New energy resource grading evaluation method based on clustering algorithm

Abstract

This invention discloses a hierarchical evaluation method for new energy resources based on clustering algorithms, specifically including the following steps: Step 1, calculating the theoretical monthly power generation hours for each new energy power station; Step 2, analyzing and correcting the data obtained in Step 1, constructing evaluation indicators for each new energy power station; Step 3, analyzing the evaluation indicators for each new energy power station obtained in Step 2 to obtain substation-level new energy evaluation indicators; Step 4, aggregating the substation-level evaluation indicators obtained in Step 3 to obtain and output provincial-level new energy evaluation indicators. This invention can provide theoretical basis and engineering guidance for new energy installed capacity planning, and is conducive to promoting the high-quality development of new energy.

Description

Technical Field

[0001] This invention belongs to the field of power system management and evaluation technology, and relates to a hierarchical evaluation method for new energy resources based on clustering algorithms. Background Technology

[0002] Currently, my country's new energy installed capacity planning and power generation grid connection allocation lack analysis of changes and flow patterns. It is impossible to reasonably plan and analyze new energy power generation based on resource changes and flow patterns throughout the year. This results in a one-sided assessment of the current new energy development level, and the imperfect new energy resource classification and evaluation methods are not conducive to guiding the future work of the power system and promoting the construction of a low-carbon, safe, and efficient modern energy system. Summary of the Invention

[0003] The purpose of this invention is to provide a hierarchical evaluation method for new energy resources based on clustering algorithms. This method can provide theoretical basis and engineering guidance for new energy installed capacity planning, and is conducive to promoting the high-quality development of new energy.

[0004] The technical solution adopted in this invention is a new energy resource hierarchical evaluation method based on clustering algorithm, which specifically includes the following steps:

[0005] Step 1: Calculate the theoretical monthly power generation hours for each new energy power station;

[0006] Step 2: Analyze and correct the data obtained in Step 1, and construct evaluation indicators for each new energy power station;

[0007] Step 3: Analyze the evaluation indicators of each new energy power station obtained in Step 2 to obtain the substation-level new energy evaluation indicators.

[0008] Step 4: Aggregate the substation-level evaluation indicators obtained in Step 3 to obtain and output the provincial-level new energy evaluation indicators.

[0009] The invention is further characterized by:

[0010] The specific process of step 1 is as follows:

[0011] Data related to new energy power plants is obtained from the EMS dispatch system, including grid-connected capacity indicators. Power generation indicators Power rationing quota Average wind speed index Average temperature index Radiation index Sunshine hours index The theoretical number of power generation hours for each station is calculated using the following formula (1):

[0012] (1);

[0013] in, It refers to the theoretical number of hours of power generation for a wind farm or photovoltaic power station. This is the total electricity generated in that month. This is the total electricity rationing for the month. This represents the actual grid-connected capacity of the power station.

[0014] The specific process of step 2 is as follows:

[0015] Step 2.1: Detect outliers using an outlier identification method based on Mahalanobis distance;

[0016] (2);

[0017] in, Represents Mahalanobis distance, Represents a sample vector. Represents the sample mean. Represents the covariance matrix between samples, when When the matrix is ​​an identity matrix, the Mahalanobis distance and the Euclidean distance are equal.

[0018] Step 2.2: Set the center spacing minDistance and check the calculation results obtained by formula (2). If the center spacing of the calculation results is less than minDistance, it indicates that there is an abnormality in the monthly theoretical power generation hours and it needs to be modified. Proceed to step 2.4; otherwise, proceed to step 2.3.

[0019] Step 2.3: Set the class center spacing maxDistance and check the results obtained in Step 2.2. If the monthly theoretical power generation hours are greater than maxDistance, it means that there is an anomaly in the monthly theoretical power generation hours and it needs to be modified. Proceed to Step 2.4; otherwise, proceed to Step 2.5.

