Industrial park load analysis method and system based on optimized K-means clustering

Abstract

The invention discloses an industrial park load analysis method based on an optimized K-means clustering algorithm, and the method comprises the steps: collecting the power load data of an industrial park, including the internal data of the total power consumption of equipment, the time-sharing power consumption, the power consumption peak value, the energy consumption of the equipment and the like, and the external data of weather data, electricity price market information, policies and regulations and the like; the method comprises the following steps of: performing unified standardization processing on data, including cleaning and normalizing the data, and filling by adopting a KNN method aiming at the problems of missing values and errors in the data, so as to avoid the problem of low baseline load analysis accuracy caused by errors in data collection; afterwards, a KPCA nonlinear dimension reduction technology is adopted to process a nonlinear relation between the power demand of the production line and external data, the complexity of a model is reduced by simplifying a data set, more nonlinear features in the external data are extracted, and the rule of the power demand of the industrial park can be better mastered; the problem that a power demand side management method is short in processing a non-linear relation is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of data model identification and energy management in power systems, and more specifically, relates to an industrial park load analysis method and system based on optimized K-means clustering. Background Art

[0002] With the national policy call for energy conservation, emission reduction and green development, industrial parks, as an important area of ​​energy consumption, have important significance for promoting sustainable economic development, improving energy efficiency and achieving environmental protection through precise management of load analysis. According to the power load management mentioned in the "Power Load Management Measures (2023 Edition)", economic, administrative and technical means can be used to regulate, control and optimize the operation of power load. Through in-depth research and precise control of the load characteristics of various equipment and systems in the park, and the reasonable planning and layout of new energy equipment, park managers can allocate energy resources more effectively, reduce unnecessary energy waste, thereby reducing operating costs and improving economic benefits. In addition, refined management of load analysis can also help the park optimize the existing power grid structure and operation mode, better participate in the demand-side response management of the power grid, and improve the stability and reliability of power supply, thereby ensuring the safe and stable operation of the power grid, maintaining a stable power supply and use order, maintaining a balance between supply and demand, and improving energy efficiency. Fine management of load analysis is the key to the construction of smart parks. It provides real-time data support for smart grids, enables the power grid to adjust faster to cope with load changes, ensures the quality of power supply, and lays the foundation for the park to further introduce more high-tech industries in the future, enhancing the attractiveness and competitiveness of the park. Fine management not only ensures the reliability of power supply, but also helps promote the green transformation of industrial parks. Through fine management, industrial parks can not only reduce the consumption of fossil energy, but also promote the integrated application of distributed photovoltaic, wind power and other renewable energy, thereby reducing carbon footprint and contributing to responding to national calls and achieving national emission reduction targets. Therefore, fine management of load analysis in industrial parks is of great significance to ensure efficient use of energy, reduce costs, improve the safety and stability of power supply, and for enterprises to fulfill their social responsibilities and respond to national energy strategies, environmental protection and sustainable development goals.

[0003] There are three common load analysis methods for industrial parks: The first is the maximum demand management method. The electricity cost of the industrial park consists of two parts: the electricity cost of the actual electricity consumption and the basic electricity cost based on the maximum electricity consumption. The maximum electricity consumption is the maximum value of the average power within 15 minutes per month. The park monitors the production line and issues an alarm when the power consumption approaches the maximum value, so as to reduce the load, lower the electricity bill, improve the efficiency of the power grid, and promote energy conservation and emission reduction; the second is the power demand side management (DSM) method, which includes load control and management (LCM) and automatic meter reading and billing (AMR). LCM is responsible for monitoring, managing and controlling the power load of the enterprise, reducing or delaying the use of non-critical equipment during peak hours, and AMR realizes the automation of electricity bill calculation, and automatically generates electricity bill reports and curves, providing detailed electricity information to users, so that enterprises can reduce loads during peak hours and increase loads during off-peak hours, thereby improving the overall efficiency of the power system; the third is the baseline load analysis method, which can detect abnormal conditions in the power system by establishing a baseline load model and comparing the difference between the actual load and the baseline load, such as equipment failure, overload power consumption, irregular power consumption patterns and other abnormal power consumption behaviors. This model can also serve as the basis for predicting future loads, helping industrial parks to better plan the power consumption plans of various enterprises and production lines.

[0004] However, the above three methods for load analysis of industrial parks all have some defects that cannot be ignored: First, with respect to the above maximum demand management method, due to the extensive application of renewable energy sources such as photovoltaics and wind power in modern industrial parks, data collection has added new data modules such as new energy production equipment data, weather conditions data, and dynamic changes in new energy power supply data, which makes it difficult for this method to identify demand peaks; Second, with respect to the above power demand side management method, the relationship between power demand and various environmental factors (such as weather conditions, electricity price market information, etc.) is often nonlinear, and the workload of power demand side management is huge and the content is complex; Third, with respect to the above baseline load analysis method, in actual production in industrial parks, technical factors such as equipment failure and communication interruption may lead to missing historical load data and data errors, which will directly reduce the accuracy and reliability of baseline load analysis. Summary of the invention

[0005] In response to the above defects or improvement needs of the prior art, the present invention provides an industrial park load analysis method and system based on optimized K-means clustering, which aims to solve the technical problem that the existing maximum demand management method has difficulty in identifying demand peaks, the technical problem that the existing power demand side management method has a huge workload and complex content, and the technical problem that the existing baseline load analysis method may result in missing historical load data and data errors due to technical factors such as equipment failure and communication interruption, which directly reduces the accuracy and reliability of baseline load analysis.

[0006] To achieve the above object, according to one aspect of the present invention, a method for analyzing industrial park load based on optimized K-means clustering is provided, comprising the following steps:

[0007] (1) Obtain a power load data set consisting of multiple power load data of an industrial park, and perform standardization processing on the power load data set to obtain a processed power load data set.

[0008] (2) classifying the power load data set processed in step (1) according to the power consumption pattern or power consumption group and using the elbow rule to obtain β optimal cluster categories and their corresponding cluster centers;

[0009] (3) performing K-means clustering analysis on the processed power load data set obtained in step (1) according to the β optimal cluster categories and their corresponding cluster centers obtained in step (2) to obtain an initial clustering result of the power load data set;

[0010] (4) using the KPCA algorithm in the KPCA-KLPP fusion model to process the initial clustering result of the power load data set obtained in step (3) to obtain the main features of the power load data set, using the KLPP algorithm in the KPCA-KLPP fusion model to process the main features of the obtained power load data set to obtain the local structure of the power load data set, and then processing the main features and local structure of the obtained power load data set to obtain the initial clustering features of the power load data set;

[0011] (5) using the DTW method to process the initial cluster features of the power load data set obtained in step (4) to obtain a cumulative distance minimum matrix;

[0012] (6) The clustering results of the power load data set obtained in step (3) are optimized by using the cumulative distance minimum matrix obtained in step (5) to obtain the reallocated clusters. The reallocated clusters are then clustered again using the same K-means clustering analysis method as in step (3) to obtain the final clustering result of the power load data set, which is the analysis result of the industrial park power load data set.

