A data-driven power distribution network photovoltaic power generation receiving capacity analysis method and device

Abstract

The application discloses a kind of power distribution network photovoltaic power generation receiving capacity analysis method and device.First, the historical data set of each node power, voltage of power distribution network is collected, and historical data set is randomly divided into training set and test set;Second, based on DELM, historical data set is trained and tested, and DELM network parameter is obtained;On this basis, SSA is used to optimize DELM network parameter, and the final DELM network parameter is obtained;According to the established DELM model, the power-voltage of photovoltaic node is fitted using nonlinear function;Finally, according to the nonlinear function of the power-voltage of photovoltaic node established, the photovoltaic power generation receiving capacity of the system under the node voltage safety constraint is extrapolated.The application models power flow analysis of power distribution network according to historical data, without the topology and parameter information of power distribution network;According to the nonlinear fitting method of node power-voltage obtained by the established DELM model, the future scene has high prediction ability.

Description

Technical Field

[0001] This invention belongs to the field of power system situational awareness, and in particular relates to a data-driven analysis of the photovoltaic power generation capacity of distribution networks. Background Technology

[0002] Power flow calculation is a prerequisite for studying the renewable energy integration capacity of distribution networks. Currently, methods for solving renewable energy integration capacity mainly include deterministic methods, probabilistic methods, and data-driven methods. Since deterministic and probabilistic methods are based on power flow models of distribution networks, they require accurate distribution network topology and parameters. However, real-world systems often suffer from inaccurate or unmeasurable basic topology and parameters, making distributed generation integration capacity analysis based on mathematical models difficult to apply to practical systems. Currently, widely used smart meters provide measurement information such as node voltage and injected power in the distribution network. This massive amount of real-time measurement information provides a data foundation for measurement modeling in distribution network power flow calculation.

[0003] Existing data-driven distribution network power flow calculation methods suffer from drawbacks such as high computational complexity and high space complexity due to the need to store all training samples, which affects the efficiency of model training when the amount of historical distribution network data is large. Current data-driven photovoltaic (PV) power integration capacity analysis often relies on extrapolation based on the linear relationship between node power and voltage, resulting in PV power integration capacity values ​​with large errors. Summary of the Invention

[0004] Purpose of the invention: The purpose of this invention is to provide a data-driven method and apparatus for analyzing the photovoltaic power generation capacity of distribution networks, which has a fast learning speed, strong generalization ability, and can be better applied to power flow modeling and situational awareness in distribution networks.

[0005] Technical solution: The data-driven distribution network photovoltaic power generation access capability analysis method of the present invention follows the steps of training to obtain a DELM network and applying the DELM network to obtain the photovoltaic power generation access limit value of each node of the target distribution network under the voltage security constraint at the target time.

[0006] Step 1: Collect historical data of each node of the target distribution network in a preset historical time period. The historical data includes active power, reactive power and voltage amplitude. Divide the historical data into training set and test set according to the ratio.

[0007] Step 2: Use the active and reactive power of each node in the training set as input and the voltage amplitude of each node as output to train the DELM network. Optimize the DELM network using the Sparrow Search Algorithm (SSA). Verify the accuracy of the DELM network using the test set. If the DELM network meets the preset accuracy, the DELM network training is complete, and proceed to Step 3. If the DELM network does not meet the preset accuracy, repeat Step 2.

[0008] Step 3: Based on the trained DELM network, obtain the photovoltaic node power-to-voltage sequence of the target distribution network, and fit the sequence with a nonlinear function to obtain the nonlinear function expression;

[0009] Step 4: Based on the obtained nonlinear function expression, obtain the photovoltaic power generation access limit value of the target distribution network under voltage security constraints.

[0010] Furthermore, step 1 specifically includes: collecting historical data including the active power, reactive power, and voltage amplitude of each node, taking the active power, reactive power, and voltage amplitude of all nodes as a set of samples, where active power and reactive power are used as data-driven inputs and voltage amplitude is the output, and randomly selecting 85% of all collected samples as the training set and 15% as the test set.