[0020] Step 2.4: If there is abnormal data, the abnormal data is corrected using the average value correction method, as shown in formula (3):

[0021] (3);

[0022] in, This is abnormal data. and For two adjacent data points of abnormal data;

[0023] Step 2.5: Weight the various new energy power station indicators obtained after the correction in Step 2.4, as shown in the following formula (4):

[0024] (4);

[0025] in, It is an average wind speed indicator. It is an average temperature index. It is an indicator of radiation levels. It is an indicator of sunshine hours. The correction factor, q, represents the comprehensive evaluation index value of the new energy power station.

[0026] The specific process of step 3 is as follows:

[0027] Step 3.1, take the result obtained in step 2.5 The data uses a neuron algorithm to calculate the discriminant function value for each input and identifies the specific neuron with the smallest discriminant function value as the winner. Each neuron... j The discriminant function is:

[0028] (5);

[0029] Among them, the input space D Dimension, input is Input unit i and neurons j The connection weights between them in the computation layer are ,in N It is the total number of neurons;

[0030] Step 3.2, find the winning neuron Update with the winning neuron The formula for calculating the update level of neighboring nodes is as follows:

[0031] (6);

[0032] in, Represents neurons With the winning neuron Squared distance in the output topological space; It is the neighborhood width. It decays over time.

[0033] Step 3.3: Adjust the connection weights of the relevant excitatory neurons. ;

[0034] Step 3.4: Return to step 3.1 until the feature mapping tends to stabilize.

[0035] The specific process of step 4 is as follows:

[0036] Step 4.1, set clustering parameters: including the drift function g(x) and the drift vector N. h(x), where n is the number of sample points (i.e., the data from the field stations), h is the radius, s is the cluster density distance, and x is the number of clusters. The cluster center is i The threshold for data density is :

[0037] (7);

[0038] In the formula, k'(x) represents the derivative of the contour function of the kernel function used in this drift algorithm;

[0039] Step 4.2, in the n-dimensional space R composed of input station-level data n In the middle, select a circle with radius h and record its center as o;

[0040] Step 4.3: Record the set to which the station-level data within radius h belongs, and define the station-level data within set N as belonging to cluster C;

[0041] Step 4.4, calculate the value N of the drift vector within cluster C. h (x);

[0042] (8);

[0043] In the formula, xi Indicates the first i Individual station-level sample data, g(x) This is expressed as the derivative of the kernel contour function. h The bandwidth of the mean-shift kernel function, i.e., the neighborhood search radius, corresponds to the set N spatial radius, x The center of the current neighborhood (the cluster center iteration point) during the mean drift iteration process is the core independent variable of the formula;

[0044] Step 4.5, calculate the value N of the drift phasor. h (x) and the threshold representing data density Compare, if ||N(x)|| < If the iteration process ends, otherwise, re-derive the new center o' and repeat the process of steps 4.3-4.5 above. During iteration, all data points encountered belong to cluster C.

[0045] Step 4.6: If, upon convergence, the distance between the point with the maximum density of the current cluster C and the other existing cluster density points is greater than the cluster density distance s, then a new class is added; if it is less than the cluster density distance s, then the two clusters are merged into one class.

[0046] Step 4.7: Repeat steps 4.2–4.6 above until all points are marked;

[0047] Step 4.8: Based on the number of accesses to each site-level data point, the category to which the point belongs is determined according to the category with the most accesses, ultimately determining the classification level i for each province. and the number of clusters x .

[0048] The beneficial effects of this invention are that the new energy resource hierarchical evaluation method based on clustering algorithm proposed in this invention can display the micro-distribution of new energy resources in different regions and resource contour maps by using site-level resources as points instead of surfaces. This can solve the problem of one-sided assessment of new energy development level, explain the laws of resource change and flow, provide theoretical basis and engineering guidance for new energy installed capacity planning, and help promote the high-quality development of new energy. Attached Figure Description

[0049] Figure 1 This is a flowchart of the new energy resource hierarchical evaluation method based on clustering algorithm of the present invention;

[0050] Figure 2 This is a diagram of the self-organizing map neural network structure in the new energy resource classification and evaluation method based on clustering algorithm of this invention. Detailed Implementation