[0013] Preferably, the power load data of the industrial park includes internal data and external data. The internal data includes the historical power consumption data, power consumption peak, power consumption parameters such as power consumption equipment type and energy consumption, power generation data and power generation efficiency of new energy equipment of each enterprise or production line in the industrial park, and the external data includes meteorological data, power and energy market prices, and policy and regulatory information on new energy.

[0014] Preferably, step (1) comprises the following sub-steps:

[0015] (1-1) obtaining a power load data set consisting of a plurality of power load data of an industrial park, and deleting abnormal power load data in the obtained power load data set to obtain a processed power load data set;

[0016] The expression of the power load data set is:

[0017] U(t)=g(t)+z(t)+p(t)

[0018] Among them, U(t) represents the total power load data set within time t, g(t) represents the basic normal power load component within time t; z(t) represents the power load component of abnormal event factors within time t; p(t) represents the power load component of random event factors within time t;

[0019] The formula for the basic normal power load component g(t) within time t is expressed as:

[0020] g(t)=g1(t)+g2(t)

[0021] in:

[0022]

[0023]

[0024] Where g1(t) is the effective value of the AC load in the basic normal power load component within time t, g2(t) is the effective value of the DC load in the basic normal power load component within time t, H and K are weight factors, respectively, and satisfy the equation H+K=1, μ1 and μ2 are abnormal load factors, and μ1∈[0,1], μ2∈[0,1], P is the number of sampling points per week, and its minimum value is 20, U dc (t) is the sampling value of the DC load within time t (i.e. the current moment), U dc (t-1) is the sampling value of the DC load within time t-1, X At is the real parameter of the AC load in time t, X Bk is the imaginary parameter of the AC load in time t, and:

[0025]

[0026] Where P is the number of sampling points per week, and its minimum value is 20. It can be optimized and adjusted according to the loads of different working frequencies. x(d) is the Fourier parameter of the dth real part parameter in the AC load, and d∈[0, P-1];

[0027] The power load component of abnormal event factors refers to the power load component generated by abnormal events such as holidays, large-scale public events and extreme weather conditions;

[0028] The power load component of random event factors presents characteristics such as unpredictability, transience and diversity.

[0029] (1-2) For the power load data set processed in step (1-1), the Euclidean distance between any two of the power load data is calculated, and a plurality of power load data corresponding to the minimum Euclidean distance is selected from the obtained plurality of Euclidean distances;

[0030] (1-3) Use the KNN classification algorithm to process the multiple power load data obtained in step (1-2) to obtain Q most similar power load data; wherein the value range of Q value is an odd number from 1 to 20, preferably 5.

[0031] (1-4) Calculate the weighted average of the Q most similar power load data obtained in step (1-3) to obtain the predicted value of the power load data

[0032] (1-5) using the predicted value of the power load data obtained in step (1-4) to replace the abnormal power load data deleted in step (1-1), and adding all the predicted values ​​to the power load data set processed in step (1-1) to obtain a power load data set after data cleaning;

[0033] (1-6) Performing maximum-minimum normalization processing on the power load data set after data cleaning obtained in step (1-5) to obtain a processed power load data set.

[0034] Preferably, the predicted value in step (1-4) is:

[0035]

[0036] where ω r is the weight of the rth power load data among the Q most similar power load data obtained in step (1-3), q r It is the first r eigenvalues, and r∈[1,Q].

[0037] The maximum-minimum normalization process in steps (1-6) is performed using the following formula:

[0038]

[0039] Wherein b is the power load data after data cleaning, min(b) and max(b) are the minimum and maximum values ​​of the power load data after data cleaning obtained in step (1-5), respectively, and b' is the processed power load data.

[0040] Preferably, step (3) is specifically as follows: first, arbitrarily select β power load data from the processed power load data set obtained in step (1); then, calculate the Euclidean distance between the first power load data and each of the β cluster centers obtained in step (2), select the cluster center corresponding to the minimum value from the multiple Euclidean distances obtained, and classify the power load data into the cluster category corresponding to the cluster center; then, repeat the above process for the remaining power load data in the β power load data, so as to obtain the initial cluster division result.

[0041] Preferably, step (4) comprises the following sub-steps:

[0042] (4-1) Process the power load data set obtained in step (3) to obtain a data sample set Y = (y1, y2, ..., y M ), y M ∈R N , and its data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )}, where R N Represents an N-dimensional space composed of real numbers, N is an arbitrary positive integer, and M represents the total number of data samples in the data sample set;

[0043] (4-2) The data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e Y ), processing the obtained projection data sample set and projection matrix to obtain the objective function equations of the KPCA algorithm and the KLPP algorithm, as well as the overall objective function equation;

[0044] (4-3) establishing a kernel function based on the data sample set mapped to the high-dimensional space obtained in step (4-1), standardizing the established kernel function to obtain its corresponding, standardized kernel matrix, and simplifying the overall objective function equation obtained in step (4-2) based on the obtained standardized kernel matrix to obtain a simplified overall objective function equation;

[0045] (4-4) The simplified overall objective function equation obtained in step (4-3) is processed to obtain a Lagrangian function expression, and the obtained Lagrangian function expression is differentiated and set to zero to obtain the eigenvalue set A = {a1, a2, ... a M} and its corresponding eigenvectors s1,s2,…s M , process the obtained eigenvalue set and its corresponding eigenvector to obtain the contribution rate of each eigenvalue, arrange the obtained contribution rates in descending order, add the first contribution rate to the second contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=2, otherwise add the first contribution rate to the second contribution rate and the third contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise add the first contribution rate to the second, third, and fourth contribution rates, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise repeat the above steps, ..., and so on, so as to finally obtain the value of n, and obtain the eigenvectors s1, s2, ...s M Select the first n eigenvectors to form the feature matrix S2 = [s1, s2, ...s n ]; According to the obtained characteristic matrix S2 = [s1, s2, ... s n ], and the projection data sample set X corresponding to the low-dimensional space obtained in step (4-2) = (x1, x2, ..., x M ) and its projection matrix E(e1,e2,…e M ), obtain the updated projection data sample set X'=(x1,x2,…,xM ).

