A Method for Classifying and Shifting Loads for Industrial Electricity Users with High Applicability and High Precision

Abstract

The present invention discloses a method for classifying and staggering electricity consumption of industrial electricity users with high applicability and high precision, which is based on clustering algorithms and Fourier transforms as features, and includes: Step S10, collecting electricity load data of target users and performing data cleaning using the load ratio law; Step S20, performing data smoothing processing on the cleaned load data using the moving average method; Step S30, performing Fourier transform on the smoothed data and using it as data features; Step S40, inputting the data after completing the feature engineering into the K-means clustering algorithm and optimizing the model parameters; Step S50, classifying the clustered users according to the load volume to obtain the results; Step S60, performing intra-class grouping on users in the same category and arranging staggered electricity consumption. The present invention solves the difficulty of complex and difficult-to-distinguish user electricity consumption behaviors, extracts the deep electricity consumption rules of users for clustering, and manages the staggered electricity consumption of users, with the advantages of clear thinking and being suitable for popularization and use.

Description

Technical Field

[0001] The present invention belongs to the field of power system analysis and calculation, and particularly relates to the technical field of peak-shaving power consumption for classifying the power consumption behaviors of different large industrial users in the power system. Background Art

[0002] In response to the national carbon neutrality strategic goal, the rational application of electric energy has become a top priority. To avoid waste of electric energy, power rationing measures have been implemented for industrial users in many provinces and cities. However, blind power rationing will have a negative impact on the economy and greatly reduce user satisfaction. Therefore, there is room for exploration, regulation, and optimization in the reasonable allocation of power consumption for large users in certain industries. At the same time, the power consumption behaviors of large users also have a huge impact on the safe and stable operation of the power grid. A reasonable power consumption arrangement will also reduce the peak load of the power grid, which is of great significance to the power grid.

[0003] To rationally utilize electric energy and scientifically arrange the power consumption of large users, a relatively reasonable method is peak-shaving power consumption. However, blind peak-shaving power consumption not only cannot reduce the power consumption peak but also will forcibly change the original production habits of users, causing dissatisfaction among users. Therefore, mastering the power consumption patterns of large users and reasonably classifying large users are the basic conditions for peak-shaving power consumption. Currently, large industrial users are only grouped together vaguely according to industry categories, and their power consumption patterns have not been analyzed, let alone substantial peak-shaving power consumption measures. To classify users according to their power consumption behaviors, it is necessary to extract power consumption behavior characteristics and calculate the similarity between users. However, the difficulty lies in that the lengths of the power consumption histories of users are mostly different and cannot be directly used to calculate similarity. Intercepting a partial length will lead to information loss and affect the calculation accuracy. Moreover, the original load data mostly has small-scale fluctuations, which have little significance in classification but will greatly affect the calculated similarity value, resulting in poor classification effects. Summary of the Invention

[0004] To solve the defects and deficiencies existing in the above-mentioned prior art and in the context of the definite need for a new method for classifying and peak-shaving power consumption of large users, after research and design by the inventor, there is now provided an industrial power user classification and peak-shaving power consumption method based on clustering algorithms and Fourier transforms, which can eliminate the influence of different historical load lengths, eliminate the calculation similarity differences caused by small fluctuations in the load, accurately classify users, optimize the use of electric energy for peak-shaving power consumption, and improve the stability of power grid operation. Specifically, this method proposes a method for classifying and peak-shaving power consumption of users' power consumption behaviors using Fourier transform to extract users' power consumption behavior characteristics and combining the k-means clustering algorithm, which can not only eliminate the influence of different historical load lengths but also eliminate the calculation similarity differences caused by small fluctuations in the load. The present invention is implemented as follows:

[0005] A method for classifying and staggering electricity consumption of industrial electricity users with high applicability and high precision, comprising the following steps:

[0006] Step S10, collecting the electricity load data of users, cleaning the points where the load change is more than 50% compared with the previous moment, and correcting the values using the Newton interpolation method;

[0007] Step S20, performing data smoothing processing on the cleaned load data by the moving average method,

[0008] Step S30, performing Fourier transform on the load data after data smoothing, and using the transformed data as new data features;

[0009] Step S40, inputting the data after completing the feature engineering into the K-means clustering algorithm, and tuning the model parameters;

[0010] Step S50, classifying the clustered users according to the load volume to obtain the results; classifying the users in the same category again according to the load volume to make the load more even when arranging the staggered electricity consumption;

[0011] Step S60, performing intra-class grouping on users in the same category and arranging staggered electricity consumption.

