A method for constructing industry prosperity index driven by power data

Abstract

The present invention relates to a method for constructing an industry prosperity index driven by power data, comprising the following steps: 1) taking enterprises belonging to the same industry as an enterprise set V, and obtaining a daily electricity consumption set S including the daily electricity consumption time series data of each enterprise in the enterprise set V; 2) performing data preprocessing on the daily electricity consumption set S, and classifying it by cluster analysis; 3) obtaining the influence weights of influencing factors on the daily electricity consumption of enterprises under each classification by correlation analysis; 4) obtaining the predicted value of electricity consumption for the next day by using a prediction model; 5) respectively constructing an industry prosperity index system H, an industry prosperity comprehensive index CI and an industry prosperity diffusion index DI. Compared with the prior art, the present invention has the advantages of high frequency, fine granularity, strong prediction and the like.

Description

Technical Field

[0001] The present invention relates to the field of electric power information big data, and in particular to a method for constructing an industry prosperity index driven by electric power data. Background Art

[0002] Problems such as overcapacity, single product structure, low technical level and low localization rate of supporting facilities are restricting the sustainable development of the local shipbuilding industry. The digital economy driven by big data is urgently needed to accelerate the transformation and upgrading of the industry.

[0003] In the field of economics, electricity consumption objectively reflects the prosperity of industrial manufacturing industries such as the shipbuilding industry and its changing trends. With the help of electricity big data such as corporate electricity consumption, business expansion data, operating capacity, contract capacity, etc., the prosperity fluctuations of the industry can be analyzed and predicted from the perspective of electricity, which will help industry participants such as the government, enterprises and banks to formulate and adjust relevant policy plans. At present, domestic authoritative indexes such as the China Export Container Freight Index, the China Coastal (Bulk) Freight Index, and the China Shipping Prosperity Index regard electricity consumption as an important part of the indicator system. However, the existing prosperity indexes, including the aforementioned, are difficult to give full play to the high frequency, multi-dimensional, and massive advantages of electricity data, which is mainly reflected in the following aspects:

[0004] First, the frequency is low. The frequency of existing prosperity indexes is mostly monthly, quarterly, and annual, while the frequency of electricity consumption is daily, hourly, and minute-level. In order to include electricity consumption in the index system, the existing prosperity index aggregates electricity consumption by time units such as month, quarter, and year, losing the daily variation characteristics of electricity consumption data.

[0005] Second, the granularity is coarse. The granularity of existing prosperity indexes is mostly at the industry level, while the granularity of electricity consumption is at the enterprise level and building level. In order to include electricity consumption in the index system, the existing prosperity index aggregates electricity consumption by industry, etc., losing the enterprise-level variation characteristics of electricity consumption data.

[0006] Third, the forecast is weak. In the existing prosperity index indicator system, electricity consumption usually plays the role of a leading indicator. Since the prosperity index is compiled based on the real data of each indicator, its ability to predict future changes in the prosperity is limited. Summary of the invention

[0007] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a high-frequency, fine-grained, and highly predictive method for constructing an industry prosperity index driven by power data.

[0008] The purpose of the present invention can be achieved by the following technical solutions:

[0009] A method for constructing an industry prosperity index driven by power data includes the following steps:

[0010] 1) Enterprises belonging to the same industry are taken as an enterprise set V, and a daily electricity consumption set S including the daily electricity consumption time series data of each enterprise in the enterprise set V is obtained;

[0011] 2) Perform data preprocessing on the daily electricity consumption set S and classify it through cluster analysis;

[0012] 3) Use the correlation analysis method to obtain the influence weights of the influencing factors on the daily electricity consumption of enterprises under each category;

[0013] 4) Use the prediction model to obtain the predicted value of electricity consumption for the next day;

[0014] 5) Construct the industry prosperity index system H, industry prosperity comprehensive index CI and industry prosperity diffusion index DI respectively.