[0046] Preferably, step (4-2) is specifically:

[0047] First, the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e M ):

[0048] X=E T φ(Y)

[0049] The superscript T represents the transpose of the matrix;

[0050] Afterwards, the objective function equation of the KPCA algorithm is established based on the obtained projection data sample set and projection matrix;

[0051]

[0052] Among them J global Represents the objective function equation constructed by the KPCA algorithm, and max represents the maximum value;

[0053] Then, the data sample set Y = (y1, y2, ..., y M ) construct a similarity matrix W, obtain a diagonal matrix D according to the similarity matrix W, and obtain a Laplace matrix L according to the diagonal matrix D;

[0054] The similarity matrix W is:

[0055]

[0056] Among them, W m,l represents the mth data sample y in the data sample set obtained in step (4-1) m and the lth data sample y l The similarity between is the mth data sample y m The θ neighborhood set of the m-th data sample is the set of θ data samples closest to the m-th data sample in the data sample set; γ is a positive real number used to adjust the distribution of similarity to ensure that the elements of the similarity matrix are within a reasonable range, m∈[1,M], l∈[1,M], θ∈[1,M], and e represents a natural constant with a value of approximately 2.71828;

[0057] A diagonal matrix D is a matrix whose elements outside the main diagonal are all 0, and whose element D in the mth row and mth column is mm for:

[0058]

[0059] The Laplace matrix L is:

[0060] L=DW

[0061] Afterwards, the objective function equation of the KLPP algorithm is established based on the obtained similarity matrix W, diagonal matrix D, Laplace matrix L, projection data sample set and projection matrix:

[0062] J local (E) = minE T φ(Y)(DW)φ T (Y)E

[0063] =minE T φ(Y)Lφ T (Y)E

[0064] Among them J local It is the objective function equation constructed by the KLPP algorithm, and min means taking the minimum value;

[0065] Finally, according to the objective function equations of the KPCA algorithm and the KLPP algorithm, the overall objective function equation is obtained:

[0066]

[0067] Preferably, step (4-3) is specifically:

[0068] First, according to the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )}Build the kernel function:

[0069] K ij =φ(y i )φ(y j )

[0070] Where φ(y i ) and φ(y i ) are respectively mapped to the i-th and j-th data samples in the high-dimensional space data sample set, K ij Represents the kernel function value between the i-th data sample and the j-th data sample in the image of M data samples mapped to the high-dimensional space, and i∈[1,M], j∈[1,M].

[0071] Then, the corresponding kernel matrix is ​​obtained according to the obtained kernel function:

[0072] K=[K ij ] M×M

[0073] Then, the obtained kernel matrix is ​​normalized to obtain a normalized kernel matrix:

[0074]

[0075] in, is the kernel matrix after normalization, I M is an M×M matrix whose elements are

[0076] Finally, the overall objective function equation obtained in step (4-2) is simplified according to the obtained standardized kernel matrix to obtain the simplified overall objective function equation:

[0077]

[0078] Preferably, in step (4-4), the Lagrangian function expression is:

[0079]

[0080] Among them, λ is the Lagrange multiplier, LA(E,λ) is the Lagrange function;

[0081] In step (4-4), the contribution rate of each eigenvalue in the eigenvector matrix is ​​calculated as:

[0082]

[0083] where ρ τ is the contribution rate of the eigenvalue of the τth normalized kernel matrix, a τ is the eigenvalue of the τth normalized kernel matrix, and τ∈[1,M];

[0084] In step (4-4), the updated projection data sample set expression is:

[0085]

[0086] Where X' is the updated projection data sample set, whose dimension is smaller than the dimension of the data sample set Y. The updated projection data sample set X' is the initial cluster feature of the power load data set.

[0087] According to another aspect of the present invention, there is provided an industrial park load analysis system based on optimized K-means clustering, comprising:

[0088] The first module is used to obtain a power load data set composed of multiple power load data of the industrial park, and perform standardization processing on the power load data set to obtain a processed power load data set.

[0089] The second module is used to classify the power load data set processed by the first module according to the power consumption mode or power consumption group and use the elbow rule to obtain β optimal cluster categories and their corresponding cluster centers;

[0090] The third module is used to perform K-means clustering analysis on the processed power load data set obtained by the first module according to the β optimal cluster categories and their corresponding cluster centers obtained by the second module, so as to obtain an initial cluster division result of the power load data set;

[0091] The fourth module is used to process the initial clustering result of the power load data set obtained by the third module using the KPCA algorithm in the KPCA-KLPP fusion model to obtain the main features of the power load data set, and use the KLPP algorithm in the KPCA-KLPP fusion model to process the main features of the obtained power load data set to obtain the local structure of the power load data set, and then process the main features and local structure of the obtained power load data set to obtain the initial clustering features of the power load data set;

[0092] The fifth module is used to process the initial cluster characteristics of the power load data set obtained in the fourth module using the DTW method to obtain a cumulative distance minimum value matrix;

[0093] The sixth module is used to use the cumulative distance minimum matrix obtained in the fifth module to perform secondary cluster category optimization processing on the cluster division results of the power load data set obtained in the third module to obtain the reallocated clusters, and use the same K-means clustering analysis method as the third module to perform secondary clustering processing on the reallocated clusters to obtain the final clustering results of the power load data set, which is the analysis result of the industrial park power load data set.

[0094] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0095] (1) Since the present invention adopts step (4), it uses the Kernel Principal Component Analysis (KPCA)-Kernel Locality Preserving Projection (KLPP) fusion model to optimize the K-means clustering algorithm. Through the data dimension reduction optimization technology, high-dimensional data can be mapped to a low-dimensional space to reduce the complexity of the data. The optimized K-means algorithm can identify different power-consuming equipment and their typical power consumption patterns by clustering the power load data, and can effectively process newly added data modules such as new energy and environment. Therefore, it can solve the technical problem that the existing maximum demand management method is difficult to effectively identify and manage demand peaks under high data dimensionality and diversified power consumption patterns;

[0096] (2) Since the present invention adopts step (4), it uses the KPCA-KLPP fusion model to optimize the K-means clustering algorithm. The algorithm can better capture the nonlinear characteristics in the data by combining with the KPCA algorithm technology, thereby solving the technical problems of huge workload and complex content in the existing power demand side management method;

[0097] (3) Since the present invention adopts steps (1) to (3), it clusters each power data set through the K-means clustering algorithm, identifies the erroneous data caused by equipment failure or technical factors and performs data cleaning, removes the erroneous data and fills the missing values ​​with the KNN method. Therefore, it can solve the problem of missing historical load data and data errors caused by technical factors such as equipment failure and communication interruption in the baseline load analysis method, which will directly reduce the accuracy and reliability of the baseline load analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is an overall flow chart of the industrial park load analysis method based on the KPCA-KLPP fusion model optimization K-means clustering algorithm of the present invention;

[0099] Figure 2 is a flow chart of extracting initial cluster features using the KPCA-KLPP fusion model in step (4) of the method of the present invention;

[0100] Figure 3 is a schematic diagram of using the elbow rule in step (2) of the method of the present invention to determine the optimal number of cluster categories;

[0101] Figure 4 is a schematic diagram of a data sample set mapped to a high dimension in step (4-1) of the method of the present invention;

[0102] Figure 5 It is a schematic diagram of the data dimensionality reduction effect of using the KPCA-KLPP fusion model in step (4-4) of the method of the present invention. DETAILED DESCRIPTION

[0103] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0104] They can be combined with each other as long as there is no conflict between them.