[0012] Further, the calculation formula of the moving average method for the data smoothing processing in the step S20 is: F t =(A t-1 +A t-2 +A t-3 +...+A t-n ) / n, where F t is the moving average value at time t, A t-i is the original data at time t-i, and n is the total number of times. Through the moving average method, the original load curve can be changed into a load trend curve, eliminating some large fluctuations;

[0013] Further, the Fourier transform formula in the step S30 is: where F(ω) is the frequency domain function, f(t) is the time domain function, t is the time, and ω is the independent variable; and the same number of terms in the front are taken for the curve waveforms of all lengths.

[0014] Further, the step S40 further includes: The K-means clustering algorithm usually uses the Euclidean distance as the metric standard, and the Euclidean distance formula in the n-dimensional space: where x and y are two points in the n-dimensional space respectively.

[0015] Furthermore, the K-means clustering algorithm belongs to an unsupervised algorithm. It can automatically cluster similar data into one category according to the characteristics of the data dimensions. The clustering categories completely depend on the similarity of the characteristics, rather than the categories defined artificially. This clustering method can classify data more objectively according to the data characteristics. The K-means algorithm also includes the steps of: creating k initial values as the initial centroids and they should be located within the data boundaries; if the cluster assignment result of any point changes: traverse each point in the dataset; traverse the k centroids: calculate the distance between the centroid and the data point; assign the data point to the cluster of the centroid with the closest distance; traverse the k clusters: calculate the mean of all points in each cluster; update the obtained k means as the new centroids. Definition of centroid: the center point of each cluster, and the average distance from all points within the cluster to this center point is the shortest.

[0016] Furthermore, the calculation method of Newton interpolation method is as follows:

[0017]

[0018] And replace the data that is more than 10 times the median with the median.

[0019] Introduction to the working principle of the present invention: Fourier transform is used to extract the data characteristics of historical load, eliminating the influence of different electricity consumption amounts of each user on classification. Moreover, the curve components extracted by Fourier transform have nothing to do with the curve length. Users with different historical load lengths can still be clustered uniformly, and common problems such as data missing or large differences in data magnitude can be easily handled. The curve components of the original load data are extracted by Fourier transform, and this component data is used as the clustering feature. The K-means clustering algorithm is used to calculate the relative distance between each component feature through the Euclidean distance. Finally, further subdivision is carried out according to the load volume within the same category. In this way, the components in the front section of the Fourier transform that have a strong influence on the curve can be intercepted as features, eliminating the influence of small load fluctuations on the similarity calculation, and the number of categories can be independently controlled according to needs. And the classification is based on the electricity consumption rules of users, and the classification effect is closer to the usage requirements. The off-peak electricity consumption arrangement is easier to be accepted by users and the off-peak electricity consumption effect is better. The present invention can eliminate the influence of different historical load lengths, can eliminate the calculation similarity differences brought by small fluctuations of the load, classify electricity users according to their electricity consumption behavior rules, and make off-peak electricity consumption arrangements according to the classification. Extract the components of the user's historical electricity load curve as clustering features through Fourier transform to eliminate the influence of different historical load lengths and eliminate the calculation similarity differences brought by small fluctuations of the load; divide the extracted curve component features into several categories as needed through the k-means clustering algorithm, with better flexibility; finally, use in-category grouping to group users in the same category for off-peak electricity consumption.

[0020] Advantages of the present invention: Under the condition of eliminating the influence of different historical load lengths, new data features are constructed using Fourier transform, the calculation similarity differences caused by small fluctuations of the load are eliminated, and the number of classifications is autonomously determined through the k-means clustering algorithm, breaking the limitation of different lengths of historical load data. Moreover, this peak-shaving power consumption method has universal applicability. The present invention has the advantages of clear thinking, good versatility, high economic value, and is suitable for popularization and use. Brief Description of the Drawings

[0021] Figure 1 It is a flowchart of an industrial electricity user classification peak-shaving power consumption method provided by the present invention based on a clustering algorithm and Fourier transform as features;

[0022] Figure 2 It is a flowchart of abnormal data cleaning based on the load ratio law provided by the present invention;

[0023] Figure 3 It is a code logic flowchart of the K-means algorithm provided by the present invention;

[0024] Figure 4 It is a result diagram of classifying the loads of large industrial users in a certain area into four categories by the classification method provided by the present invention;