[0051] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0052] This invention relates to a new energy resource hierarchical evaluation method based on clustering algorithms, the process of which is as follows: Figure 1 As shown, the specific steps are as follows:

[0053] Step 1: Obtain information on each renewable energy power generation unit in a certain region's renewable energy power plants from the Electric Dispatch Automation System (EMS), and calculate the theoretical monthly power generation hours for each plant based on the obtained data; specifically: obtain relevant data on renewable energy power plants from the EMS dispatch system, including: grid-connected capacity indicators. Power generation indicators Power rationing quota Average wind speed index Average temperature index Radiation index Sunshine hours index Establish a tiered evaluation index system. Calculate the theoretical power generation hours for each power station:

[0054] (1);

[0055] in It refers to the theoretical number of hours of power generation for a wind farm or photovoltaic power station. This is the total electricity generated in that month. This is the total electricity rationing for the month. This represents the actual grid-connected capacity of the power station.

[0056] Step 2 involves analyzing the monthly theoretical power generation hours for each power station obtained in Step 1, identifying and correcting outlier data, and then constructing evaluation indicators for each renewable energy power station based on these indicators. Specifically, the basic information of renewable energy at each power station, including erroneous and missing data such as turbine type, capacity, grid connection location, and voltage level, is cleaned.

[0057] Step 2.1: Detect outlier data in the monthly theoretical power generation hours of each power station using an outlier identification method based on Mahalanobis distance.

[0058] First, the Mahalanobis distance is obtained based on formula (2).

[0059] (2);

[0060] in, Represents Mahalanobis distance, x This indicates that the sample vector is calculated from the result of step 1. middle, This indicates that the sample mean is derived from step 1.1. get, express and The covariance matrix when When the matrix is ​​an identity matrix, the Mahalanobis distance and the Euclidean distance are equal.

[0061] Step 2.2: Set the center spacing minDistance and check the calculation results obtained by formula (2). If the center spacing of the calculation results is less than minDistance, it indicates that there is an abnormality in the monthly theoretical power generation hours and it needs to be modified. Proceed to step 2.4; otherwise, proceed to step 2.3.

[0062] Step 2.3: Set the class center spacing maxDistance and check the results obtained in Step 2.2. If the monthly theoretical power generation hours are greater than maxDistance, it means that there is an anomaly in the monthly theoretical power generation hours and it needs to be modified. Proceed to Step 2.4; otherwise, proceed to Step 2.5.

[0063] Step 2.4: If there is abnormal data, the abnormal data is corrected using the average value correction method, as shown in formula (3).

[0064] (3);

[0065] in, This is abnormal data. and These are two adjacent data points to the abnormal data.

[0066] Step 2.5: Weight the various new energy power station indicators obtained after the correction in Step 2.4, taking into account the influence of each parameter.

[0067] (4);

[0068] in, It is the average wind speed index. It is an average temperature index. It is a radiation index, It is an indicator of sunshine hours. Correction factor ( , , , , ), where q represents the comprehensive evaluation index value of the new energy power station;

[0069] Step 3: Analyze the evaluation indicators of each new energy power station obtained in Step 2 to obtain substation-level new energy evaluation indicators; use a self-organizing map neural network (SOM) to cluster the theoretical hours of all power stations to obtain the classification results of 330kV substations. The structure of the SOM is shown in [link to SOM]. Figure 2 Specifically:

[0070] Step 3.1, network structure initialization, using the data obtained in step 2.5. The data uses a neuron algorithm to calculate the discriminant function value for each input and declares the specific neuron with the smallest discriminant function value as the winner. Each neuron... j The discriminant function is

[0071] (5);

[0072] Among them, the input space D Dimension, input is Input unit i and neurons j The connection weights between them in the computation layer are ,in N It is the total number of neurons;

[0073] Step 3.2, find the winning neuron Update the nodes and their neighboring nodes. The formula for calculating the update level is as follows:

[0074] (6);

[0075] in, Represents neurons With the winning neuron Squared distance in the output topological space; It is the neighborhood width. As time decays, the further away the node, the less it is updated, and the greater the penalty.