[0105] The present invention proposes an industrial park load analysis method based on an optimized K-means clustering algorithm. The method optimizes the K-means clustering algorithm from three aspects, thereby improving the accuracy and efficiency of the industrial park load analysis. First, after collecting the power load data of the industrial park, including internal data such as total power consumption of equipment, time-sharing power consumption, peak power consumption, equipment energy consumption, and external data such as weather data, electricity price market information, policies and regulations, etc., they are uniformly standardized, including data cleaning and normalization. For missing values ​​and errors in the data, the KNN method is used to fill them, so as to avoid the problem of low accuracy of baseline load analysis caused by errors in data collection; then, the KPCA nonlinear dimensionality reduction technology is used to process the nonlinear relationship between the power demand of the production line and external data (weather data, electricity price market data, new energy policy data, etc.), and the complexity of the model is reduced by simplifying the data set, and more nonlinear data in the external data are extracted. The characteristics of the industrial park electricity demand are used to better understand the law of power demand in the industrial park and solve the shortcomings of the power demand side management method in dealing with nonlinear relationships. Subsequently, in view of the limitations of the maximum demand management method in dealing with high-dimensional data and identifying multiple power consumption patterns, the KPCA-KLPP joint optimization model is used to extract the eigenvalues ​​of the initial clustering results, and the global characteristics of the data set are described by KPCA, and the local domain information preservation characteristics of the data set are performed by KLPP. The joint optimization model of the two can retain the characteristic information of the data (such as the peak-to-valley difference, load rate, and load curve of the power system) to a greater extent through data dimensionality reduction, and identify the typical power consumption patterns of various equipment; then, the dynamic time warping method (Dynamic Time Warping, referred to as DTW) is used to perform similarity analysis on the extracted feature information and establish the cumulative distance minimum matrix, and then the K-means clustering algorithm is optimized for the second time. The defects of the K-means clustering algorithm are improved, the clustering quality of K-means is significantly improved, and the stability and robustness of data analysis are significantly improved. After obtaining the final clustering results, the electricity consumption patterns of different types of loads in the industrial park can be analyzed, and enterprises or production lines with similar electricity consumption can be identified and unified in demand response management, so as to better grasp the dynamic changes in the electricity demand of each enterprise and achieve more effective power resource management and optimized scheduling in the industrial park.

[0106] like Figure 1 As shown, the present invention provides an industrial park load analysis method based on optimized K-means clustering, comprising the following steps:

[0107] (1) Obtain a power load data set consisting of multiple power load data of an industrial park, and perform standardization processing on the power load data set to obtain a processed power load data set.

[0108] Specifically, the power load data of industrial parks include internal data and external data. Internal data include historical power consumption data of each enterprise or production line in the industrial park (including total power consumption and time-sharing power consumption on a daily, weekly, monthly and annual basis), peak power consumption (including the highest power consumption period and maximum power consumption on a daily, weekly and monthly basis), power consumption parameters such as power equipment type and energy consumption, power generation data and power generation efficiency of new energy equipment, and external data include meteorological data (sunshine time, light intensity, temperature and humidity, etc.), power and energy market prices, and information on policies and regulations on new energy.

[0109] In order to eliminate the impact of dimension and improve the accuracy of the algorithm, the acquired power load data needs to be standardized.

[0110] Specifically, standardization processing includes data cleaning and maximum-minimum normalization. Data cleaning refers to deleting outliers in the data and filling missing values ​​with the K-Nearest Neighbors (KNN) method. Maximum-minimum normalization can deal with the scale problem of eigenvalues, making the numerical ranges of different features consistent, which is convenient for comparison and calculation. Since external factor data (meteorological data, electricity and energy market prices, and new energy policies and regulations, etc.) will affect the use and electricity consumption of power equipment, it is necessary to unify the scale of the various external data sources and internal power equipment usage data samples, and conduct similarity analysis and comprehensive evaluation.

[0111] Specifically, this step includes the following sub-steps:

[0112] (1-1) obtaining a power load data set consisting of a plurality of power load data of an industrial park, and deleting abnormal power load data in the obtained power load data set to obtain a processed power load data set;

[0113] Specifically,

[0114] The expression of the power load data set is:

[0115] U(t)=g(t)+z(t)+p(t)

[0116] Among them, U(t) represents the power load data set within time t, g(t) represents the basic normal power load component within time t; z(t) represents the power load component of abnormal event factors within time t; p(t) represents the power load component of random event factors within time t, and the value of t is preferably 24h or 168h;

[0117] Abnormal power load data in the power load data refers to power load data that is significantly deviated from other power load data in the power load data set, which may be caused by equipment failure, measurement error or other unknown factors.

[0118] The formula for the basic normal power load component g(t) within time t is expressed as:

[0119] g(t)=g1(t)+g2(t)

[0120] in:

[0121]

[0122] Where g1(t) is the effective value of the AC load in the basic normal power load component within time t, g2(t) is the effective value of the DC load in the basic normal power load component within time t, H and K are weight factors, respectively, and satisfy the equation H+K=1, μ1 and μ2 are abnormal load factors, and μ1∈[0,1], μ2∈[0,1], P is the number of sampling points per week, and its minimum value is 20, U dc (t) is the sampling value of the DC load within time t (i.e. the current moment), U dc (t-1) is the sampling value of the DC load within time t-1, X Ak is the real parameter of the AC load in time t, X Bk is the imaginary parameter of the AC load in time t, and:

[0123]

[0124] Where P is the number of sampling points in a week, and its minimum value is 20. It can be optimized and adjusted according to the load of different working frequencies. x(d) is the Fourier parameter of the dth real parameter in the AC load, and d∈[0, P-1]. The variable sampling period Fourier algorithm can improve the stability of the load sampling process and effectively filter out various types of harmonic interference. With the addition of variable sampling points, the algorithm can better adapt to frequency changes and the calculated results are more accurate.

