself requires a certain amount of time in the process of conversion. The distributed photovoltaic power source has its own characteristics during the access process, which can change the power flow direction within the power system. When the power supply is accessed, the stability of the access cannot be ensured. Therefore, the stability and safety of the power system must be ensured. If the relay protection equipment is damaged, it will affect the normal operation of the entire power system.

[0056] However, distributed photovoltaics will generate certain pollution during the access process, which is collectively called harmonic pollution. Since the type of current changes from the original direct current to alternating current, it is easy to generate harmonics during the change process. When a large number of photovoltaic power sources are connected, the number of harmonics will continue to increase, causing the frequency of harmonics to increase in amplitude. The generated harmonic pollution will not only affect the stability of the distribution network, but also generate certain radiation, affecting people's physical safety.

[0057] The large-scale access of distributed photovoltaics will also affect the reliability of power supply in the distribution network. Distributed photovoltaic power sources are not accessed step by step and in an orderly manner during the access process. They are extremely unstable. The occurrence of network losses will not only affect the internal technical indicators, but also the development of the industry economy.

[0058] First of all, a unified design can be carried out in the early stage of the power network. The design should be carried out according to the development needs of related industries, and then a series of plans can be carried out to ensure reasonable operation and reasonable control of some inevitable factors. If the development of the low-voltage distribution network cannot meet the needs of social development in the new stage, the network structure will be continuously upgraded and the original operation plan will be optimized so that users can have a better experience in the process of use and participation. In this way, it can not only better meet the needs of the development of the power market, but also make full use of renewable energy and improve people's quality of life. The large-scale access of distributed photovoltaics can better improve the planning and design of the distribution network, and make the operation more scientific and reasonable, and also reduce the risks of power production and transmission.

[0059] On the other hand, after the distributed photovoltaic grid is connected, a distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting provided in this embodiment is adopted. By analyzing the distributed photovoltaic output characteristics, the distribution network load data under different light intensities are collected in time series, and the collected load data is processed; the load data is decomposed by singular spectrum analysis, and a BP neural network is established to predict the load after the distributed photovoltaic grid is connected; based on the load forecasting results, a distributed photovoltaic grid-connected power grid impact quantitative model is established, and finally the distributed photovoltaic grid-connected distribution network risk assessment is performed. According to the risk assessment results, the distribution network is managed and controlled.

[0060] During this process, the load characteristics of the current power grid are analyzed to generate a load fluctuation signal; if the load fluctuation signal is a low fluctuation signal, the load is predicted; if the load fluctuation signal is a high fluctuation signal, a risk warning signal is immediately output to control the distribution network load.

[0061] When controlling the load of the power grid, first of all, the quality of electric energy plays a very important role in the distribution network. The intervention of distributed photovoltaics will affect the voltage distribution of the distribution network. The access of distributed photovoltaics will change the original flow direction, causing large fluctuations in the distribution network and causing local voltage increases or decreases. Therefore, the load of the distribution network can be controlled by reasonably controlling the voltage.

[0062] Specifically, the photovoltaic power inverter can be used. The photovoltaic power inverter can better collect shock waves and improve the power quality of the power grid. When the energy storage device is combined with the photovoltaic power supply, the voltage of the distribution network can be reasonably controlled, which can provide better help for the regulation frequency of the grid voltage. For the voltage imbalance phenomenon that occurs during the normal operation of the distribution network, the photovoltaic power inverter can be used to effectively divide the current and distribute the voltage to some unevenly distributed areas.

[0063] This embodiment adopts a distributed photovoltaic grid-connected distribution network risk assessment method based on load forecasting, and establishes a three-phase load imbalance optimization model for a distributed photovoltaic energy storage distribution network that takes into account voltage over-limit risks. Based on the load forecast results, a photovoltaic grid-connected voltage impact quantification model is established, and the control of distributed photovoltaic grid-connected reactive power is considered to suppress the voltage over-limit risk, so as to better adapt to the operation changes caused by "multi-source" grid connection in medium and low voltage distribution networks.