[0076] Step 3.3: Appropriately adjust the connection weights of relevant excitatory neurons. This enhances the response of the winning neuron to subsequent applications of similar input patterns; the value of the winning neuron is classified according to its numerical value, and it is assigned to that class.

[0077] Step 3.4: Return to step 3.1 until the feature mapping stabilizes. After the iteration ends, the neurons activated for each sample represent its corresponding category, which can be divided into a total of 4 categories.

[0078] Step 4: Aggregate the substation-level evaluation indicators obtained in Step 3 to obtain and output the provincial-level new energy evaluation indicators, specifically as follows:

[0079] Step 4.1, set clustering parameters: including the drift function g(x) and the drift vector N. h (x), where n is the number of sample points (i.e., the data from the field stations), h is the radius, s is the cluster density distance, and x is the number of clusters. The cluster center is i The threshold for data density is :

[0080] (7);

[0081] In the formula, k'(x) represents the derivative of the contour function of the kernel function used in this drift algorithm.

[0082] Step 4.2, in the n-dimensional space R composed of input station-level data n In the middle, select a circle with radius h and record its center as o;

[0083] Step 4.3: Record the set to which the station-level data within radius h belongs, and define the station-level data within set N as belonging to cluster C;

[0084] Step 4.4, calculate the value N of the drift vector within cluster C. h (x);

[0085] (8);

[0086] In the formula, xi Indicates the first i For each station-level sample data point, g(x) represents the derivative of the kernel profile function. h The bandwidth of the mean-shift kernel function, i.e., the neighborhood search radius, corresponds to the set N spatial radius,x The center of the current neighborhood during the mean drift iteration process, i.e., the cluster center iteration point, is the core independent variable of the formula;

[0087] Step 4.5, calculate the value N of the drift phasor. h (x) and the threshold representing data density Compare them. If ||N(x)|| < If the iteration ends, the process ends; otherwise, the new center o' is re-derived, and the process of steps 4.3-4.5 above is repeated; during iteration, all data points encountered belong to cluster C;

[0088] Step 4.6: If, upon convergence, the distance between the point with the highest density in the current cluster C and the density points of other existing clusters is greater than the cluster density distance s, then a new cluster is added; if the distance is less than the cluster density distance s, then the two clusters are merged into one cluster.

[0089] Step 4.7: Repeat steps 4.2–4.6 above until all points are marked;

[0090] Step 4.8: For each site-level data access count, the category to which that point belongs is determined by the category with the most access counts, ultimately determining the classification level i for each province. and the number of clusters x ;

[0091] Step 4.9: Combine the provincial classification results obtained in Step 4.8 with the power grid wiring diagram to draw new energy resource maps on a monthly and quarterly basis.

Claims

1. A method for hierarchical evaluation of new energy resources based on clustering algorithm, characterized in that, Specifically, the steps include the following: Step 1: Calculate the theoretical monthly power generation hours for each renewable energy power station. The specific process of Step 1 involves obtaining relevant data from the EMS dispatch system, including grid-connected capacity indicators. Power generation indicators Power rationing quota Average wind speed index Average temperature index Radiation index Sunshine hours index The theoretical number of power generation hours for each station is calculated using the following formula (1): （1）； in, It refers to the theoretical number of hours of power generation for a wind farm or photovoltaic power station. This is the total electricity generated in that month. This is the total electricity rationing for the month. This represents the actual grid-connected capacity of the power station; Step 2 involves analyzing and correcting the data obtained in Step 1 to construct evaluation indicators for each new energy power station. The specific process of Step 2 is as follows: Step 2.1: Detect outliers using an outlier identification method based on Mahalanobis distance; （2）； in, Represents Mahalanobis distance, Represents a sample vector. Represents the sample mean. Represents the covariance matrix between samples, when When the matrix is ​​an identity matrix, the Mahalanobis distance and the Euclidean distance are equal. Step 2.2: Set the center spacing minDistance and check the calculation results obtained by formula (2). If the center spacing of the calculation results is less than minDistance, it indicates that there is an abnormality in the monthly theoretical power generation hours and it needs to be modified. Proceed to step 2.4; otherwise, proceed to step 2.