[0125] The power load component of abnormal event factors refers to the power load component generated by abnormal events such as holidays, large-scale public events and extreme weather conditions;

[0126] The power load component of random event factors presents characteristics such as unpredictability, transience and diversity.

[0127] (1-2) For the power load data set processed in step (1-1), the Euclidean distance between any two of the power load data is calculated, and a plurality of power load data corresponding to the minimum Euclidean distance is selected from the obtained plurality of Euclidean distances;

[0128] (1-3) using a KNN classification algorithm to process the multiple power load data obtained in step (1-2) to obtain Q most similar power load data;

[0129] Specifically, the Q value ranges from an odd number of 1 to 20, preferably 5.

[0130] (1-4) Calculate the weighted average of the Q most similar power load data obtained in step (1-3) to obtain the predicted value of the power load data

[0131] Specifically, the predicted values ​​are:

[0132]

[0133] Among them, ω r is the weight of the rth power load data among the Q most similar power load data obtained in step (1-3), q r It is the rth eigenvalue in the power load data, and r∈[1,Q].

[0134] (1-5) using the predicted value of the power load data obtained in step (1-4) to replace the abnormal power load data deleted in step (1-1), and adding all the predicted values ​​to the power load data set processed in step (1-1) to obtain a power load data set after data cleaning;

[0135] (1-6) performing maximum-minimum normalization processing on the power load data set after data cleaning obtained in step (1-5) to obtain a processed power load data set;

[0136] Specifically, the maximum-minimum normalization process uses the following formula:

[0137]

[0138] Wherein: b is the power load data after data cleaning, min(b) and max(b) are the minimum and maximum values ​​of the power load data after data cleaning obtained in step (1-5), and b' is the processed power load data.

[0139] (2) Classify the power load data set processed in step (1) according to the power consumption mode or power consumption group and use the elbow method to obtain β optimal cluster categories and their corresponding cluster centers (such as Figure 4 shown);

[0140] The advantage of the above step (2) is that the elbow rule provides a visual method to evaluate the impact of different β values ​​on the clustering effect, and determines the optimal cluster category based on the characteristics of the power load data itself. It can be both intuitive and objective, reducing the influence of human factors, making the clustering results more objective and reliable, and providing data support for industrial parks to identify different types of power consumption patterns or user groups, simplify data, and provide targeted power supply services and strategies for park managers.

[0141] (3) performing K-means clustering analysis on the processed power load data set obtained in step (1) according to the β optimal cluster categories and their corresponding cluster centers obtained in step (2) to obtain an initial clustering result of the power load data set;

[0142] Specifically, first, β power load data are randomly selected from the processed power load data set obtained in step (1); then, the Euclidean distance between the first power load data and each of the β cluster centers obtained in step (2) is calculated, and the cluster center corresponding to the minimum value is selected from the multiple Euclidean distances obtained, and the power load data is classified into the cluster category corresponding to the cluster center; then, the above process is repeated for the remaining power load data in the β power load data to obtain the initial cluster division result.

[0143] In the power load analysis of industrial parks, the optimal cluster categories were determined, which can obtain different power consumption patterns after β initial classifications, including traditional high-energy-consuming user groups (such as clusters with large power consumption and peak time concentrated in daytime working hours), distributed energy user groups (such as clusters where power consumption fluctuates with lighting conditions and may have a net power consumption of zero or even negative (i.e., supplying power to the grid)), and electric vehicle charging station user groups (such as clusters where power consumption fluctuates with charging demand, especially during peak hours in the morning and evening (commuting time) and at night (parking and charging time)).

[0144] (4) using the KPCA algorithm in the KPCA-KLPP fusion model to process the initial clustering result of the power load data set obtained in step (3) to obtain the main features of the power load data set, using the KLPP algorithm in the KPCA-KLPP fusion model to process the main features of the obtained power load data set to obtain the local structure of the power load data set, and then processing the main features and local structure of the obtained power load data set to obtain the initial clustering features of the power load data set;

[0145] This step uses the KPCA algorithm and the KLPP algorithm to establish a fused data feature extraction model. The KPCA algorithm is used to determine the global objective function, and the main features are found from the most representative feature dimensions of the original multi-dimensional power load data set, such as the power consumption data features of the time series, the data features of various power consumption indicators, and the environmental data features; based on KPCA, the local objective function is determined by KLPP, and further optimization is performed from the above main features to retain the local structure of the power load data set. The two parts jointly establish a new objective function to extract the initial cluster features of the power load data set.

[0146] The above process of using the KPCA-KLPP fusion model to obtain the initial cluster characteristics of the power load data set is as follows: Figure 2 shown.

[0147] Specifically, this step includes the following sub-steps:

[0148] (4-1) Process the power load data set obtained in step (3) to obtain a data sample set Y = (y1, y2, ..., y M ), y M ∈R N , and its data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )}, where R N Represents an N-dimensional space composed of real numbers, N is an arbitrary positive integer, and M represents the total number of data samples in the data sample set;

[0149] (4-2) The data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e Y ), processing the obtained projection data sample set and projection matrix to obtain the objective function equations of the KPCA algorithm and the KLPP algorithm, as well as the overall objective function equation;

[0150] Specifically, this step is as follows: first, the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e M ):

[0151] X=ET φ(Y) (3)

[0152] The superscript T represents the transpose of the matrix;

[0153] Afterwards, the objective function equation of the KPCA algorithm is established based on the obtained projection data sample set and projection matrix;

[0154] The objective function equation of the KPCA algorithm is shown in equation (4):

[0155]

[0156] Among them J global Represents the objective function equation constructed by the KPCA algorithm, and max represents the maximum value;

[0157] Then, the data sample set Y = (y1, y2, ..., y M ) construct a similarity matrix W, obtain a diagonal matrix D according to the similarity matrix W, and obtain a Laplace matrix L according to the diagonal matrix D;

[0158] In this step, the similarity matrix W is shown in the following formula (5):

[0159]

[0160] Among them, W m,l represents the mth data sample y in the data sample set obtained in step (4-1) m and the lth data sample y l The similarity between is the mth data sample y m The θ neighborhood set of the m-th data sample is the set of θ data samples closest to the m-th data sample in the data sample set; γ is a positive real number used to adjust the distribution of similarity to ensure that the elements of the similarity matrix are within a reasonable range, m∈[1,M], l∈[1,M], θ∈[1,M], and e represents a natural constant with a value of approximately 2.71828;

[0161] In this step, the diagonal matrix D is a matrix whose elements outside the main diagonal are all 0, and its elements in the mth row and mth column (i.e., the elements on the diagonal) D mm for:

[0162]

[0163] In this step, the Laplace matrix L is shown in the following formula (7):

