Multivariable low-light remote sensing urban power energy consumption prediction method and device

Abstract

The invention provides a multivariable low-light remote sensing city power energy consumption prediction method and device, and belongs to the field of low-light remote sensing power industry application, and the method comprises the steps: data normalization processing: carrying out the processing of a low-light remote sensing image, and obtaining the night light data of a city, collecting urban power energy consumption statistical data, a second industry domestic total production value proportion and a third industry GDP proportion, and normalizing the four data item by item; city classification is carried out based on a hierarchical clustering method of time sequence consistency evaluation, cities are classified by adopting the hierarchical clustering method year by year based on the data, the maximum jump distance is calculated by adopting a connection matrix, and a final city category is obtained by calculating a consistency score; and carrying out city EPC prediction based on a multivariable regression method, establishing city EPC prediction models for different types of cities according to a city classification result of hierarchical clustering based on time sequence consistency evaluation, and obtaining a city EPC prediction result based on a low-light image.

Description

Technical Field

[0001] The present invention belongs to the application field of low-light remote sensing in the power industry, and further belongs to the urban electric power consumption (EPC) simulation and prediction technology, and specifically relates to a multivariable low-light remote sensing urban electric power consumption prediction method and device. Background Art

[0002] Traditional electricity consumption data is mostly based on statistical data from administrative units. This data suffers from time lags and missing data, and cannot fully reveal spatial variations in electricity consumption within administrative regions. Remote sensing technology offers new methods for predicting electricity consumption, and the use of low-light-level remote sensing imagery has proven to be a convenient and effective method. Currently, the primary remote sensing data sources used for EPC prediction come from DMSP / OLS and NPP / VIIRS nighttime light data. EPC is the sum of residential and industrial electricity consumption within a given time period in a given region and serves as a fundamental indicator for quantitatively evaluating regional electricity consumption. However, simply establishing a fitting prediction model between nighttime light data and electricity consumption leads to limitations in EPC simulation and prediction accuracy.

[0003] Taking into account the regional development level and pattern of the study area, scholars use methods to classify cities according to different standards to estimate EPC, including the Boston matrix classification method, K-Means classification algorithm and improved general matrix classification method.

[0004] (1) Boston Matrix Classification

[0005] Common classification methods in statistics and economics typically use "urban population growth rate" and "relative urban population ratio" as absolute and relative indicators, respectively, to construct the Boston Matrix, which categorizes cities into four categories: star cities, cash cow cities, problem cities, and dog cities. The Boston Matrix has a fixed number of categories and cannot flexibly adapt to sample differences and sample size.

[0006] (2) K-Means algorithm

[0007] K-Means is an iterative cluster analysis algorithm. Its core concept is to partition n objects in a dataset into k clusters, minimizing the sum of the distances from each object to the center of its cluster. The K-Means algorithm iteratively optimizes the clustering results, ensuring that objects within each cluster are as close together as possible and objects within different clusters are as far apart as possible. However, this algorithm has limitations: while it allows for flexible selection of the number of clusters, it uses fewer reference indicators when determining the number of cluster centers, making it less comprehensive.

[0008] (3) Hierarchical clustering method

[0009] Hierarchical clustering is a method that groups data by calculating the distances between different data points to construct a cluster tree. It does not require a predefined number of clusters. By generating a dendrogram, it clearly demonstrates the relationships between data points and within clusters. It is generally suitable for small-scale data and can handle non-spherical clusters. However, this method has limitations: it relies on subjective judgment, which may affect the objectivity of the results.