3. Step 2.3: Set the class center spacing maxDistance and check the results obtained in Step 2.

2. If the monthly theoretical power generation hours are greater than maxDistance, it means that there is an anomaly in the monthly theoretical power generation hours and it needs to be modified. Proceed to Step 2.4; otherwise, proceed to Step 2.

5. Step 2.4: If there is abnormal data, the abnormal data is corrected using the average value correction method, as shown in formula (3): （3） in, This is abnormal data. and For two adjacent data points of abnormal data; Step 2.5: Weight the various new energy power station indicators obtained after the correction in Step 2.4, as shown in the following formula (4): （4） in, It is an average wind speed indicator. It is an average temperature index. It is an indicator of radiation levels. It is an indicator of sunshine hours. The correction factor, q, represents the comprehensive evaluation index value of the new energy power station; Step 3 involves analyzing the evaluation indicators for each new energy power station obtained in Step 2 to obtain substation-level new energy evaluation indicators. The specific process of Step 3 is as follows: Step 3.1, take the result obtained in step 2.5 The data uses a neuron algorithm to calculate the discriminant function value for each input and identifies the specific neuron with the smallest discriminant function value as the winner. Each neuron... j The discriminant function is: （5） Among them, the input space D Dimension, input is Input unit i and neurons j The connection weights between them in the computation layer are ,in N It is the total number of neurons; Step 3.2, find the winning neuron Update with the winning neuron The formula for calculating the update level of neighboring nodes is as follows: （6）； in, Represents neurons With the winning neuron Squared distance in the output topological space; It is the neighborhood width. Decays over time; Step 3.3: Adjust the connection weights of the relevant excitatory neurons. ; Step 3.4: Return to step 3.1 until the feature mapping tends to stabilize; Step 4: Aggregate the substation-level evaluation indicators obtained in Step 3 to obtain and output the provincial-level new energy evaluation indicators; the specific process of Step 4 is as follows: Step 4.1, set clustering parameters: including the drift function g(x) and the drift vector N. h (x), where n is the number of sample points (i.e., the data from the field stations), h is the radius, s is the cluster density distance, and x is the number of clusters. The cluster center is i The threshold for data density is : （7） In the formula, k'(x) represents the derivative of the contour function of the kernel function used in this drift algorithm; Step 4.2, in the n-dimensional space R composed of input station-level data n In the middle, select a circle with radius h and record its center as o; Step 4.3 Record the set to which the station-level data within radius h belongs, and define the station-level data within set N as belonging to cluster C; Step 4.4, calculate the value N of the drift vector within cluster C. h (x); （8）； In the formula, xi Indicates the first i For each station-level sample data point, g(x) represents the derivative of the kernel profile function. h The bandwidth of the mean-shift kernel function, i.e., the neighborhood search radius, corresponds to the set N spatial radius, x The center of the current neighborhood during the mean drift iteration process, i.e., the cluster center iteration point, is the core independent variable of the formula; Step 4.5 calculates the value N of the drift phasor. h (x) and the threshold representing data density Compare, if ||N(x)|| < If the iteration process ends, otherwise, re-derive the new center o' and repeat the process of steps 4.3-4.5 above. During iteration, all data points encountered belong to cluster C. Step 4.6 If, upon convergence, the distance between the point with the maximum density of the current cluster C and the other existing cluster density points is greater than the cluster density distance s, then a new class is added; if it is less than the cluster density distance s, then the two clusters are merged into one class. Step 4.7: Repeat steps 4.2–4.6 above until all points are marked; Step 4.8: Based on the number of accesses to each site-level data point, the category to which the point belongs is determined according to the category with the most accesses, ultimately determining the classification level i for each province. and the number of clusters x .

Images