[0164] L=DW (7)

[0165] Afterwards, the objective function equation of the KLPP algorithm is established based on the obtained similarity matrix W, diagonal matrix D, Laplace matrix L, projection data sample set and projection matrix as shown in the following equation (8):

[0166]

[0167] Among them J local It is the objective function equation constructed by the KLPP algorithm, and min means taking the minimum value;

[0168] Finally, according to the objective function equations of the KPCA algorithm and the KLPP algorithm, the overall objective function equation is obtained as shown in the following equation (9):

[0169]

[0170] (4-3) establishing a kernel function based on the data sample set mapped to the high-dimensional space obtained in step (4-1), standardizing the established kernel function to obtain its corresponding, standardized kernel matrix, and simplifying the overall objective function equation obtained in step (4-2) based on the obtained standardized kernel matrix to obtain a simplified overall objective function equation;

[0171] Specifically, this step is as follows: first, according to the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} The kernel function is established as shown in the following formula (10):

[0172] K ij =φ(y i )φ(y j ) (10)

[0173] Where φ(y i ) and φ(y i ) are respectively mapped to the i-th and j-th data samples in the high-dimensional space data sample set, K ij Represents the kernel function value between the i-th data sample and the j-th data sample in the image of M data samples mapped to the high-dimensional space, and i∈[1,M], j∈[1,M].

[0174] Then, the corresponding kernel matrix is ​​obtained according to the obtained kernel function as shown in the following formula (11):

[0175] K=[K ij ] M×M (11)

[0176] Then, the obtained kernel matrix is ​​normalized to obtain the normalized kernel matrix as shown in the following formula (12):

[0177]

[0178] in, is the kernel matrix after normalization, I M is an M×M matrix whose elements are

[0179] Finally, the overall objective function equation obtained in step (4-2) is simplified according to the obtained standardized kernel matrix to obtain the simplified overall objective function equation as shown in the following equation (13):

[0180]

[0181] (4-4) The simplified overall objective function equation obtained in step (4-3) is processed to obtain a Lagrangian function expression, and the obtained Lagrangian function expression is differentiated and set to zero to obtain the eigenvalue set A = {a1, a2, ... a M} and its corresponding eigenvectors s1,s2,…s M , process the obtained eigenvalue set and its corresponding eigenvector to obtain the contribution rate of each eigenvalue, arrange the obtained contribution rates in descending order, add the first contribution rate to the second contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=2, otherwise add the first contribution rate to the second contribution rate and the third contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise add the first contribution rate to the second, third, and fourth contribution rates, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise repeat the above steps, ..., and so on, so as to finally obtain the value of n, and obtain the eigenvectors s1, s2, ...s M Select the first n eigenvectors to form the feature matrix S2 = [s1, s2, ...s n ]; According to the obtained characteristic matrix S2 = [s1, s2, ... s n ], and the projection data sample set X corresponding to the low-dimensional space obtained in step (4-2) = (x1, x2, ..., x M ) and its projection matrix E(e1,e2,…e M ), obtain the updated projection data sample set X'=(x1,x2,…,x M ).

[0182] In this step, the Lagrangian function expression is shown in the following formula (14):

[0183]

[0184] Among them, λ is the Lagrange multiplier, LA(E,λ) is the Lagrange function;

[0185] In this step, the contribution rate of each eigenvalue in the eigenvector matrix is ​​calculated as shown in the following formula (15):

[0186]

[0187] where ρ τ is the contribution rate of the eigenvalue of the τth normalized kernel matrix, a τ is the eigenvalue of the τth normalized kernel matrix, and τ∈[1,M];

[0188] In this step, the updated projection data sample set expression is shown in the following formula (16):

[0189]

[0190] Where X' is the updated projection data sample set, whose dimension is smaller than the dimension of the data sample set Y. The updated projection data sample set X' is the initial cluster feature of the power load data set.

[0191] The advantage of the above step (4) is that the KPCA-KLPP fusion model can ensure the global structural characteristics of the power load data set while maintaining the local neighborhood structure, making the feature extraction more comprehensive and retaining important feature information. The main features extracted by the KPCA part include: changes in power consumption patterns in different time periods (such as the power consumption period of traditional high-energy-consuming users, the power consumption period of distributed energy users, and the power consumption period of electric vehicle charging station users) and seasonal power consumption patterns, the duration of daily power consumption peaks, and the nonlinear effects of external factors such as meteorological conditions and electricity price strategies on power consumption; the local structures extracted by the KLPP part include: changes in power consumption patterns in the short term, individual user power consumption habits, and the capture of local abnormal conditions.

[0192] (5) using the DTW method to process the initial cluster features of the power load data set obtained in step (4) to obtain a cumulative distance minimum matrix;

[0193] The advantage of the above step (5) is that using the DTW algorithm to optimize the K-means clustering algorithm is an efficient method for processing time series data. It can effectively handle the clustering problem of time series data, better capture the dynamic characteristics of time series data, and identify different types of power load patterns. By improving the accuracy and efficiency of the clustering algorithm, it can further analyze the impact of external data (weather data, electricity price information data, new energy policy data, etc.) on power load, making data matching more accurate and enhancing model robustness.

[0194] (6) The clustering results of the power load data set obtained in step (3) are optimized by using the cumulative distance minimum matrix obtained in step (5) to obtain the reallocated clusters, and the reallocated clusters are clustered again by using the K-means clustering analysis method in step (3) to obtain the final clustering result of the power load data set, which is the analysis result of the industrial park power load data set.

[0195] Specifically, the analysis results of the industrial park power load data set obtained in this step can more accurately identify the power consumption pattern in the power load data, thereby assisting the decision-making of power system operation and planning, optimizing resource scheduling and allocation, and formulating more reasonable power supply plans and service strategies for industrial parks to achieve the goal of energy conservation and emission reduction.

[0196] Experimental Results

[0197] Figure 3 In order to establish the optimal number of clusters for the initial optimization of the K-means clustering algorithm based on the elbow rule in Matlab, a curve of the Within-Cluster Sum of Squares (WCSS) versus the β value is established. As the cluster category β increases, the WCSS decreases, so the power load data is more likely to be assigned to a more adjacent cluster category. However, when β increases to a certain extent, the rate of decline of WCSS slows down significantly, forming an elbow point. This point is usually considered to be the optimal number of cluster categories.

[0198] Figure 5 This is a diagram of the dimensionality reduction effect of the KPCA-KLPP fusion model on the data sample set mapped to a high dimension. It can be seen that the power load data is not only maintained in the global structure, but also well preserved in the local neighborhood structure. After dimensionality reduction, K-means clustering is performed. The visualization of the clustering results may show that different categories of power load data are clearly distinguished to form different clusters. The boundaries of these clusters are clear, indicating that the dimensionality reduction process helps to improve the accuracy of the clustering results.