[0011] Furthermore, in step 2, training the DELM network specifically includes: using active power and reactive power as input samples X of the DELM, and voltage amplitude as output samples Y of the DELM. The DELM input samples X are used to obtain the first weight matrix w1 and threshold matrix b1 according to ELM-AE theory, and then the feature vector H1 of the hidden layer is obtained; thus, the input weight matrix w1 of the M layer is finally obtained. M and eigenvector H M The feature vector H of each layer can be expressed as:

[0012] H = G(wX + b) (1)

[0013] In the formula: G is the activation function; w is the weight from each node in the input layer to each node in the hidden layer; b is the threshold matrix for each node in the hidden layer;

[0014] The model obtains the parameters of the DELM network by setting the number of neurons in each hidden layer. The DELM network is set to have 3 hidden layers, with 150 neurons in each layer.

[0015] Furthermore, in step 2, optimizing the DELM network using the Sparrow Search Algorithm (SSA) specifically includes: optimizing the DELM network using the SSA algorithm, where the parameters to be optimized are the input weights w1 and threshold matrix b1 of the first hidden layer of the DELM network in step 2, denoted as a = [w1, b1]. T Several groups of weights w1 and thresholds b1 are randomly assigned to form a population A. The sparrow search algorithm is a discoverer-follower-guarder model, where the position of each individual corresponds to a solution. The position of the discoverer is updated as follows:

[0016]

[0017] Where \(t\) is the current iteration number, iter max is the set maximum number of iterations, is a random number within \([0, 1]\), \(Q\) is a random number from a normal distribution, \(R\) is a uniform random number within the \([0, 1]\) interval, serving as the warning value; \(ST\) is the set safety value. When \(R < ST\), it indicates that the environment is safe and the individual moves to a position near the current location; when \(R\geq ST\), it indicates that the environment is dangerous and the individual randomly leaves the current location;

[0018] The method for updating the position of the followers in the population is as follows:

[0019]

[0020] Where \(A\) best and \(A\) worst are respectively the optimal position and the worst position occupied by the discoverer; \(N\) is the total number of individuals; \(A\) i is sorted by fitness value. If \(i > N / 2\), it is considered that the \(i\)-th follower has a poor fitness value and its position is randomly changed. Otherwise, the follower moves near the optimal position;

[0021] The position of the early warning individuals in the population is randomly generated, accounting for 10%-20% of the total. Their position description is as follows:

[0022]

[0023] Where \(\beta\) is the step size control parameter; \(K\) is a random number within the \([-1, 1]\) interval; \(f\) i is the fitness value of the current individual, \(f\) g and \(f\) w are respectively the optimal and worst fitness values; \(\varepsilon\) is a constant. When this early warning individual is at the optimal position, it will move to other positions near itself. Otherwise, it will move towards the optimal position;

[0024] The fitness function \(f\) selects the mean square error of the test samples after training, that is:

[0025]

[0026] where \(m\) is the number of samples participating in training; \(n\) is the number of individuals within the sample; \(y\) ij is the actual value, is the predicted value. The smaller \(f\) is, the higher the prediction accuracy. The optimization termination condition is that the fitness reaches the specified value or the number of iterations reaches 50 times.

[0027] Furthermore, step 3 specifically includes:

[0028] The power of the photovoltaic power supply access point is perturbed, and the voltage of each node in the system is obtained according to the DELM network optimized by SSA. The photovoltaic output power of the node is continuously increased within the historical data range to obtain the photovoltaic node power-voltage sequence of each node. Based on this sequence, a fourth-order nonlinear function fitting is used to obtain the nonlinear function expression of the photovoltaic node's power-voltage sensitivity to each node.

[0029] The fourth-order nonlinear function is expressed as:

[0030]

[0031] Where i represents the i-th node, P PV Let be the photovoltaic power generation capacity, and a, b, c, d, e be constant coefficients.

[0032] Furthermore, step 4 specifically includes: solving the photovoltaic access limit of the target distribution network under voltage security constraints when photovoltaic power sources are connected at single and multiple points, wherein:

[0033] When a photovoltaic power source is connected to a single point, the photovoltaic power is continuously increased based on the fourth-order nonlinear function of photovoltaic power versus node voltage obtained by fitting. The photovoltaic power limit under the corresponding node voltage safety is obtained with the node whose voltage first exceeds the safety threshold as the benchmark.