[0010] (4) Improved general matrix classification method

[0011] The initial goal was to use a weighted scoring method to evaluate the industry attractiveness and corporate strength of various products. Regarding city classification, economic and demographic factors can reflect a city's overall strength and attractiveness. The improved general matrix classification method replaces the original corporate strength with comprehensive city strength, primarily based on economic factors, and replaces industry attractiveness with city attractiveness, primarily based on demographic factors. This method uses economic and demographic factors as primary evaluation indicators and further uses secondary evaluation indicators such as per capita GDP and urban population ratio to achieve city classification. Based on the resulting city attractiveness and comprehensive strength scores, cities are divided into three types: Type A, Type B, and Type C. However, the improved general matrix classification method only rigidly divides samples into three categories, with a fixed number of categories, and lacks flexibility to adapt to sample differences and changes in sample size.

[0012] To improve the accuracy of EPC predictions, existing technologies also incorporate other factors (such as regional GDP, population density, and vegetation index) into the model for correction, thereby improving the accuracy of simulations and predictions. However, current EPC simulation and prediction research based on low-light imagery is mostly focused on the national, provincial, or single-city levels. City classification methods lack flexibility in the number of clusters, do not consider the temporal consistency of urban development processes, and the indicator system used to determine the number of clusters is incomplete, resulting in low accuracy in city EPC predictions. Summary of the Invention

[0013] To solve the above technical problems, the present invention proposes a multivariable low-light remote sensing urban power energy consumption prediction method and device. The specific technical solution is as follows:

[0014] A multivariate low-light remote sensing urban power energy consumption prediction method includes the following steps:

[0015] Step 1: Data normalization, including: pre-processing low-light remote sensing images to obtain nighttime light data for each city, as well as collecting statistical data on each city's electric energy consumption (EPC), the proportion of the secondary industry in GDP, and the proportion of the tertiary industry in GDP, and normalizing the above data;

[0016] Step 2: Classify cities using a hierarchical clustering method based on temporal consistency evaluation to obtain city classification results. This includes: clustering and classifying cities year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, secondary industry GDP share data, and tertiary industry GDP share data to determine the optimal city classification results.

[0017] Step 3: Predict the EPC of urban electric energy consumption based on the multivariate regression method, including: establishing a multivariate linear regression model for predicting the EPC of electric energy consumption for different categories of cities based on the city classification results of the hierarchical clustering of the time series consistency evaluation, and predicting the electric energy consumption of each city.

[0018] A multivariable low-light remote sensing urban power energy consumption prediction device includes the following modules:

[0019] The data normalization processing module is used to pre-process low-light remote sensing images to obtain nighttime light data for each city; collect statistical data on power energy consumption (EPC), the proportion of secondary industry GDP, and the proportion of tertiary industry GDP for each city, and normalize the above data;

[0020] The city classification module is used to classify cities using a hierarchical clustering method based on temporal consistency evaluation. The results include: clustering and classifying cities year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, secondary industry GDP share data, and tertiary industry GDP share data. The optimal city classification result is determined based on temporal consistency evaluation.

[0021] The EPC prediction model construction and prediction module is used to: perform EPC prediction of urban electric energy consumption based on the multivariate regression method, including: establishing a multivariate linear regression model for EPC prediction for different categories of cities based on the city classification results based on time series consistency evaluation, and predicting the electric energy consumption of each city.

[0022] An electronic device comprises: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0023] A computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to implement the above method.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] (1) The present invention proposes a city classification algorithm based on hierarchical clustering, which performs refined city classification by integrating multi-dimensional spatiotemporal data. The method integrates the NTL data, electric power consumption statistics (EPC data), the proportion of GDP of the secondary industry, and the proportion of GDP of the tertiary industry of cities in different years, constructs a multi-dimensional feature space for hierarchical clustering, automatically determines the number of city clusters based on the maximum jump distance, and uses the adjusted Rand index to measure the consistency of the city clustering results in different years. Finally, the optimal clustering result is determined by constructing a consistency matrix for calculation and analysis. By classifying cities, the present invention avoids the differences in EPC rules caused by different urban industrial structures and economic development levels, solves the problems of fixed number of classifications and lack of comprehensiveness in existing city classification methods, and directly obtains clustering results by designing automated judgment criteria, thereby improving the computational efficiency of obtaining clustering results.