[0025] Figure 5 It is a comparison diagram of the peak-shaving power consumption hourly load and the non-peak-shaving power consumption effect of large industrial users in a certain area by the industrial electricity user classification peak-shaving power consumption method provided by the present invention, where the load on the left is the non-peak-shaving power consumption and the load on the right is the peak-shaving power consumption. Detailed Embodiment

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the detailed embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention. Please refer to Figure 1 , Figure 1 It is a flowchart of an industrial electricity user classification peak-shaving power consumption method provided by the present invention based on a clustering algorithm and Fourier transform as features, mainly including the following steps: Figure 2 It is a data cleaning method, which can fill and modify data according to the trends of the front and back points of the load data, better restore the authenticity of the data, and avoid data jumps.

[0027] In step S10, electricity load data of target users is collected and data cleaning is performed using the load ratio law. The original 15-minute load data is collected. After data analysis, it is found that the change in electricity load of industrial users will not exceed 50% except during holidays or maintenance. Therefore, according to this law, points where the load change is greater than 50% compared with the previous moment can be cleaned and the values are corrected using the Newton interpolation method. Because load data often has distortion such as jumps and mutations, in order to make the data restoration closer to the truth, the Newton interpolation method can be used to make the interpolated data follow the change trend of the previous and next data. Of course, the Lagrange interpolation method can also be used. The interpolated data of the Lagrange interpolation method is the median of the connected data after signing. The calculation method of the Newton interpolation method is:

[0028]

[0029] Before calculating the data expectation and variance, in order to prevent a few extremely large outliers from having a huge impact on the data expectation and variance, the data that is more than 10 times the median is first replaced with the median;

[0030] Step S20, the cleaned load data is smoothed using the moving average method. The calculation formula of the moving average method is: F t =(A t-1 +A t-2 +A t-3 +...+A t-n ) / n, where F t is the moving average at time t, A t-i is the original data at time ti, and n is the total number of time. The original load curve can be transformed into a load trend curve by the moving average method, eliminating some large fluctuations;

[0031] Step S30: Perform Fourier transform on the smoothed load data and use the transformed data as new data features. The Fourier transform formula is: Where F(ω) is a frequency domain function, f(t) is a time domain function, t is time, and ω is the independent variable. In the Fourier transform results, the frontmost terms contain more information, which means they have a greater impact on the original curve waveform. Taking the same number of frontmost terms for all curve waveform lengths can minimize the difference in similarity calculation caused by small load fluctuations and unify curves of all lengths to the same length.

[0032] Step S40: Input the feature-engineered data into the K-means clustering algorithm and tune the model parameters. The K-means clustering algorithm usually uses Euclidean distance as the metric. The Euclidean distance formula in n-dimensional space is: Where x and y are two points in an n-dimensional space respectively. The K-means algorithm first randomly selects K objects as the initial clustering centers. Then it calculates the distance between each object and each seed clustering center, and assigns each object to the clustering center closest to it. The clustering centers and the objects assigned to them represent a cluster. Once all the objects are assigned, the clustering center of each cluster will be recalculated based on the existing objects in the cluster. This process will be repeated continuously until a certain termination condition is met. For the process flow of this step, please refer to Figure 3 ;

[0033] Step S50: Classify the clustered users again according to the load volume to obtain the results. Even if users with different volumes stagger their electricity consumption, it is not very reasonable because their impacts on the load are different. Therefore, it is necessary to classify the users in the same category again according to the load volume so that the load is more evenly distributed when arranging peak-shaving electricity consumption. That is, when we cluster the users in the same cluster into 5 categories, then classify the users in each of these 5 categories again according to the electricity consumption volume, and divide the users with the same magnitude into small categories under these 5 categories. For the process flow of electricity users in this step, please refer to Figure 4 ;

[0034] Step S60: Group users within the same category and arrange peak-shaving electricity consumption. Since the electricity consumption behaviors of users in the same category are very similar, if they consume electricity simultaneously, it will cause a sharp increase in the grid load and a corresponding increase in the demand for power generation. Therefore, it is necessary to group users in the same category and arrange their production activities at different times, that is, peak-shaving electricity consumption.

[0035] The present invention can eliminate the influence of different historical load lengths, can eliminate the calculation similarity differences caused by small fluctuations in the load, classify electricity users according to the electricity consumption behavior rules, and arrange peak-shaving electricity consumption according to the classification.

