CN110111003A - A kind of new energy typical scene construction method based on improvement FCM clustering algorithm - Google Patents

Abstract

The invention discloses a new energy typical scene construction method based on the improved FCM clustering algorithm, which includes the following steps S1: improving the FCM clustering algorithm by establishing a clustering validity index function; S2: using the improved FCM clustering algorithm to analyze the new energy Clustering and division of output historical data; S3: Select typical scenarios of new energy output in each category after clustering. This invention takes the new energy output characteristics of new energy-rich areas as the research object, uses the improved FCM algorithm to cluster and analyze the historical time-series output data of new energy, and reduces and merges the original large-scale scenes to obtain several representative ones. The collection of new energy output scenarios has certain practical application value.

Description

A new energy typical scene construction method based on improved FCM clustering algorithm

Technical field:

The invention relates to a construction method of a typical new energy output scene, in particular to a construction method of a new energy typical scene based on an improved FCM clustering algorithm.

Background technique

With the large-scale development and utilization of new energy in my country, new energy power generation continues to develop rapidly, and the scale of installed capacity growth continues to expand. With the continuous growth of new energy installed capacity and the continuous increase of the proportion of new energy in the grid power supply, the demand for new energy consumption has put forward higher requirements for the economic operation of the power system, the evaluation of the new energy consumption capacity of the power grid, and the formulation of power grid dispatching plans. requirements. Therefore, considering the seasonality and periodicity of new energy output such as wind power and photovoltaics, representative typical output scenarios are extracted from historical output data, and these typical output scenarios are used to reflect the characteristics of new energy output in the medium and long term. The power supply planning of the power system with a high proportion of new energy is of great significance.

In recent years, domestic and foreign researchers have conducted research on the construction of typical new energy output scenarios. At present, there are generally three methods for selecting medium and long-term new energy output scenarios or load characteristics that are widely used: typical day method, time series simulation method and clustering algorithm. The typical day method usually refers to taking the output characteristics of the day closest to the average value in a certain period as the typical daily output scenario, or selecting a representative day in a certain period as the typical daily output scenario. It is simple and quick to obtain the output characteristics of new energy through the typical Japanese method, but because the scenarios are not rich enough to reflect the changing characteristics of new energy output throughout the year, there are large errors in the mid- and long-term power planning calculations. The time series simulation method refers to the time series data of actual output of historical new energy sources, and then adjusted according to changes in installed capacity and other factors to obtain a time series of simulated output. The annual output time series obtained by the time series simulation method is close to the actual output characteristics of new energy sources such as wind power and photovoltaics every day, and the results are accurate and reliable. The disadvantage is that the calculation efficiency is low. The clustering algorithm extracts, classifies, and simplifies the information of the long-term time-series new energy actual output scenarios through the method of cluster analysis, and then obtains a set of typical scenarios. The clustering algorithm not only guarantees the original characteristics of the output data, but also takes into account the computational efficiency. At present, most studies have analyzed a large amount of actual load or wind power data through classical clustering algorithms, and obtained user load and wind power output scene sets that can more accurately reflect the actual characteristics. , Conduct in-depth discussions on the correlation between multiple types of new energy in the same region.

In view of this, it is necessary to improve and optimize the traditional FCM clustering algorithm, so as to cluster and analyze the historical time-series output data of new energy in an area rich in new energy, generate a set of typical new energy output scenarios in this area, and analyze the actual problems The validity of the verification method in.

Contents of the invention

Purpose of the invention: The purpose of the invention is to provide a new energy typical scene construction method based on the improved FCM clustering algorithm that can solve the defects in the prior art.

Technical scheme: in order to achieve this goal, the present invention adopts following technical scheme:

A method for constructing a typical new energy scenario based on an improved FCM clustering algorithm, the method comprising the following steps:

S1: Improve the FCM clustering algorithm by establishing a clustering effectiveness index function;

S2: Use the improved FCM clustering algorithm to cluster and divide the historical data of new energy output;

S3: Select typical scenarios of new energy output in each category after clustering.

In the described new energy typical scene construction method based on the improved FCM clustering algorithm, the clustering validity index function CH ⁽⁺⁾ in step S1 is:

Among them, c is the number of clusters, n is the number of samples, T _c , P _c are the sum of the squares of inter-class and intra-class deviations.