[0015] Furthermore, the data preprocessing specifically includes:

[0016] 21) Using the quartile method, remove the outliers of each daily power consumption time series data in the daily power consumption set S;

[0017] 22) Using the moving average method, fill in the missing points of each daily power consumption time series data in the daily power consumption set S;

[0018] 23) Using the K-Shape curve clustering analysis method and the daily electricity consumption curve, the daily electricity consumption set S is classified;

[0019] 24) Correct the abnormal points in the time series data of daily electricity consumption of enterprises in each category.

[0020] Furthermore, the step 23) specifically includes:

[0021] 231) Construct a power consumption matrix U corresponding to each category, where the i-th row vector in the power consumption matrix U is the i-th power consumption time series data in the corresponding category;

[0022] 232) Get the number of clusters k and initialize the cluster center C of each category k , where the cluster center C k is the zero vector;

[0023] 233) Calculate the distance from each vector in the power consumption matrix U to each cluster center C k The shape similarity distance is used to classify each vector into the category with the smallest shape similarity distance;

[0024] 234) Update the cluster center C of each category according to the clustering results k , so that the cluster center C k The sum of shape similarity distances with all vectors in this category is the shortest;

[0025] 235) Repeat steps 233) and 234) until the preset maximum number of iterations is reached or the clustering result no longer changes.

[0026] Furthermore, the step 24) only corrects the abnormal points in the missing points of the daily power consumption time series data filled in step 22), which specifically includes:

[0027] 241) Obtain the power consumption matrix U corresponding to any category G in the daily power consumption set S;

[0028] 242) traverse each row of the power consumption matrix U, and search for the row with the least missing points and outliers as the standard row;

[0029] 243) Traverse the power consumption matrix U from left to right and calculate the cosine similarity between the standard row and other rows;

[0030] 244) If the cosine similarity between a row and the standard row is lower than the set threshold, some point values ​​in the row are modified so that the cosine similarity between the row and the standard row is higher than the set threshold, thus completing the correction of the abnormal points.

[0031] Furthermore, the daily electricity consumption curve is a curve of the daily electricity consumption time series data of each enterprise in the two-dimensional coordinate system of "date-electricity consumption".

[0032] Furthermore, the influencing factors include meteorological information and holiday information. The meteorological information includes temperature, humidity, wind speed and rainfall. The holiday information includes time series data of statutory holidays and weekends. The influence weight of the influencing factors on the daily electricity consumption of the enterprise is calculated by the Pearson correlation coefficient.

[0033] Furthermore, the prediction model adopts a Seq2Seq prediction model, whose input includes daily electricity consumption time series data, holiday time series data and meteorological information.

[0034] Furthermore, each enterprise in the enterprise set V is an indicator in the industry prosperity index system H, and the current value of each indicator in the industry prosperity index system H is the ratio of the next day's electricity consumption forecast value of the corresponding enterprise to the actual electricity consumption on the day, and the weight of each indicator is the operating capacity or contract capacity of the corresponding enterprise.

[0035] Furthermore, the current value of the comprehensive industry prosperity index CI is the normalized value of the sum of the products of the current values ​​of all indicators in the industry prosperity index system H and the corresponding weights, and its calculation formula is:

[0036]

[0037] Among them, p i is the weight of the i-th indicator in the industry prosperity index system H, qi It is the current value of the i-th indicator in the industry prosperity index system H.

[0038] Furthermore, the current value of the industry prosperity diffusion index DI is the normalized value of the sum of the comparison results of the current value and the previous value of all indicators in the industry prosperity index system H, and its calculation formula is:

[0039]

[0040]

[0041] Among them, x i is the previous value of the i-th indicator in the industry prosperity indicator system H, q i is the current value of the i-th indicator in the industry prosperity index system H, I(x i ,qi) is a piecewise continuous binary function, which maps the comparison result of the previous value and the current value of the i-th indicator to the interval [0,1].