[0034] When photovoltaic power sources are connected at multiple points, the solution steps are as follows:

[0035] (1) Obtain the fourth-order function fitting curve of photovoltaic output power versus node voltage sensitivity at each photovoltaic power source access point;

[0036] (2) Sort the sensitivity by magnitude;

[0037] (3) Install photovoltaic power at the node with the lowest sensitivity according to the maximum physical limit set for that node, and check whether the voltage of each node in the distribution network exceeds the limit; if it does not exceed the limit, install photovoltaic power at the node with the second lowest sensitivity according to the maximum limit, and check whether the voltage of each node in the distribution network exceeds the limit; and so on. When the installation of photovoltaic power at a node causes the voltage of a node in the system to exceed the limit, the photovoltaic power that can be installed is determined based on the node whose voltage exceeds the limit first.

[0038] (4) The photovoltaic power of each node is added together to get the maximum photovoltaic power that the system can accept.

[0039] Furthermore, the voltage safety threshold is 1.07U. n .

[0040] This invention also discloses a data-driven distribution network photovoltaic power generation acceptance capacity analysis device, comprising:

[0041] Acquisition Unit: Includes acquiring historical data of active power, reactive power, and voltage amplitude of each node, and dividing the historical data into training set and test set;

[0042] DELM Network Training Unit: The active and reactive power of historical data are used as input data, and the voltage amplitude is used as output data. The number of hidden layers, the number of neurons in each layer, and the activation function of the DELM network are set to train the DELM network and obtain network parameters. The SSA method is used to optimize the DELM network and obtain the optimized DELM network parameters.

[0043] Fitting Unit: Based on the DELM network obtained through optimized training, the power perturbation of photovoltaic nodes is continuously increased to obtain the power-voltage sequence of each photovoltaic node, and the sequence is fitted using a fourth-order nonlinear function;

[0044] Calculation unit: Sets voltage safety threshold and obtains the access limit of single-point or multi-point photovoltaic power sources based on the installation location of the photovoltaic power source.

[0045] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0046] (1) This invention models the power flow analysis of the distribution network based on historical data, without requiring the topology and parameter information of the distribution network;

[0047] (2) SSA was used to optimize DELM parameters, which improved the performance of DELM network for power flow analysis of distribution network;

[0048] (3) The node power-voltage nonlinear fitting method obtained from the established DELM model has a high predictive ability for future scenarios and provides guidance on installation schemes for the large-scale promotion of photovoltaic power in distribution networks. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the data-driven method for analyzing the photovoltaic power generation capacity of distribution networks according to the present invention.

[0050] Figure 2 This is a topology diagram of a power distribution network area in Hebei Province.

[0051] Figure 3 This is a diagram of the DELM network structure.

[0052] Figure 4 This is the power-voltage sensitivity curve. Detailed Implementation

[0053] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0054] Example 1

[0055] The implementation process is attached. Figure 1 .

[0056] S1. Collect historical data on power and voltage of distribution network nodes, and randomly divide the historical data into 85% training set and 15% test set.

[0057] The examples are attached. Figure 2 The example shown is from a distribution network area in Hebei Province. The rated capacity of the transformer in this area is 300kVA. The distribution network has a total of 40 nodes, of which 20 are load nodes and the rest are intermediate nodes. The load information of each node is shown in Table 1.

[0058] Table 1 Load Characteristics and Design Power

[0059]

[0060] The collected historical data consisted of active power P, reactive power Q, and voltage amplitude at each node for 96 points (sampling interval of 15 minutes) throughout 2021 (365 years).

[0061] The {active power, reactive power, voltage amplitude} of all nodes are treated as a set of samples, where {active power, reactive power} are the data-driven inputs and {voltage amplitude} is the output. 85% of the collected samples are randomly selected as the training set, and 15% as the test set.

[0062] S2, use historical data to train a Deep Extreme Learning Machine (DELM) network, and obtain the parameters of the DELM network accordingly.