[0199] The above simulation results show that compared with the traditional industrial park load analysis method, the industrial park load analysis method based on optimized K-means clustering proposed in the present invention improves the accuracy of load pattern recognition on the basis of ensuring system stability and optimizing the performance of K-means clustering algorithm, and has the function of abnormal power consumption pattern detection and diagnosis. Through the optimized clustering algorithm results, it can help industrial parks to better optimize resource allocation and achieve more efficient energy utilization. For identifying the power consumption behavior patterns of different enterprises, it can provide more targeted energy-saving measures and personalized service opinions for each enterprise.

[0200] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An industrial park load analysis method based on optimized K-means clustering, characterized in that: The following steps are involved: (1) Obtain a power load data set consisting of multiple power load data of an industrial park, and perform standardization processing on the power load data set to obtain a processed power load data set. (2) classifying the power load data set processed in step (1) according to the power consumption pattern or power consumption group and using the elbow rule to obtain β optimal cluster categories and their corresponding cluster centers; (3) performing K-means clustering analysis on the processed power load data set obtained in step (1) according to the β optimal cluster categories and their corresponding cluster centers obtained in step (2) to obtain an initial clustering result of the power load data set; (4) using the KPCA algorithm in the KPCA-KLPP fusion model to process the initial clustering result of the power load data set obtained in step (3) to obtain the main features of the power load data set, using the KLPP algorithm in the KPCA-KLPP fusion model to process the main features of the obtained power load data set to obtain the local structure of the power load data set, and then processing the main features and local structure of the obtained power load data set to obtain the initial clustering features of the power load data set; (5) using the DTW method to process the initial cluster features of the power load data set obtained in step (4) to obtain a cumulative distance minimum matrix; (6) The clustering results of the power load data set obtained in step (3) are optimized by using the cumulative distance minimum matrix obtained in step (5) to obtain the reallocated clusters. The reallocated clusters are then clustered again using the same K-means clustering analysis method as in step (3) to obtain the final clustering result of the power load data set, which is the analysis result of the industrial park power load data set.

2. The industrial park load analysis method based on optimized K-means clustering according to claim 1 is characterized in that: The power load data of the industrial park includes internal data and external data. The internal data includes the historical power consumption data, peak power consumption, power consumption parameters such as power equipment type and energy consumption, power generation data and power generation efficiency of each enterprise or production line in the industrial park, and the external data includes meteorological data, power and energy market prices, and policy and regulatory information on new energy. The expression of the power load data set is: U(t)=g(t)+z(t)+p(t) Among them, U(t) represents the power load data set within time t, g(t) represents the basic normal power load component within time t; z(t) represents the power load component of abnormal event factors within time t; p(t) represents the power load component of random event factors within time t, and the value of t is preferably 24h or 168h; The formula for the basic normal power load component g(t) within time t is expressed as: g(t)=g1(t)+g2(t) in: Where g1(t) is the effective value of the AC load in the basic normal power load component within time t, g2(t) is the effective value of the DC load in the basic normal power load component within time t, H and K are weight factors, respectively, and satisfy the equation H+K=1, μ1 and μ2 are abnormal load factors, and μ1∈[0,1], μ2∈[0,1], P is the number of sampling points per week, and its minimum value is 20, U dc (t) is the sampling value of DC load within time t, U dc (t-1) is the sampling value of the DC load within time t-1, X Ak is the real parameter of the AC load in time t, X Bk is the imaginary parameter of the AC load in time t, and: Where P is the number of sampling points in a week, and its minimum value is 20. x(d) is the Fourier parameter of the dth real parameter in the AC load, and d∈[0, P-1].

3. The industrial park load analysis method based on optimized K-means clustering according to claim 1 or 2, characterized in that: Step (1) includes the following sub-steps: (1-1) obtaining a power load data set consisting of a plurality of power load data of an industrial park, and deleting abnormal power load data in the obtained power load data set to obtain a processed power load data set; (1-2) For the power load data set processed in step (1-1), the Euclidean distance between any two of the power load data is calculated, and a plurality of power load data corresponding to the minimum Euclidean distance is selected from the obtained plurality of Euclidean distances; (1-3) using a KNN classification algorithm to process the multiple power load data obtained in step (1-2) to obtain Q most similar power load data; The Q value ranges from 1 to 20, preferably 5. (1-4) Calculate the weighted average of the Q most similar power load data obtained in step (1-3) to obtain the predicted value of the power load data (1-5) using the predicted value of the power load data obtained in step (1-4) to replace the abnormal power load data deleted in step (1-1), and adding all the predicted values ​​to the power load data set processed in step (1-1) to obtain a power load data set after data cleaning; (1-6) Performing maximum-minimum normalization processing on the power load data set after data cleaning obtained in step (1-5) to obtain a processed power load data set.

4. The industrial park load analysis method based on optimized K-means clustering according to any one of claims 1 to 3, characterized in that: The predicted values ​​in steps (1-4) are: where ω r is the weight of the rth power load data among the Q most similar power load data obtained in step (1-3), q r It is the first r eigenvalues, and r∈[1,Q]. The maximum-minimum normalization process in steps (1-6) is performed using the following formula: Wherein b is the power load data after data cleaning, min(b) and max(b) are the minimum and maximum values ​​of the power load data after data cleaning obtained in step (1-5), respectively, and b' is the processed power load data.

5. The industrial park load analysis method based on optimized K-means clustering according to claim 4 is characterized in that: Step (3) specifically comprises the following steps: first, arbitrarily selecting β power load data from the processed power load data set obtained in step (1); then, calculating the Euclidean distance between the first power load data and each of the β cluster centers obtained in step (2), selecting the cluster center corresponding to the minimum value from the obtained multiple Euclidean distances, and classifying the power load data into the cluster category corresponding to the cluster center; Subsequently, the above process is repeated for the remaining power load data among the β power load data, thereby obtaining an initial cluster division result.