[0042] Compared with the prior art, the present invention has the following advantages:

[0043] 1) High frequency: The daily comprehensive index CI and daily diffusion index DI constructed by the present invention utilize the collection frequency of electricity consumption to update and predict the daily frequency, which can obtain the daily change characteristics of the industry prosperity index, and help industry entities such as governments, enterprises, and financial institutions to carry out daily decision-making and analysis work;

[0044] 2) Fine-grained: The industry prosperity index system H constructed by the present invention is an enterprise-level index system, in which each index corresponds to each enterprise one by one, depicting the industry prosperity status from a micro level, which is helpful to analyze the prosperity covariation between industry subcategories by combining the upstream and downstream relationships in the industry;

[0045] 3) Strong prediction: The next-day electricity consumption prediction model constructed by the present invention comprehensively uses cluster analysis technology, data cleaning technology, time series prediction technology, etc. to effectively pre-process the input data, with high prediction accuracy, so that the daily comprehensive index CI and daily diffusion index DI constructed on its basis can more truly reflect the short-term future changes in the industry's prosperity. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0047] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should belong to the scope of protection of the present invention.

[0048] Example

[0049] like Figure 1 As shown, the present invention provides a method for constructing an industry prosperity index driven by power data, which first establishes three sets, namely, an enterprise set V, a daily power consumption set S, and a power consumption time series curve set C, wherein the enterprise set V = {V1, V2, ..., Vm} is a set of enterprises belonging to a certain industry, the daily power consumption set S = {S1, S2, ..., Sm} is the daily power consumption time series data of each enterprise in the enterprise set V, and the power consumption time series curve set C = {C1, C2, ..., Cm} is the corresponding curve of the daily power consumption time series data of each day in the daily power consumption set S in the "date-power consumption" two-dimensional coordinate system. The construction method of the present invention performs the following steps in sequence:

[0050] S1: The quartile method is used to remove outliers in each daily electricity consumption time series data in the daily electricity consumption set S.

[0051] Specifically: for any time series data S in the daily power consumption set S i , calculate its first quartile P, median Q and third quartile R. Let IQR = RP, and convert the time series data S i Points with values ​​outside [P-1.5*IQR, R+1.5*IQR] are set as missing points.

[0052] S2: Use the moving average method to fill in the missing points of each daily electricity consumption time series data in the daily electricity consumption set S.

[0053] Specifically: for any time series data S in the daily power consumption set S i , traverse the time series data S from left to right i The elements in the window are calculated, and the moving average of the window is w. If S i [j] is a missing value, then S i [j] Assigned to window S i The average number of elements in [jw,j-w+1,...,j-1].

[0054] S3: Use K-Shape curve clustering analysis method to classify the daily electricity consumption set S, so that the corresponding curves of the daily electricity consumption time series data belonging to the same classification in the electricity time series curve set C have similar morphological characteristics.

[0055] Specifically:

[0056] S301: constructing a power consumption matrix U corresponding to each category, wherein the i-th row vector in the power consumption matrix U is the i-th power consumption time series data in the corresponding category;

[0057] S302: Specify the number of clusters k and initialize the cluster center C of each category k , where the cluster center C k is the zero vector;

[0058] S303: Calculate the distance from each vector in the power consumption matrix U to each cluster center C k The shape similarity distance is used to classify each vector into the class with the smallest shape similarity distance. Among them, the two vectors S in the power consumption matrix U i and S j The calculation formula of the shape similarity distance between is:

[0059]

[0060] Among them, S i and S j are two vectors in the electricity consumption matrix U, ED, MD, and DSD are the Euclidean distance, Manhattan distance, and absolute value of the sum of the elements of the difference vector between the two vectors, respectively;

[0061] S304: Update the cluster center C of each category according to the clustering results k , so that the cluster center C k The sum of shape similarity distances with all vectors in this category is the shortest;

[0062] S305: Repeat steps S303-S304 until a preset maximum number of iterations is reached or the clustering result does not change any more.

[0063] S4: For each category in the daily electricity consumption set S, the cosine similarity method is used to correct the abnormal points of the daily electricity consumption time series data of each enterprise in the classification, wherein the modifiable points must be the missing points filled in step S2.