[0063] The DELM uses {active power, reactive power} as input samples X and {voltage amplitude} as output samples Y. The DELM has 3 hidden layers, each containing 150 neurons. Training... Figure 3 The DELM network shown is trained as follows:

[0064] The DELM input sample X is used to obtain the first weight matrix w1 and threshold matrix b1 according to the ELM-AE theory, followed by the feature vector H1 of the hidden layer; and so on, finally obtaining the input weight matrix w of the M layer. M and eigenvector H M The feature vector H of each layer can be represented as:

[0065] H = G(wX + b) (1)

[0066] In the formula: G is the activation function; w is the weight from each node in the input layer to each node in the hidden layer; b is the threshold matrix for each node in the hidden layer.

[0067] S3. The first hidden layer input weights and thresholds of the DELM network are optimized using the Sparrow Search Algorithm (SSA) to obtain the final DELM network parameters.

[0068] The population size of SSA is 20, the number of iterations is 50, the safety value is set to 0.6, the proportion of warning individuals is set to 20%, and the proportion of discoverers is set to 70%.

[0069] When using the SSA algorithm to optimize the DELM network, the parameters to be optimized are the input weights w1 of the first hidden layer of the DELM network and the threshold matrix b1 in step S2, denoted as a = [w1, b1]. T . A number of groups of weights w1 and thresholds b1 are randomly set to form a population A. The Sparrow Search Algorithm is a discoverer-follower-warning model, and the position of each individual corresponds to a solution. The update method for the position of the discoverer is as follows:

[0070]

[0071] In the formula, t is the current iteration number, iter max is the set maximum iteration number, is a random number within [0, 1], Q is a random number from a normal distribution, R is a uniform random number within the interval [0, 1], serving as the warning value; ST is the set safety value. When R < ST, it indicates that the environment is safe and the individual moves to the vicinity of the current position; when R ≥ ST, it indicates that the environment is dangerous and the individual will randomly leave the current position.

[0072] The update method for the position of the followers in the population is as follows:

[0073]

[0074] In the formula, A best and A worst are respectively the optimal position and the worst position occupied by the discoverers; N is the total number of individuals; A i is sorted by fitness value. If i > N / 2, it is considered that the fitness value of the i-th follower is poor, and the position is randomly changed. Otherwise, the follower moves to the vicinity of the optimal position.

[0075] The position of the warning individuals in the population is randomly generated, accounting for 10% - 20% of the total. Their position description is as follows:

[0076]

[0077] In the formula, β is the step size control parameter; K is a random number within the interval [-1, 1]; f i is the fitness value of the current individual, f g and f wThese represent the optimal and worst fitness values, respectively; ε is a constant. When the alerter is in the optimal position, it will move to another nearby position; otherwise, it will move closer to the optimal position.

[0078] The fitness function f is selected from the mean squared error of the test samples after training, that is:

[0079]

[0080] Where m is the number of samples participating in the training; n is the number of individuals in the sample; y ij This is the actual value. The value is the prediction value; a smaller f indicates higher prediction accuracy. The optimization termination condition is that the fitness reaches a specified value or the number of iterations reaches 50.

[0081] S4, the power of the photovoltaic power supply access point is perturbed, and the voltage of each node in the system is obtained according to the DELM network optimized by SSA; the photovoltaic output power of the node is continuously increased within the historical data range to obtain the photovoltaic node power-node voltage sequence. Based on this sequence, a fourth-order nonlinear function fitting is used to obtain the nonlinear function expression of the photovoltaic node's power-voltage sensitivity to each node.

[0082] Assuming that nodes 1-20 are connected to photovoltaic power sources, the nonlinear expressions for photovoltaic power versus corresponding node voltage sensitivity are obtained based on the established SSA-DELM model.

[0083] S5. Based on the nonlinear function of node power-voltage obtained by fitting, extrapolate the node power-voltage sensitivity, and solve the system photovoltaic access limit under voltage safety constraints when connecting photovoltaic devices at single and multiple points.

[0084] When a photovoltaic (PV) power source is connected at a single point, the PV power is continuously increased based on the fourth-order nonlinear function of PV power versus node voltage obtained from the fitting algorithm, until the voltage exceeds the limit first (set the threshold to be 1.07U). n Using the nodes as a reference, the photovoltaic power limit under the corresponding voltage is obtained.