[0026] (2) This patent further considers the temporal characteristics of urban development, with the input data being city-related indicators accumulated year by year. At the same time, a temporal evaluation matrix is constructed, effectively improving the consistency of the classification results over time. This method avoids the noise interference caused by unified modeling of global data, thereby obtaining city classification results that are closer to reality.

[0027] (3) The present invention proposes an urban EPC prediction method based on the results of hierarchical clustering city classification based on time series consistency evaluation. This method integrates NTL, regional GDP and population density data to construct a multivariate EPC regression equation, which fully considers the impact of regional GDP and population density on EPC. Compared with the linear model, exponential model and power model that only consider the single relationship between EPC and NTL, a more realistic urban electricity consumption prediction system is established, which improves the overall accuracy of the urban EPC model. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a multivariable low-light remote sensing urban power energy consumption prediction method of the present invention;

[0029] Figure 2 Flowchart of the city classification algorithm based on hierarchical clustering with temporal consistency evaluation. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions, and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate 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 may be combined with each other as long as they do not conflict with each other. To achieve the above-mentioned objectives, the present invention adopts the following technical solutions.

[0031] The present invention proposes a multivariable low-light remote sensing urban power energy consumption prediction method, the flow chart is as follows Figure 1 As shown. Includes:

[0032] Step 1: Data preprocessing and normalization, including: preprocessing low-light remote sensing images to obtain nighttime light data for each city, collecting statistical data on power consumption (EPC), the proportion of secondary and tertiary industries in GDP, and normalizing these data.

[0033] Step 2: City classification, including: hierarchical clustering of city indicators year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, secondary industry GDP share data, and tertiary industry GDP share data. The optimal city classification result is determined based on the evaluation matrix generated by the temporal consistency evaluation.

[0034] Step 3: EPC prediction model construction and prediction: Based on the classification results, a multivariate regression model for EPC prediction is established for cities of different categories to predict the electric energy consumption of each city.

[0035] Step 1 includes:

[0036] Step 1.1: Night light image processing and total night light extraction;

[0037] After the required denoising and desaturation preprocessing of the low-light remote sensing image, the total nighttime light data (NTL) within the administrative area of each prefecture-level city in the study area was extracted. The calculation formula is shown in formula (1):

[0038] (1)

[0039] Where, For the The total amount of nighttime lights in a city; For the Within the city's administrative area The brightness value of each pixel; is the number of pixels within the city's administrative area.

[0040] Step 1.2, statistical data and related auxiliary data;

[0041] This invention requires statistical data such as power consumption (EPC statistics), the percentage of GDP in the secondary industry (%), the percentage of GDP in the tertiary industry (%), regional GDP (in 100 million yuan), and population density (people per square kilometer). These data can be obtained from statistical yearbooks. For cities where population density data is not directly published in the city statistical yearbooks, the ratio of the year-end permanent population to land area can be calculated as the city's population density for that year. Administrative division vector data can be obtained from the 1:4,000,000 county-level administrative region vector data published by the National Basic Geographic Information Center.

[0042] Normalize the EPC statistical data, secondary industry GDP share, tertiary industry GDP share, and NTL data of all cities one by one to obtain the normalized EPC statistical data, normalized secondary industry GDP share, normalized tertiary industry GDP share, and normalized NTL data. The four normalized indicators are used as input features for step 2.

[0043] The flowchart for step 2 is as follows Figure 2 As shown, the following steps are included:

[0044] Step 2.1: Use the hierarchical clustering method to obtain the dendrogram of the four normalized indicators for each modeling year for all cities studied, and calculate the maximum jump distance of the dendrogram through the connection matrix. As the maximum cutoff distance, the number of cluster categories for each year and the corresponding city clustering results are automatically divided. The calculation formulas are shown in (2) and (3).

[0045] (2)

[0046] Where, and Representative Second and The distance when the secondary dendrograms are merged, Represents the difference between two adjacent merge distances.