[0064] Specifically include:

[0065] S401: Obtaining the power consumption matrix U corresponding to any category G in the daily power consumption set S;

[0066] S402: traverse each row of the power consumption matrix U, and find the row with the least missing points and outliers before cleaning as the standard row.

[0067] S403: Traverse the power consumption matrix U from left to right, and calculate the cosine similarity between the standard row and other rows. If the cosine similarity between a row and the standard row is lower than the threshold, modify some point values ​​in the row so that the cosine similarity between the row and the standard row is higher than the preset threshold. The modifiable points must be the missing points filled in step S2.

[0068] S5: For each category of the daily electricity consumption set S, the correlation analysis method is used to calculate the influence weight of the influencing factors on the daily electricity consumption of each enterprise in this category.

[0069] The influencing factors include meteorological information and holiday information. The meteorological information includes temperature, humidity, wind speed and rainfall. The holiday information includes the time series data of statutory holidays and weekends. The influence weight of the influencing factors on the daily electricity consumption of the enterprise is calculated by the Pearson correlation coefficient, and the calculation formula is:

[0070]

[0071] Among them, X and is the daily electricity consumption time series data and its mean, Y and It is the time series data of a certain influencing factor and its mean.

[0072] S6: For each daily electricity consumption time series data of the daily electricity consumption set S, a Seq2Seq prediction model is used to calculate the predicted value of the next day's electricity consumption. The input of the prediction model includes the daily electricity consumption time series data, the time series data of meteorological factors, holidays and other factors, and their influence weights, specifically including: (1) electricity consumption time series data; (2) time series data of holidays such as the Spring Festival and weekends, with 1 and 0 representing whether a certain day is a holiday; (3) time series data of meteorological factors such as temperature, humidity, wind speed, and rainfall, and the influence weights of each factor on electricity in the prediction model are initialized with the correlation coefficient calculated in step S5.

[0073] S7: Construct an industry prosperity index system H. Each enterprise in the enterprise set V is an indicator in the industry prosperity index system H. The current value of each indicator is the ratio of the next day's predicted electricity consumption of the corresponding enterprise to the actual electricity consumption on the current day. The weight of each indicator is the operating capacity or contract capacity of the corresponding enterprise.

[0074] S8: Construct the industry prosperity comprehensive index CI. The current value of the industry prosperity comprehensive index CI is the normalized value of the sum of the current values ​​of all indicators in H and the product of the corresponding weights. The calculation formula is:

[0075]

[0076] Among them, p i is the weight of the i-th indicator in the industry prosperity index system H, q iIt is the current value of the i-th indicator in the industry prosperity index system H.

[0077] S9: Construct the industry prosperity diffusion index DI. The current value of the industry prosperity diffusion index DI is the normalized value of the sum of the comparison results of the current value of all indicators in H with the previous value. The calculation formula is:

[0078]

[0079]

[0080] Among them, x i is the previous value of the i-th indicator in the industry prosperity indicator system H, q i is the current value of the i-th indicator in the industry prosperity index system H, and the piecewise continuous binary function I(x i ,q i ) maps the comparison result of the previous value and the current value of the ith indicator to the interval [0,1].