[0085] Assume that nodes 1-20 are connected to photovoltaic power sources, and the photovoltaic acceptance capacity of each node is shown in Table 2.

[0086] Table 2. Regional Absorption Capacity of a Single Node When Connected to Photovoltaic Power Supply

[0087]

[0088] When photovoltaic power is connected to multiple points, the solution steps are as follows: (1) Obtain the fourth-order function fitting curve of photovoltaic output power - node voltage sensitivity of each photovoltaic access node; (2) Sort the sensitivity by size; (3) Install photovoltaic power at the node with the smallest sensitivity according to the maximum physical limit set for that node, and check whether the voltage of each node in the system exceeds the limit; if it does not exceed the limit, install photovoltaic power at the node with the second smallest sensitivity according to the maximum limit, and check whether the voltage of each node in the system exceeds the limit; and so on. When the installation of photovoltaic power at a certain node causes the node voltage in the system to exceed the limit, the upper limit of the first node voltage to exceed the limit is used as the benchmark to obtain the photovoltaic power that can be installed; (4) Add the photovoltaic power of each node to get the maximum photovoltaic value that the system can accept.

[0089] When multiple nodes are connected to photovoltaic power sources, for ease of solution, Figure 3 The power grid topology of the distribution area is artificially divided into four branches: Branch 1 contains nodes {1,3,4,5,10,11}, Branch 2 contains nodes {2,6,7}, Branch 3 contains nodes {8,12,14,17,18,19,20}, and Branch 4 contains nodes {9,13,15,16}. Assume that nodes {1,3} in Branch 1, {2,6} in Branch 2, {8,12} in Branch 3, and {9,13} in Branch 4 are connected to photovoltaic (PV) power sources. Due to physical limitations, the physical capacity of PV power sources that can be connected to each node is limited to 40kW for residential load nodes and 80kW for commercial load nodes. Using the aforementioned multi-node capacity calculation method, taking node {1,3} connected to PV power sources as an example, see the attached diagram. Figure 4 P is given respectively PV1 -U1、P PV1 -U3、P PV3 -U1 and P PV3 -U3 sensitivity fitting curve.

[0090] First, install photovoltaic (PV) power at node 1, which has the lowest sensitivity, up to its upper limit of 40kW. Then, install PV power at node 3, which has the highest sensitivity. When the PV power reaches 13.38kW, the voltage at node 1 exceeds the limit earlier than that at node 3. Therefore, the upper limit of PV power for branch 1 is 53.38(40+13.38)kW.

[0091] The photovoltaic (PV) integration capacity of other branches, calculated using this method, is 66.95 kW for branch 2, 74.04 kW for branch 3, and 104.45 kW for branch 4. According to Table 1, the sum of the design power of each branch node is less than or close to the integration limit of that branch. Therefore, it can be considered that when PV power is connected to nodes {1,3} of branch 1, {2,6} of branch 2, {8,12} of branch 3, and {9,13} of branch 4, the power is fed back to the grid. The system's PV power integration capacity is 53.38 + 66.95 + 74.04 + 104.45 = 298.82 kW.

[0092] Example 2

[0093] This invention provides a data-driven distribution network photovoltaic power generation acceptance capacity analysis device, comprising:

[0094] Acquisition Unit: This includes acquiring historical data on active power, reactive power, and voltage amplitude of each node, and randomly dividing the historical data into an 85% training set and a 15% test set.

[0095] The DELM network training unit uses historical data (active power and reactive power) as input data and voltage amplitude as output data. It sets the number of hidden layers, the number of neurons in each layer, and the activation function to train the DELM network and obtain its parameters. The SSA algorithm is then used to optimize the first-layer input weights w1 and the threshold matrix b1, using mean squared error as the fitness function to determine the final DELM network parameters.