[0047] (3)

[0048] Where, Represents the index where the distance difference between adjacent merges is maximized. Operator that obtains the maximum value.

[0049] Step 2.2: After obtaining the clusters corresponding to cities in each year, we further calculated the adjusted Rand index of the clustering results of each year to construct a clustering result consistency evaluation matrix. By summing the results of each row of the consistency evaluation matrix, we can obtain the comprehensive score of the consistency between each year's clustering results and other years. , select the year with the maximum value of the sum of the rows of the consistency assessment matrix The corresponding clustering results are used as the final city clustering results. If there are multiple identical row maximum values, the clustering result closest to the predicted year is taken as the final city classification result. The specific calculation process is shown in formulas (4), (5), and (6).

[0050] (4)

[0051] Where, stands for Adjusted Rand Index Calculation, 、 Representative Hedi The clustering results of the year, Indicates that for all Years, all belong to the modeling year set , the value range of i, j is from 1 to T. Represents the consistency assessment matrix Row and The elements of the column are the adjusted Rand index values of the clustering results in year i and year j.

[0052] (5)

[0053] Where, Represents the Comprehensive score of consistency evaluation of annual clustering results: The adjusted Rand index value representing the clustering results of year i and year j, Represents the consistency assessment matrix All conditions met arrive and of Perform the summation.

[0054] (6)

[0055] Where, Representative consistency evaluation comprehensive score The year corresponding to the maximum value. Represents the consistency evaluation score When it reaches its maximum value, Corresponding year .

[0056] Step 3 includes, based on the results of step 2, integrating NTL, GDP data and population density data, and constructing a multivariate regression equation for each category of cities to establish a city EPC prediction model, as shown in Equation (7).

[0057] (7)

[0058] Where, 、 、 、 They represent the NTL regression coefficient, GDP regression coefficient, population density regression coefficient and constant term of the c-th type city respectively. 、 、 、 Representing category c The EPC value, NTL value, GDP data and population density value of each city, among which the GDP data is the total GDP of a city.

[0059] The mean absolute relative error (MARE) was used to evaluate the prediction effect of the model.

[0060] (8)

[0061] (9)

[0062] Where, is the EPC simulation value of city i; is the EPC statistical value of city i; is the relative error value of the i-th data; n is the total number of prefecture-level cities participating in the calculation.

[0063] The specific implementation plan takes the data of cities in the Yangtze River Delta urban agglomeration from 2017 to 2022 as an example, where the data from 2017 to 2021 are used to obtain the classification results of cities in the urban agglomeration and the EPC prediction model, and the EPC data in 2022 is used to verify the accuracy of the EPC prediction model. The implementation steps of this technical solution are further elaborated in detail.

[0064] Step 1, specifically:

[0065] Using NPP / VIIRS as a low-light remote sensing data source, we processed NPP / VIIRS images using an empirical threshold method. We selected 12 as the minimum threshold for NPP / VIIRS annual images and assigned a value of 0 to pixels with light brightness values less than 12 in NPP / VIIRS nighttime light images. We also performed cropping and reprojection on the remote sensing images. Based on the preprocessed NPP / VIIRS nighttime light remote sensing data, we extracted the NTL within the administrative areas of prefecture-level cities within the Yangtze River Delta urban agglomeration according to formula (1). We normalized the NTL, EPC, secondary industry GDP share, and tertiary industry GDP share for different cities in the Yangtze River Delta over the five-year period from 2017 to 2021.

[0066] Step 2, specifically:

[0067] Hierarchical clustering was performed on the normalized indicators of the Yangtze River Delta urban agglomeration cities from 2017 to 2021. Combining the proposed intelligent classification method for urban agglomeration cities based on hierarchical clustering, the urban agglomeration clustering results for each year were obtained according to formulas (2) and (3). The adjusted Rand index of the clustering results for each year was calculated, and a clustering result consistency evaluation matrix was constructed. The final city clustering results were obtained according to formulas (4), (5), and (6), as shown in Table 1.