[0081] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for constructing an industry prosperity index driven by power data, characterized in that: The following steps are involved: 1) Enterprises belonging to the same industry are taken as an enterprise set V, and a daily electricity consumption set S including the daily electricity consumption time series data of each enterprise in the enterprise set V is obtained; 2) Perform data preprocessing on the daily electricity consumption set S and classify it through cluster analysis; 3) Use the correlation analysis method to obtain the influence weights of the influencing factors on the daily electricity consumption of enterprises under each category; 4) Use the prediction model to obtain the predicted value of electricity consumption for the next day; 5) Construct the industry prosperity index system H, industry prosperity comprehensive index CI and industry prosperity diffusion index DI respectively; The data preprocessing specifically includes: 21) Using the quartile method, remove the outliers of each daily power consumption time series data in the daily power consumption set S; 22) Using the moving average method, fill in the missing points of each daily power consumption time series data in the daily power consumption set S; 23) Using the K-Shape curve clustering analysis method and the daily electricity consumption curve, the daily electricity consumption set S is classified; 24) Correct the abnormal points of the time series data of daily electricity consumption of enterprises in each category; The step 23) specifically includes: 231) Construct a power consumption matrix U corresponding to each category, where the i-th row vector in the power consumption matrix U is the i-th power consumption time series data in the corresponding category; 232) Get the number of clusters k and initialize the cluster center C of each category k , where the cluster center C k is the zero vector; 233) Calculate the distance from each vector in the power consumption matrix U to each cluster center C k The shape similarity distance is used to classify each vector into the category with the smallest shape similarity distance; 234) Update the cluster center C of each category according to the clustering results k , so that the cluster center C k The sum of shape similarity distances with all vectors in this category is the shortest; 235) repeating steps 233) and 234) until a preset maximum number of iterations is reached or the clustering result no longer changes; The step 24) only corrects the abnormal points in the missing points of the daily power consumption time series data filled in step 22), which specifically includes: 241) Obtain the power consumption matrix U corresponding to any category G in the daily power consumption set S; 242) traverse each row of the power consumption matrix U, and search for the row with the least missing points and outliers as the standard row; 243) Traverse the power consumption matrix U from left to right and calculate the cosine similarity between the standard row and other rows; 244) If the cosine similarity between a row and the standard row is lower than the set threshold, some point values ​​in the row are modified so that the cosine similarity between the row and the standard row is higher than the set threshold, thus completing the correction of the abnormal points.

2. The method for constructing an industry prosperity index driven by power data according to claim 1, characterized in that: The daily electricity consumption curve is a curve of the daily electricity consumption time series data of each enterprise in the two-dimensional coordinate system of "date-electricity consumption".

3. The method for constructing an industry prosperity index driven by power data according to claim 1, characterized in that: The influencing factors include meteorological information and holiday information. The meteorological information includes temperature, humidity, wind speed and rainfall. The holiday information includes time series data of statutory holidays and weekends. The influence weight of the influencing factors on the daily electricity consumption of the enterprise is calculated by the Pearson correlation coefficient.

4. The method for constructing an industry prosperity index driven by power data according to claim 3 is characterized in that: The prediction model adopts a Seq2Seq prediction model, whose input includes daily electricity consumption time series data, holiday time series data and meteorological information.

5. The method for constructing an industry prosperity index driven by power data according to claim 3 is characterized in that: Each enterprise in the enterprise set V is an indicator in the industry prosperity index system H. The current value of each indicator in the industry prosperity index system H is the ratio of the next day's electricity consumption forecast value of the corresponding enterprise to the actual electricity consumption on the day. The weight of each indicator is the operating capacity or contract capacity of the corresponding enterprise.

6. The method for constructing an industry prosperity index driven by power data according to claim 5, characterized in that: The current value of the comprehensive industry prosperity index CI is the normalized value of the sum of the current values ​​of all indicators in the industry prosperity index system H and the product of the corresponding weights, and its calculation formula is: Among them, p i is the weight of the i-th indicator in the industry prosperity index system H, q i It is the current value of the i-th indicator in the industry prosperity index system H.

7. The method for constructing an industry prosperity index driven by power data according to claim 5, characterized in that: The current value of the industry prosperity diffusion index DI is the normalized value of the sum of the comparison results of the current value and the previous value of all indicators in the industry prosperity index system H, and its calculation formula is: Among them, x i is the previous value of the i-th indicator in the industry prosperity indicator system H, q i is the current value of the ith indicator in the industry prosperity index system H, and I(xi,qi) is a piecewise continuous binary function, which maps the comparison result between the previous value and the current value of the ith indicator to the interval [0,1].

Images