[0064] Accurate load forecasting can not only play an important role in the stable operation and safety of the power grid, but also provide guidance for the future planning and construction of the power grid.

[0065] At the same time, with the promotion and application of distributed photovoltaic power sources, we know that more and more photovoltaic power sources are connected to the power grid. However, due to the intermittent and random characteristics of distributed photovoltaic power sources, node load prediction has become a difficult problem that needs to be solved after the large-scale access of photovoltaic power sources. Through risk assessment, we can pay attention to the safety of the distribution network in a timely and effective manner, so that the operation of the distribution network can be better guaranteed. When a large number of distributed power sources are connected to the distribution network, the influence of various factors should be reasonably considered so that emergency measures can be taken when problems arise. While providing a safe operating environment for the distribution network, the electric power should also be effectively distributed.

[0066] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A distributed photovoltaic grid-connected distribution network risk assessment method, characterized in that: include: S1: Analyze the output characteristics of distributed photovoltaics, collect the load data of the distribution network under different light intensities in a time series, and process the collected load data; S2: Use singular spectrum analysis to decompose load data, establish BP neural network, and predict the load after distributed photovoltaic grid connection; S3: Based on the load forecast results, a quantitative model of the impact of distributed photovoltaic grid connection on the power grid is established to conduct a risk assessment of the distribution network of distributed photovoltaic grid connection.

2. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, characterized in that: The load characteristics of the power grid at the current moment are analyzed to generate a load fluctuation signal; if the load fluctuation signal is a low fluctuation signal, the load is predicted; if the load fluctuation signal is a high fluctuation signal, a risk warning signal is immediately output to control the power grid load.

3. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, characterized in that: The decomposition of load data by singular spectrum analysis includes: performing singular value decomposition on the load data, deleting the noise sequence, and obtaining a plurality of subsequences; using BP neural network to perform load forecasting on each subsequence respectively, superimposing the load forecasting results of each sequence, and obtaining the load forecasting value.

4. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, 2 or 3, characterized in that: The quantification model of the impact of distributed photovoltaic grid-connected power grid is: the risk of the distribution network is equal to the product of the first load factor, the second load factor and the severity of the consequence, the first load factor is the integral of the load forecast result over the time interval T, the second load factor is the product of the factors affecting the load, and the severity of the consequence is the product of the importance of each device in the distribution network and the degree of equipment loss.

5. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, 2 or 3, characterized in that: The step S3 includes using a random forest algorithm to establish a quantitative model of the impact of distributed photovoltaic grid-connected power grids and perform a distribution network voltage over-limit risk assessment: establishing a data set, the data set including historical load data; selecting a base learner, determining the number of decision trees to be constructed, and training a distribution network voltage over-limit risk random forest model for different time periods in the future; and evaluating the obtained random forest model.

6. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, 2 or 3, characterized in that: Determine the current moment and the predicted moment, and use the load fluctuation of the same period and the same light intensity in the historical data as the reference value. If the difference between the load prediction value and the load fluctuation of the current moment and the reference value is greater than the preset threshold, the current distribution network is directly determined to be high risk. If the difference between the load prediction value and the load fluctuation of the current moment and the reference value is less than the preset threshold, the current distribution network is low risk and the subsequent steps are continued.

7. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 5, characterized in that: The data set samples are subjected to unbalanced processing and standardized processing, wherein the unbalanced processing includes under-sampling the normal voltage data and over-sampling the low voltage over-limit and high voltage over-limit voltage data.

8. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 1, 2 or 3, characterized in that: The risk of voltage exceeding the limit is suppressed by controlling the reactive power of distributed photovoltaic grid-connected power.

9. A method for risk assessment of distributed photovoltaic grid-connected distribution network based on load forecasting according to claim 1, 2 or 3, characterized in that: The time series acquisition includes placing the data collected at the same time together, dividing the collected data according to date, and arranging and encoding them in chronological order.

10. A distributed photovoltaic grid-connected distribution network risk assessment method according to claim 5, characterized in that: The importance of variables in each random forest model is calculated to determine the impact of each variable on the voltage over-limit of the distribution network.

Images