[0068] Table 1

[0069] Step 3, specifically:

[0070] Based on the city classification results of the hierarchical clustering algorithm based on temporal consistency evaluation (Table 1), a multivariate regression equation for EPC prediction was constructed by integrating the city's NTL data, regional GDP data, and population density data to obtain the EPC prediction results for each city within the urban agglomeration. The MARE method was used to measure prediction accuracy, and the linear, exponential, and power models using NTL and EPC modeling were compared. As shown in Table 2 (2022 Prediction Accuracy Comparison Table for the Yangtze River Delta Urban Agglomeration), the proposed method achieves the lowest mean absolute relative error in both the first and second categories, and approaches the optimal accuracy in the third category. Furthermore, the proposed method outperforms all compared methods for all cities, achieving a 1.69% improvement over the second-best method, demonstrating its effectiveness.

[0071] Table 2

[0072] In another aspect, the present invention provides a multivariable low-light remote sensing urban power energy consumption prediction device, comprising the following modules:

[0073] The data normalization processing module is used to pre-process low-light remote sensing images to obtain nighttime light data for each city within the urban agglomeration; collect statistical data on each city's electric energy consumption (EPC), the proportion of the secondary industry in GDP, and the proportion of the tertiary industry in GDP, and normalize the above data;

[0074] The city classification module is used to classify cities using a hierarchical clustering method based on temporal consistency evaluation, and obtain the classification results of cities within the urban agglomeration. This includes: clustering and classifying cities within the urban agglomeration year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, data on the proportion of GDP in the secondary industry, and data on the proportion of GDP in the tertiary industry, to determine the optimal city classification results.

[0075] The EPC prediction model construction and prediction module is used to: perform EPC prediction of urban electric energy consumption based on the multivariate regression method, including: establishing a multivariate linear regression model for EPC prediction for different categories of cities according to the classification results, and predicting the electric energy consumption of each city in the urban agglomeration.

[0076] On the other hand, the present invention provides an electronic device comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0077] In another aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the above method.

Claims

1. A multivariate low-light remote sensing urban power energy consumption prediction method, characterized by: include: Step 1: Data normalization, including: pre-processing low-light remote sensing images to obtain nighttime light data for each city, as well as collecting statistical data on each city's electric energy consumption (EPC), the proportion of the secondary industry in GDP, and the proportion of the tertiary industry in GDP, and normalizing these four data items for each city one by one; Step 2: Classify cities using a hierarchical clustering method based on temporal consistency evaluation to obtain city classification results. This includes: performing hierarchical clustering classification on cities year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, secondary industry GDP share data, and tertiary industry GDP share data. The clustering results for different years are then adjusted to obtain a consistency evaluation matrix to determine the optimal city classification result. Step 3: Predicting the EPC of urban electric energy consumption based on the multivariate regression method, including: establishing a multivariate regression model for predicting the EPC of electric energy consumption for different categories of cities based on the city classification results of the hierarchical clustering of the time series consistency evaluation, and predicting the electric energy consumption of each city.

2. The multivariate low-light remote sensing urban power energy consumption prediction method according to claim 1 is characterized in that: The preprocessing of low-light-level remote sensing images includes denoising and desaturation.

3. The multivariate low-light remote sensing urban power energy consumption prediction method according to claim 1 is characterized in that: Obtaining the nighttime light data of each city includes: extracting the total nighttime light data NTL, and the calculation formula is shown in formula (1): （1） Where, For the The total amount of nighttime lights in a city; For the Within the city's administrative area The brightness value of each pixel; is the number of pixels within the city's administrative area.