6. The industrial park load analysis method based on optimized K-means clustering according to claim 5 is characterized in that: Step (4) includes the following sub-steps: (4-1) Process the power load data set obtained in step (3) to obtain a data sample set Y = (y1, y2, ..., y M ), y M ∈R N , and its data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )}, where R N Represents an N-dimensional space composed of real numbers, N is an arbitrary positive integer, and M represents the total number of data samples in the data sample set; (4-2) The data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e Y ), processing the obtained projection data sample set and projection matrix to obtain the objective function equations of the KPCA algorithm and the KLPP algorithm, as well as the overall objective function equation; (4-3) establishing a kernel function based on the data sample set mapped to the high-dimensional space obtained in step (4-1), standardizing the established kernel function to obtain its corresponding, standardized kernel matrix, and simplifying the overall objective function equation obtained in step (4-2) based on the obtained standardized kernel matrix to obtain a simplified overall objective function equation; (4-4) The simplified overall objective function equation obtained in step (4-3) is processed to obtain a Lagrangian function expression, and the obtained Lagrangian function expression is differentiated and set to zero to obtain the eigenvalue set A = {a1, a2, ... a M } and its corresponding eigenvectors s1,s2,…s M , process the obtained eigenvalue set and its corresponding eigenvector to obtain the contribution rate of each eigenvalue, arrange the obtained contribution rates in descending order, add the first contribution rate to the second contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=2, otherwise add the first contribution rate to the second contribution rate and the third contribution rate, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise add the first contribution rate to the second, third, and fourth contribution rates, and judge whether the sum is greater than 85%. If so, end the judgment and set n=3, otherwise repeat the above steps, ..., and so on, so as to finally obtain the value of n, and obtain the eigenvectors s1, s2, ...s M Select the first n eigenvectors to form the feature matrix S2 = [s1, s2, ...s n ]; According to the obtained characteristic matrix S2 = [s1, s2, ... s n ], and the projection data sample set X corresponding to the low-dimensional space obtained in step (4-2) = (x1, x2, ..., x M ) and its projection matrix E(e1,e2,…e M ), obtain the updated projection data sample set X'=(x1,x2,…,x M ).

7. The industrial park load analysis method based on optimized K-means clustering according to claim 6 is characterized in that: Step (4-2) is specifically: First, the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )} to perform projection processing to obtain the projection data sample set X=(x1,x2,…,x M ) and its projection matrix E(e1,e2,…e M ): X=E T φ(Y) The superscript T represents the transpose of the matrix; Afterwards, the objective function equation of the KPCA algorithm is established based on the obtained projection data sample set and projection matrix; Among them J global Represents the objective function equation constructed by the KPCA algorithm, and max represents the maximum value; Then, the data sample set Y = (y1, y2, ..., y M ) construct a similarity matrix W, obtain a diagonal matrix D according to the similarity matrix W, and obtain a Laplace matrix L according to the diagonal matrix D; The similarity matrix W is: Among them, W m,l represents the mth data sample y in the data sample set obtained in step (4-1) m and the lth data sample y l The similarity between is the mth data sample y m The θ neighborhood set of , that is, the set of θ data samples closest to the mth data sample in the data sample set; γ is a positive real number used to adjust the distribution of similarity and ensure that the elements of the similarity matrix are within a reasonable range. m∈[1,M], l∈[1,M], θ∈[1,M], and e represents a natural constant with a value of approximately 2.71828. A diagonal matrix D is a matrix whose elements outside the main diagonal are all 0, and whose element D in the mth row and mth column is mm for: The Laplace matrix L is: L=DW Afterwards, the objective function equation of the KLPP algorithm is established based on the obtained similarity matrix W, diagonal matrix D, Laplace matrix L, projection data sample set and projection matrix: J local (E)=minE T φ(Y)(DW)φ T (Y)E =minE T φ(Y)Lφ T (Y)E Among them J local It is the objective function equation constructed by the KLPP algorithm, and min means taking the minimum value; Finally, according to the objective function equations of the KPCA algorithm and the KLPP algorithm, the overall objective function equation is obtained:

8. The industrial park load analysis method based on optimized K-means clustering according to claim 7 is characterized in that: Step (4-3) is specifically: First, according to the data sample set φ(Y)={φ(y1),φ(y2),…,φ(y M )}Build the kernel function: K ij =φ(y i )φ(y j ) Where φ(y i ) and φ(y i ) are respectively mapped to the i-th and j-th data samples in the high-dimensional space data sample set, K ij Represents the kernel function value between the i-th data sample and the j-th data sample in the image of M data samples mapped to the high-dimensional space, and i∈[1,M], j∈[1,M]. Then, the corresponding kernel matrix is ​​obtained according to the obtained kernel function: K=[K ij ] M×M Then, the obtained kernel matrix is ​​normalized to obtain a normalized kernel matrix: in, is the kernel matrix after normalization, I M is an M×M matrix whose elements are Finally, the overall objective function equation obtained in step (4-2) is simplified according to the obtained standardized kernel matrix to obtain the simplified overall objective function equation:

9. The industrial park load analysis method based on optimized K-means clustering according to claim 8, characterized in that: In step (4-4), the Lagrangian function expression is: Among them, λ is the Lagrange multiplier, LA(E,λ) is the Lagrange function; In step (4-4), the contribution rate of each eigenvalue in the eigenvector matrix is ​​calculated as: where ρ τ is the contribution rate of the eigenvalue of the τth normalized kernel matrix, a τ is the eigenvalue of the τth normalized kernel matrix, and τ∈[1,M]; In step (4-4), the updated projection data sample set expression is: Where X' is the updated projection data sample set, whose dimension is smaller than the dimension of the data sample set Y. The updated projection data sample set X' is the initial cluster feature of the power load data set.

10. An industrial park load analysis system based on optimized K-means clustering, characterized in that: include: The first module is used to obtain a power load data set composed of multiple power load data of the industrial park, and perform standardization processing on the power load data set to obtain a processed power load data set. The second module is used to classify the power load data set processed by the first module according to the power consumption mode or power consumption group and use the elbow rule to obtain β optimal cluster categories and their corresponding cluster centers; The third module is used to perform K-means clustering analysis on the processed power load data set obtained by the first module according to the β optimal cluster categories and their corresponding cluster centers obtained by the second module, so as to obtain an initial cluster division result of the power load data set; The fourth module is used to process the initial clustering result of the power load data set obtained by the third module using the KPCA algorithm in the KPCA-KLPP fusion model to obtain the main features of the power load data set, and use the KLPP algorithm in the KPCA-KLPP fusion model to process the main features of the obtained power load data set to obtain the local structure of the power load data set, and then process the main features and local structure of the obtained power load data set to obtain the initial clustering features of the power load data set; The fifth module is used to process the initial cluster characteristics of the power load data set obtained in the fourth module using the DTW method to obtain a cumulative distance minimum matrix; The sixth module is used to use the cumulative distance minimum matrix obtained in the fifth module to perform secondary cluster category optimization processing on the cluster division results of the power load data set obtained in the third module to obtain the reallocated clusters, and use the same K-means clustering analysis method as the third module to perform secondary clustering processing on the reallocated clusters to obtain the final clustering results of the power load data set, which is the analysis result of the industrial park power load data set.

Images