In the method for constructing a typical new energy scenario based on the improved FCM clustering algorithm, the improved FCM clustering algorithm flow by establishing a clustering effectiveness index function in step S1 is as follows:

1) Set the variation range of the number of clusters c

2) Call the FCM clustering algorithm to update the clustering center until the objective function converges;

3) Calculate the CH ⁽⁺⁾ index;

4) c=c+1, turn to step 2), until

5) Compare the size of the CH ⁽⁺⁾ index under different c values to determine the optimal number of clusters;

6) Output the clustering results under the optimal clustering number.

In the method for constructing typical new energy scenarios based on the improved FCM clustering algorithm, the new energy includes wind energy and light energy, and the process of clustering and dividing the historical data of new energy output by using the improved FCM clustering algorithm in step S2 is as follows:

1) Collect historical output data of wind power and photovoltaics;

2) Preprocess the historical output data of new energy sources, including the correction of missing data and mutation data, etc., and obtain the output data set of wind farms with n consecutive days and m equal time intervals per day as follows: The photovoltaic power station output data set of the photovoltaic power station corresponding to the time interval is in, Indicates the wind power output data on the nth day, Indicates the photoelectric output data of the nth day;

3) Use the improved FCM clustering algorithm proposed in step S1 to cluster and divide the historical data X ^w and X ^s of wind power and photovoltaic output.

In the method for constructing typical scenarios of new energy based on the improved FCM clustering algorithm, the selection process of typical scenarios of new energy output in each category after clustering in step S3 is as follows:

1) Read the output data of each historical day in this category;

2) Calculate the average of all historical daily output power that will belong to this class j is the total number of historical days belonging to this category, P _i is the daily output power of historical day i;

3) Calculate the distance d _i =|P _i -P _avg | between the output power and the average power of each historical day;

4) Sort d _i in ascending order, and select the closest historical daily output curve as a typical scenario of this type.

Beneficial effects: The invention takes the output characteristics of new energy sources in regions with abundant new energy resources as the research object, and proposes a construction method of typical new energy scene sets based on the improved FCM clustering algorithm and an operation cost optimization model for power systems containing high proportions of new energy sources. Using the typical scene set construction method based on the improved FCM clustering algorithm to cluster and analyze the historical time-series output data of new energy sources in this area, a typical output scene set for this area is generated. The simulation results of specific implementations show that the wind power and photoelectric typical scenes generated by the new energy typical scene set construction method based on the improved FCM clustering algorithm have obvious annual characteristics, conform to the actual output situation, and have high computational efficiency and low error in actual engineering applications. small advantage.

Description of drawings

Fig. 1 is the clustering effectiveness index of wind power output scene in the specific embodiment of the present invention;

Fig. 2 is the typical output curve of wind power spring and autumn in the specific embodiment of the present invention;

Fig. 3 is the typical output curve of wind power in summer in the embodiment of the present invention;

Fig. 4 is the typical output curve of wind power in winter in the embodiment of the present invention;

Fig. 5 is the typical output curve of photoelectric spring and autumn in the embodiment of the present invention;

Fig. 6 is the typical summer output curve of photoelectricity in the embodiment of the present invention;

Fig. 7 is a typical output curve of photovoltaics in winter in a specific embodiment of the present invention.

Detailed ways

The technical solution of the present invention will be further introduced below in combination with specific embodiments.

The new energy typical scene construction method based on the improved FCM clustering algorithm described in the present invention comprises the following steps:

S1: Improve the FCM clustering algorithm by establishing a clustering effectiveness index function;

S2: Use the improved FCM clustering algorithm to cluster and divide the historical data of new energy output;

S3: Select typical scenarios of new energy output in each category after clustering.

Furthermore, the FCM clustering algorithm in step S1 is a partition-based clustering algorithm. By introducing the concept of membership function, the relationship between objects and clusters is extended to any It is described by the value of the membership function, which cluster the object is more likely to belong to by judging the value of the membership function.