4. The multivariate low-light remote sensing urban power energy consumption prediction method according to claim 1 is characterized in that: Step 2 includes: Step 2.1: For all cities in each modeling year, the normalized night light data, power energy consumption EPC statistics, secondary industry GDP share data, and tertiary industry GDP share data are clustered using the hierarchical clustering method to obtain a dendrogram. The maximum jump distance of the dendrogram is calculated using the connection matrix. As the maximum cutoff distance, the number of cluster categories for each year and the corresponding city clustering results are automatically divided. The calculation formula is shown in (2) (3): （2） Where, and Representative Second and The distance when the secondary dendrograms are merged, Represents the difference between two adjacent merge distances; （3） Where, Represents the index where the distance difference between adjacent merges is maximized; argmax is the Operator to obtain the maximum value; Step 2.2: After obtaining the clusters corresponding to each year, the clustering result consistency evaluation matrix is constructed by calculating the adjusted Rand index of the clustering results of each year. By summing the results of each row of the temporal consistency evaluation matrix, we can obtain the comprehensive score of the consistency between each year's clustering results and other years. , select the year with the maximum value of the sum of the rows of the consistency assessment matrix The corresponding clustering results are used as the final city clustering results. If there are multiple identical row maximum values, the clustering result closest to the predicted year is taken as the final city classification result. The specific calculation process is shown in formulas (4), (5), and (6): （4） Where, stands for Adjusted Rand Index Calculation, 、 Representative Hedi The clustering results of the year, Indicates that for all Years, all belong to the modeling year set , the value range of i, j is from 1 to T, Represents the consistency assessment matrix Row and The elements of the column are the adjusted Rand index values of the clustering results in year i and year j; （5） Where, represents the comprehensive score of the consistency evaluation of the clustering results in year i; The adjusted Rand index value representing the clustering results of year i and year j, Represents the consistency assessment matrix All conditions met arrive and of Perform the summation: （6） Where, Representative consistency evaluation comprehensive score is the year corresponding to the maximum value, Represents the consistency evaluation score When it reaches its maximum value, Corresponding year .

5. The multivariable low-light remote sensing urban power energy consumption prediction method according to claim 4 is characterized in that: Step 3 includes: Based on the results of step 2, integrating NTL data, GDP data, and population density data, constructing a multivariate regression equation for each category of cities to establish a city EPC prediction model, as shown in formula (7): （7） Where, 、 、 、 They represent the NTL regression coefficient, GDP regression coefficient, population density regression coefficient and constant term of the c-th type city respectively; 、 、 、 Representing category c EPC value, NTL value, GDP data and population density value of each city.

6. The multivariate low-light remote sensing urban power energy consumption prediction method according to claim 1 is characterized in that: The average absolute relative error is used to evaluate the prediction effect: （8） （9） Where, is the EPC simulation value of city i; For the city EPC statistical value; For the The relative error value of each data; is the total number of prefecture-level cities involved in the calculation.

7. A multivariable low-light remote sensing urban power energy consumption prediction device, characterized in that: Includes the following modules: The data normalization processing module is used to pre-process low-light remote sensing images to obtain nighttime light data for each city; collect statistical data on power energy consumption (EPC), the proportion of secondary industry GDP, and the proportion of tertiary industry GDP for each city, and normalize the above data; The intelligent city classification module is used to classify cities using a hierarchical clustering method based on temporal consistency evaluation. The results include: hierarchical clustering of cities year by year based on normalized nighttime light data, electric power consumption (EPC) statistics, secondary industry GDP share data, and tertiary industry GDP share data, and determining the optimal city classification result based on consistency evaluation. The EPC prediction model construction and prediction module is used to: perform EPC prediction of urban electric energy consumption based on the multivariate regression method, including: establishing a multivariate regression model for EPC prediction for different categories of cities based on the city classification results based on time series consistency evaluation, and predicting the electric energy consumption of each city.

8. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the multivariate low-light remote sensing urban power energy consumption prediction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that Executable instructions are stored thereon, and when the instructions are executed by a processor, the processor implements the multivariable low-light remote sensing urban electric energy consumption prediction method according to any one of claims 1 to 6.

Images