[0096] Fitting Unit: Based on the DELM network obtained through optimized training, the power perturbation of photovoltaic nodes is continuously increased to obtain the power-voltage sequence of each photovoltaic node, and the sequence is fitted using a fourth-order nonlinear function;

[0097] Calculation unit: Using 1.07 times the rated voltage as a threshold, the grid connection limit of single-point or multi-point photovoltaic power sources is obtained based on the installation location of the photovoltaic power source.

[0098] For specific implementation schemes of each module of the device of the present invention, please refer to the specific implementation process of the above method.

[0099] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0101] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0102] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0103] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A data-driven method for analyzing the photovoltaic power generation acceptance capacity of a distribution network, characterized in that, The following steps are followed to train and obtain a DELM network, and then the DELM network is applied to obtain the photovoltaic power generation access limit value under the target distribution network voltage security constraint. Step 1: Collect historical data of each node of the target distribution network in a preset historical time period. The historical data includes the active power, reactive power and voltage amplitude of the node, forming a sample set, and then divide the sample set into a training set and a test set. Step 2: Use the active power and reactive power of each node in the training set as input and the voltage amplitude of each node as output to train the DELM network. During the training process, the Sparrow Search Algorithm (SSA) is used to optimize the DELM network. The accuracy of the DELM network is verified through the test set. If the DELM network meets the preset accuracy, the DELM network training is complete. Proceed to Step 3. If the DELM network does not meet the preset accuracy, continue training until the preset accuracy is met; Step 3: Based on the trained DELM network, obtain the photovoltaic node power-voltage sequence of the target distribution network, and fit the sequence with a nonlinear function to obtain the nonlinear function expression; Step 3 specifically includes: The power of the photovoltaic power supply access point is perturbed, and the voltage of each node in the system is obtained according to the DELM network obtained by SSA optimization. The photovoltaic output power of the node is continuously increased within the historical data range to obtain the photovoltaic node power-voltage sequence of each node. Based on this sequence, a fourth-order nonlinear function fitting is used to obtain the nonlinear function expression of the photovoltaic node's power-voltage sensitivity to each node. The fourth-order nonlinear function is expressed as: (6) Where i is the i-th point, P PV Let a, b, c, d, and e represent the photovoltaic power generation capacity, and a, b, c, d, and e be constant coefficients. Step 4: Based on the obtained nonlinear function expression, obtain the photovoltaic power generation access limit value of the target distribution network under voltage security constraints; Step 4 specifically includes: solving the photovoltaic access limit value of the target distribution network voltage under safety constraints when connecting photovoltaic power sources at single and multiple points, wherein: When a photovoltaic power source is connected to a single point, the photovoltaic power is continuously increased based on the fourth-order nonlinear function of photovoltaic power versus node voltage obtained by fitting. The photovoltaic power is then used as the reference for the node whose voltage first exceeds the safety threshold, and the photovoltaic power limit value under the corresponding node voltage safety is obtained. When photovoltaic power sources are connected at multiple points, the solution steps are as follows: (1) Obtain the fourth-order function fitting curve of photovoltaic output power-node voltage sensitivity at each photovoltaic power source access point; (2) Sort the sensitivities by magnitude; (3) Install photovoltaic power at the node with the lowest sensitivity according to the maximum physical limit set for that node, and check whether the voltage of each node in the target distribution network exceeds the limit; if it does not exceed the limit, install photovoltaic power at the node with the second lowest sensitivity according to the maximum limit, and check whether the voltage of each node in the system exceeds the limit; and so on. When the installation of photovoltaic power at a node causes the voltage of the node in the system to exceed the limit, the upper limit of the voltage of the first node to exceed the limit is used as the benchmark to obtain the photovoltaic power that can be installed. (4) The photovoltaic power of each node above is added together to get the maximum photovoltaic power that the system can accept.

2. The data-driven method for analyzing the photovoltaic power generation capacity of a distribution network according to claim 1, characterized in that, Step 1 specifically includes: collecting historical data including the active power, reactive power and voltage amplitude of each node, taking the active power, reactive power and voltage amplitude of all nodes as a set of samples, where active power and reactive power are used as data-driven inputs and voltage amplitude is the output, and randomly selecting 85% of all collected samples as the training set and 15% as the test set.