First, the concept of fuzzy subset is defined as:

For any x∈U, a number μ _A (x)∈[0,1] can be determined to describe the degree of x belonging to A, which is defined as the membership function of A, and the membership of the elements in U to the fuzzy subset A The degree is described by the constant μ _A (x). The closer the membership degree of μ _A (x) is to 0, the smaller the degree that x belongs to A; the closer μ _A (x) is to 1, the greater the degree that x belongs to A.

For a data set X={x ₁ ,x ₂ ,x ₃ ,…,x _n }, use FCM clustering to divide the data set X into c categories (2≤c≤n), where the set of cluster centers can be Expressed as V={v ₁ ,v ₂ ,...,v _c }, in fuzzy division, a certain membership degree is used to describe each data object belongs to a certain category, and it is not strictly divided into a certain category.

The membership degree μ _ij of the i-th data sample x _i in the data set X belonging to the j-th class is expressed by the following numerical relationship:

Initialize the membership degree matrix U={μ _i,j } with random numbers in the interval [0,1], and solve the minimum value of the objective function on the basis of satisfying the constraints in the above formula:

The membership value μ _ij is calculated as follows:

In the formula, F(X,V) represents the weighted sum of the squares of distances from the data sample to the cluster center, the weight is the f power of the membership degree μ _ij of the j-th class of the data sample x _i , and the smoothing factor (fuzzy weighting parameter) f∈ [1,+∞), the index of the adjusted membership degree can be used to control the smoothness of the adjusted membership degree. In general calculation examples, if there is no special requirement, the smoothing factor f is generally taken as 2. U={μ _i,j } is the fuzzy membership matrix. d _ij =|| _xi -v _j || represents the Euclidean distance from the i-th data sample to the j-th cluster center.

The flow of the FCM clustering algorithm can be summarized as follows:

1) Set the number of clusters c, the number of iterations k=0, the maximum number of iterations T, the termination error ε, and initialize the cluster center V ⁰ ;

2) Update the membership degree matrix U ^k with formula (4);

3) Update the next clustering center V ^k+1 with formula (3) and formula (4);

4) If ||U ^k+1 -U ^k ||<ε, end the algorithm. Otherwise, let k=k+1, return to step 2).

Further, the clustering validity index function CH ⁽⁺⁾ in step S1 is:

Among them, T _c , P _c are the sum of squares of deviations between classes and within classes. The sum of squares of inter-class deviations can reflect the differences between classes, and the larger the value, the better; the sum of squares of intra-class deviations reflects the differences between samples of the same class, and the smaller the value, the better. Therefore, finding the optimal number of clusters c can be transformed into seeking the maximum value of this index.

Further, the process of the improved FCM clustering algorithm by establishing the clustering effectiveness index function in step S1 is summarized as follows:

1) Set the variation range of the number of clusters c

2) Call the FCM clustering algorithm to update the clustering center until the objective function converges;

3) Calculate the CH ⁽⁺⁾ index;

4) c=c+1, turn to step 2), until

5) Compare the size of the CH ⁽⁺⁾ index under different c values to determine the optimal number of clusters;

6) Output the clustering results under the optimal clustering number.

Further, the new energy in step S2 includes wind energy and light energy, and the process of clustering and dividing the historical data of new energy output by using the improved FCM clustering algorithm is as follows:

1) Collect historical output data of wind power and photovoltaics;

2) Preprocess the historical output data of new energy sources, including the correction of missing data and mutation data, etc., and obtain the output data set of wind farms with n consecutive days and m equal time intervals per day as follows: The photovoltaic power station output data set of the photovoltaic power station corresponding to the time interval is in, Indicates the wind power output data on the nth day, Indicates the photoelectric output data of the nth day;

3) Use the improved FCM clustering algorithm proposed in step S1 to cluster and divide the historical data X ^w and X ^s of wind power and photovoltaic output.

Further, the selection process of typical scenarios of new energy output in each category after clustering in step S3 is as follows:

1) Read the output data of each historical day in this category;

2) Calculate the average of all historical daily output power that will belong to this class j is the total number of historical days belonging to this category, P _i is the daily output power of historical day i;

3) Calculate the distance d _i =|P _i -P _avg | between the output power and the average power of each historical day;

4) Sort d _i in ascending order, and select the closest historical daily output curve as a typical scenario of this type.