[0036] Embodiment 1

[0037] A method for classifying and arranging peak-shaving electricity consumption for industrial electricity users based on a clustering algorithm and Fourier transform as features, which classifies and arranges peak-shaving electricity consumption based on the load data of large industrial users in a certain area under a certain power grid, and presents the theoretically peak-shaving results. The specific steps are as follows:

[0038] (1) Obtain the original load data of large industrial users in a certain area from January 2015 to December 2020 in the metering automation system, with a collection frequency of 15 minutes.

[0039] (2) Use the load ratio law to check for outliers, and substitute the data into x i -x i-1 <x i-1×50%. When the difference between the load value at the next moment and the load value at the previous moment exceeds 50% of the load at the previous moment, it is regarded as abnormal data, and it is replaced by calculation according to the data before and after using Newton interpolation method.

[0040] (3) Perform moving average calculation on the original load data through the pandas framework in the python language library, and then use the calculated data as the new data input.

[0041] (4) Perform fast Fourier transform on the modified load data through the scipy.fftpack framework in the python language library, and take the first 6 items of the transformed data as the new data features.

[0042] (5) Use the skit-learn framework in the python language library to establish a K-means algorithm model, input the first 6 items of the data after Fourier transform into the algorithm model, set the distance metric to Euclidean distance to start calculating similarity and clustering, and we set the number of clusters to 4 according to needs.

[0043] (6) Among the 4 categories of users divided, each category of users is further classified according to the electricity consumption level. We set that those with an average electricity load fluctuation below 50,000 megawatts are classified into one category, those with 50,000 - 100,000 megawatts are classified into one category, and those above 100,000 megawatts are classified into one category.

[0044] (7) Take out one category of the users who have completed the secondary classification in the previous step to test the peak-shaving power consumption management. For the three sub-categories of this category of users, half of each sub-category of users are arranged to use electricity at 5 pm, and the other half of the users use electricity as usual. The final result is shown in Figure 5 .

[0045] As can be seen from the above calculation process, an industrial electricity user classification and peak-shaving power consumption method based on clustering algorithm and Fourier transform as features can classify users more reasonably and flexibly, and arrange peak-shaving power consumption, providing a basis for the safe and stable operation of the power system.

[0046] In addition, those skilled in the art can also make other changes within the spirit of the present invention. Of course, these changes made according to the spirit of the present invention should be included within the scope claimed by the present invention.

Claims

1. A method for classifying and staggering power consumption of industrial electricity users with high applicability and high precision, characterized in that, It includes the following steps: Step S10: Collect the user's electricity load data, clean the points where the load change is more than 50% greater than the previous moment, and correct the values using Newton interpolation method; Step S20, perform data smoothing on the cleaned load data using the moving average method. The calculation formula of the moving average method for data smoothing is: Ft = (A t-1 + A t-2 + A t-3 +... + A t-i ) / n, where F t is the moving average value at time t, A t-i is the original data at time t - i, and n is the total number of time instances; by using the moving average method, the original load curve can be transformed into a load trend curve, eliminating some large fluctuations; Step S30: Perform Fourier transform on the load data after data smoothing, and use the transformed data as new data features. The Fourier transform formula is: where F(ω) is the frequency-domain function, f(t) is the time-domain function, t is time, and ω is the independent variable; and for curve waveforms of all lengths, take the same number of leading terms. Step S40: Input the data after feature engineering into the K-means clustering algorithm and tune the model parameters. The K-means clustering algorithm usually uses the Euclidean distance as the metric standard. The Euclidean distance formula in an n-dimensional space is: where x and y are two points in the n-dimensional space respectively; Step S50: Classify the clustered users again according to the load volume to obtain the results; Classify the users in the same category again according to the load volume to make the load more evenly distributed during the off-peak electricity consumption arrangement; Step S60: Group the users in the same category within the class and arrange off-peak electricity consumption.

2. The method for classifying and staggering electricity consumption of industrial electricity users with high applicability and high precision according to claim 1, characterized in that The K-means algorithm further includes the steps: Create k initial values as the initial centroids and they should be located within the data boundaries; if the cluster assignment result of any point changes: Traverse each point in the dataset: Traverse the k centroids: Calculate the distance between the centroid and the data point; Assign the data point to the cluster of the centroid closest to it; Traverse the k clusters: Calculate the mean of all points in each cluster; Update the obtained k means as the new centroids.

3. The high-applicability and high-precision industrial electricity user classification and peak-shaving electricity consumption method according to claim 1, characterized in that The calculation method of Newton interpolation method is: And replace the data more than 10 times the median with the median.

Images