3. The data-driven method for analyzing the photovoltaic power generation capacity of a distribution network according to claim 1, characterized in that, In step 2, training the DELM network specifically includes: using active power and reactive power as input samples X of the DELM, and voltage amplitude as output samples Y of the DELM. The DELM input samples X are used to obtain the first weight matrix w1 and threshold matrix b1 according to ELM-AE theory, followed by obtaining the feature vector H1 of the hidden layer; thus, the input weight matrix w1 of the M layer is finally obtained. M and eigenvector H M The feature vector H of each layer is represented as: (1) In the formula: G is the activation function; w is the weight from each node in the input layer to each node in the hidden layer; b is the threshold matrix for each node in the hidden layer; The model obtains the parameters of the DELM network by setting the number of neurons in each hidden layer. The DELM network is set to have 3 hidden layers, with 150 neurons in each layer.

4. The data-driven method for analyzing the photovoltaic power generation capacity of a distribution network according to claim 3, characterized in that, In step 2, the Sparrow Search Algorithm (SSA) is used to optimize the DELM network. Specifically, this includes optimizing the DELM network using the SSA algorithm, where the parameters to be optimized are the input weights w1 and threshold matrix b1 of the first hidden layer of the DELM network in step 2, denoted as a = [w1, b1]. T Several groups of weights w1 and thresholds b1 are randomly assigned to form a population A. The sparrow search algorithm is a discoverer-follower-guarder model, where the position of each individual corresponds to a solution. The position of the discoverer is updated as follows: (2) where \(t\) is the current iteration number, iter max is the set maximum number of iterations, \(\varphi\) is a random number within \([0, 1]\), \(Q\) is a random number from a normal distribution, \(R\) is a uniform random number within the interval \([0, 1]\) and serves as the warning value; \(ST\) is the set safety value. When \(R < ST\), it indicates that the environment is safe and the individual moves to the vicinity of the current position; when \(R\geq ST\), it indicates that the environment is dangerous and the individual randomly leaves the current position; The method for updating the position of followers in a population is as follows: (3) In the formula, A best and A worst These represent the best and worst positions occupied by the discoverer, respectively; N is the total number of individuals; A i Sort by fitness value. If i>N / 2, the i-th follower is considered to have poor fitness value and its position is randomly changed. Otherwise, the follower moves to the vicinity of the optimal position. The location of early warning individuals within the population is randomly generated, accounting for 10%-20% of the total population, and their locations are described as follows: (4) In the formula, β is the step size control parameter; K is a random number in the interval [-1, 1]; f i f is the current individual fitness value. g with f w These are the optimal and worst fitness values, respectively. The value is a constant. When the alerter is in the optimal position, it will move to another position near itself; otherwise, it will move closer to the optimal position. The fitness function f is selected from the mean squared error of the test samples after training, that is: (5) Where m is the number of samples participating in the training; n is the number of individuals in the sample; This is the actual value. The smaller the value of f, the higher the prediction accuracy. The optimization termination condition is that the fitness reaches a specified value or the number of iterations reaches 50.

5. The data-driven method for analyzing the photovoltaic power generation capacity of a distribution network according to claim 1, characterized in that, The voltage safety threshold is 1.07U. n .

6. A data-driven distribution network photovoltaic power generation acceptance capacity analysis device, used to implement the method as described in claim 1, characterized in that, include: Acquisition Unit: Includes acquiring historical data of active power, reactive power, and voltage amplitude of each node, and dividing the historical data into training set and test set; DELM Network Training Unit: The active and reactive power of historical data are determined as input data, and the voltage amplitude is determined as output data. The number of hidden layers, the number of neurons in each layer, and the activation function of the DELM network are set to train the DELM network and obtain network parameters. The SSA method was used to optimize the DELM network, and the optimized DELM network parameters were obtained. Fitting Unit: Based on the DELM network obtained through optimized training, the power of photovoltaic nodes is continuously increased to obtain the power-voltage sequence of each node, and the sequence is fitted using a fourth-order nonlinear function; Calculation unit: Sets voltage over-limit thresholds and obtains the target distribution network photovoltaic acceptance limit under single-point or multi-point photovoltaic power supply access based on the photovoltaic power supply installation location.

Images