Taking a new energy-rich area in Zhejiang Province as an example, use the construction method of the typical scene set of new energy output based on the improved FCM algorithm proposed in this paper to construct a typical scene set of new energy in this area. The installed capacity of wind power in this area is 50MW, and the installed capacity of photovoltaic power generation is 320MW, accounting for about 25% of the installed capacity in the whole region. The output data of wind power and photovoltaic power generation in this area from June 1, 2017 to May 31, 2018 are selected as samples. Considering the wind power , photoelectric output has a periodicity related to the season. In this specific implementation, the wind power and photoelectric historical output data samples are divided into three categories: spring and autumn, summer and winter, and then a typical scene is constructed, and the constructed new energy typical The scene set will be applied to the practical problem of medium and long-term operation cost optimization of a high-proportion new energy power system, and the practical application value of this method will be evaluated.

Example 1:

Example 1 uses the improved FCM clustering algorithm to divide the historical output data of wind power and photovoltaics into scenarios. In the clustering process, the clustering validity index calculation is performed first, and then the scenarios are divided after obtaining the optimal number of clusters. Taking the wind power in this area as an example, the clustering effectiveness CH ⁽⁺⁾ index of wind power output scenarios in each season is calculated. In this paper, the extreme value normalization method is used to process the CH ⁽⁺⁾ index, as shown in the following formula:

The clustering effectiveness CH ⁽⁺⁾ index of the processed wind power output scenario is shown in Figure 1.

It can be seen from Figure 1 that the clustering effectiveness index CH ⁽⁺⁾ of wind power in each season takes its maximum value at 2, that is, the optimal clustering number of wind power output scenarios in each season is 2. Similarly, the optimal clustering number of output scenarios of photovoltaics in each season is also 2. Figures 2 to 4 show the typical output curves of wind power in spring and autumn, summer and winter, and Figures 5 to 7 show the typical output curves of photovoltaics in spring and autumn, summer and winter. It can be seen from the output curve above that the clustering results are roughly divided according to the level of output. Table 1 and Table 2 respectively give the probability distribution of typical scenarios of wind power and photovoltaic output in each season.

Table 1 Probability distribution of typical scenarios of wind power output in each season

Table 2 Probability distribution of typical scenarios of photoelectric output in each season

The output curves in Fig. 2 to Fig. 7 intuitively represent the typical output size and change trend of wind power and photovoltaic in each season. Wind power has strong volatility, and the peak and valley change process and reverse peak regulation characteristics are obvious. Photoelectric output also has a certain degree of volatility, and its output is closely related to the light intensity, and the maximum value is taken around 13:00 every day. It can be seen from the output curve above that the clustering results are roughly divided according to the level of output. Table 1 and Table 2 respectively give the probability distribution of typical scenarios of wind power and photovoltaic output in each season.

Example 2:

Example 2 applies the typical scenario of wind power/photovoltaic output to the field of operating cost optimization of high-proportion new energy power systems. The minimum sum of operating costs is the optimization goal, and the maximum consumption of new energy is taken into account while economical operation is carried out. The objective function is:

Among them, T represents the total number of simulation periods, c _w is the wind power penalty coefficient, c _s is the photoelectric penalty coefficient, Indicates the predicted output of wind power in period t, Indicates the actual output of wind power in period t, Indicates the predicted output of the photoelectricity in the period t, Indicates the actual output of the photoelectricity in the period t, Indicates the actual output of thermal power in period t, a _th , b _th , c _th are the cost coefficients of thermal power generation, a _s , b _s , c _s are the cost coefficients of wind power generation, a _w , b _w , c _w are the cost coefficients of photovoltaic power generation .

The constraints considered by the optimization model include:

1) Constraints on power balance: the total power generation of regional thermal power, wind power, and photovoltaics is balanced with user loads.

Among them, u represents the user load in this area.

2) Generating power constraints: the generating capacity of each type of generating unit satisfies the respective maximum and minimum generating capacity constraints.

Among them, q _thmin represents the minimum power generation of thermal power plants, and q _thmax represents the maximum power generation of thermal power plants. Similarly, q _wmin , q _wmax , q _wmin , and q _wmax represent the maximum and minimum power generation of wind power and photovoltaic power.

3) Thermal power unit climbing constraints:

Among them, γ _down and γ _up represent the down-slope rate and up-slope rate of the thermal power unit, respectively, and Respectively represent the power generation of the thermal power plant during t-1 and t period.

Use the above power system operation cost optimization model to verify and compare the scenarios constructed by the annual time series method, the improved FCM algorithm and the typical day method. In the typical day method, the output of wind power and photovoltaic power in spring, autumn, summer and winter is closest to the average value. A day's output is used as a typical scenario. Taking the annual output scenario constructed by the annual timing method as the benchmark, the performance comparison of the three scenario construction methods is shown in Table 3.

Table 3 Performance comparison of scene construction methods

It can be seen that the scene construction method based on the improved FCM clustering algorithm and the typical Japanese method has significantly improved the computational efficiency due to the large reduction in the number of scenes. However, compared with the annual output scenarios of wind power and photovoltaics constructed by the annual time series method, the scenarios constructed by the improved FCM clustering algorithm and the typical daily method have certain errors. Among them, compared with the scene construction method based on the improved FCM clustering algorithm and the annual time series method, the error rates of wind power and photovoltaic prediction results are 33.4% and 27.1%, respectively. Compared with wind power and photovoltaic prediction results, the error rates are 49.8% and 51.9%, respectively. In comparison, the scene construction method based on the improved FCM clustering algorithm can accurately reflect the annual wind power and photovoltaic performance while ensuring computing efficiency. The operation situation has practical application value.

Claims (5)

1. A kind of new energy typical scene construction method based on improved FCM clustering algorithm, it is characterized in that: comprise the following steps: S1: Improve the FCM clustering algorithm by establishing a clustering effectiveness index function; S2: Use the improved FCM clustering algorithm to cluster and divide the historical data of new energy output; S3: Select typical scenarios of new energy output in each category after clustering. 2. The new energy typical scene construction method based on the improved FCM clustering algorithm according to claim 1, characterized in that: the clustering validity index function CH ⁽⁺⁾ in step S1 is: Among them, c is the number of clusters, n is the number of samples, T _c , P _c are the sum of the squares of inter-class and intra-class deviations. 3. The new energy typical scene construction method based on the improved FCM clustering algorithm according to claim 1, characterized in that: in step S1, the improved FCM clustering algorithm process by establishing the clustering effectiveness index function is as follows: 1) Set the variation range of the number of clusters c 2) Call the FCM clustering algorithm to update the clustering center until the objective function converges; 3) Calculate the CH ⁽⁺⁾ index; 4) c=c+1, turn to step 2), until 5) Compare the size of the CH ⁽⁺⁾ index under different c values to determine the optimal number of clusters; 6) Output the clustering results under the optimal clustering number. 4. The method for constructing typical scenarios of new energy based on the improved FCM clustering algorithm according to claim 1, characterized in that: new energy includes wind energy and light energy, and the use of the improved FCM clustering algorithm in step S2 to analyze the output history of new energy The process of clustering and dividing data is as follows: 1) Collect historical output data of wind power and photovoltaics; 2) Preprocess the historical output data of new energy sources, including the correction of missing data and mutation data, etc., and obtain the output data set of wind farms with n consecutive days and m equal time intervals per day as follows: The photovoltaic power station output data set of the photovoltaic power station corresponding to the time interval is in, Indicates the wind power output data on the nth day, Indicates the photoelectric output data of the nth day; 3) Use the improved FCM clustering algorithm proposed in step S1 to cluster and divide the historical data X ^w and X ^s of wind power and photovoltaic output. 5. The method for constructing typical scenarios of new energy based on the improved FCM clustering algorithm according to claim 1, characterized in that: the selection process of typical scenarios of new energy output in each category after clustering in step S3 is: 1) Read the output data of each historical day in this category; 2) Calculate the average of all historical daily output power that will belong to this class j is the total number of historical days belonging to this category, P _i is the daily output power of historical day i; 3) Calculate the distance d _i =|P _i -P _avg | between the output power and the average power of each historical day; 4) Sort d _i in ascending order, and select the closest historical daily output curve as a typical scenario of this type.