Power customer order multi-dimensional portrait construction method and system based on big data

Abstract

The invention discloses a power customer order multi-dimensional portrait construction method and system based on big data, and relates to the technical field of power data processing prediction.The method comprises the steps that a power utilization mode of a customer is constructed through power data feature values, and a power utilization volatility level is evaluated according to the power utilization mode of the customer; therefore, the electricity utilization mode of the customer and the fluctuation condition of the mode are accurately captured, and a basis is provided for subsequent region similarity. Similar power customers in the similar areas are determined by comparing the similarity between the power utilization modes and the difference of the power utilization volatility levels, so that similar and similar customers are screened out. The division proportion of the data set is formulated through the influence relation and the regional similarity coefficient, the customer order portrait model is trained through the training set and the test set, and multiple customer dimensions, regional dimensions and external data dimensions are considered in the process of dividing the data set, so that the adaptability and precision of customer power order portrait construction are improved, and the customer power order portrait construction efficiency is improved. And the matching performance of customized electric power services is ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power data processing and prediction, and in particular to a method and system for constructing a multi-dimensional portrait of power customer orders based on big data. Background Art

[0002] The solution for constructing a multi-dimensional portrait of power customer orders based on big data mainly relies on the rapid development of big data processing and analysis technology. With the acceleration of the digitalization process of the power industry, massive power customer data has been accumulated, which provides the possibility for accurately portraying the portrait of power customer orders. The solution aims to analyze power customer orders by integrating data from multiple power-related information systems such as power supply services, marketing systems, and user collection, and using big data technologies such as cluster analysis and comprehensive evaluation. By mining the correlation and regularity in the data, an accurate portrait of power customer order behavior, demand characteristics, etc. can be achieved, providing data support and scientific basis for corporate decision-making, optimizing power resource allocation, and improving customer service levels and operational efficiency. The application of this technology not only promotes the development of the power industry towards intelligence and refinement, but also provides power customers with a more personalized and efficient service experience.

[0003] In the existing technology, customer profiles are often only constructed based on their electricity consumption, without considering data in other dimensions. This results in low accuracy and poor adaptability in constructing customer electricity order profiles, which cannot meet the increasingly complex and changeable customer electricity order needs.

[0004] Therefore, how to improve the accuracy and adaptability of building customer electricity order profiles is a technical problem that needs to be solved. Summary of the invention

[0005] The purpose of the present invention is to solve the problems of low accuracy and poor adaptability in the construction of customer power order portraits in the prior art, and propose a method for constructing a multi-dimensional portrait of power customer orders based on big data, including:

[0006] Obtaining the internal power data of the power customer, and calculating the power data characteristic value based on the power customer's internal power data, constructing the customer's power consumption pattern through the power data characteristic value, and evaluating the power consumption volatility level according to the customer's power consumption pattern;

[0007] Obtain the electricity consumption location information of all electricity customers, determine similar electricity customers in similar areas by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set;

[0008] Obtain the external power data of the power customer, analyze the influence relationship between the external power data and the internal power data, formulate the division ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of the power customer, and divide the feature sets into training sets and test sets according to the division ratio of the data set;

[0009] The customer order profile model is trained through training sets and test sets, and customized electricity services are provided to customers based on the customer order profile model.

[0010] In some embodiments of the present application, the power data characteristic value is calculated based on the power internal data of the power customer, and the power consumption pattern of the customer is constructed through the power data characteristic value, including:

[0011] Calculate the characteristic value of each type of power data, which includes the maximum value, the minimum value, the mode value and the average value, construct a curve of the change of each type of power data over time, randomly select a number of marking points on the power data change curve, and mark the points corresponding to the characteristic values ​​on the power data change curve, and record the two types of marking points as the first marking points and the second marking points respectively;

[0012] Determine the number of splits according to the difference between the mode value and the maximum value and the minimum value, split the power data change curve into a corresponding number of sub-curves according to the number of splits, recalculate the average value and standard deviation of each sub-curve, compare the average value and standard deviation of each sub-curve to obtain the change stability of the sub-curve, and predict the second mark point at the future time node according to the change stability of the sub-curve and the second mark point;

[0013] The size of random noise is determined by the standard deviation of the sub-curve, and the first marked point in the sub-curve is combined to predict the first marked point at the future time node. The interpolation method is used to smoothly and continuously connect the first marked point and the second marked point at the future time node to obtain the future prediction curve of each customer's power data, thereby constructing the customer's power consumption pattern.

[0014] In some embodiments of the present application, the power consumption volatility level is evaluated based on the power consumption pattern of the customer, including:

[0015] Determine the standard deviation, mean value, peak-to-valley difference and fluctuation frequency in the future prediction curve of each type of customer's electricity data, and determine the electricity consumption volatility level based on the standard deviation, mean value, peak-to-valley difference and fluctuation frequency;

[0016]

[0017] Among them, L is the level of electricity consumption volatility, α1 is the conversion coefficient of the coefficient of variation, Q1 is the standard deviation, Q2 is the average value, α2 is the conversion coefficient of the peak-to-valley difference, Q3 is the peak-to-valley difference, n is the number of time periods of the future forecast curve, and f ti is the fluctuation frequency in the ti-th period, a1 and a2 are the first constant and the second constant respectively, and [] is the rounding symbol.

[0018] In some embodiments of the present application, similar power customers in a nearby area are determined by comparing the similarity between power consumption patterns and the difference in power consumption volatility levels, and a similar set of regional power customers is obtained, including:

[0019] Adjacent customers are divided into adjacent customer sets according to their electricity usage location information. The similarities of future prediction curves of electricity data between customers are compared in the adjacent customer sets. The differences in electricity usage volatility levels between customers are compared in the adjacent customer sets. Similarity indicators are defined based on the similarities of future prediction curves and the differences in electricity usage volatility levels. Similar electricity customers in similar areas are determined based on the similarity indicators, and regional electricity customer similarity sets are constructed based on similar electricity customers in similar areas.

[0020] In some embodiments of the present application, the regional similarity coefficient of the regional power customer similarity set is determined by relying on the regional power customer similarity set, including:

[0021] According to the union of the electricity usage location information of each customer in the regional electricity customer similarity set, the regional electricity customer similarity area is planned, and the center point of the regional electricity customer similarity area is determined, and the distance between the electricity usage location of each customer in each regional electricity customer similarity set and the center point is calculated, and the regional similarity coefficient of the regional electricity customer similarity set is determined in combination with the similarity index;

[0022]

[0023] in, is the regional similarity coefficient of the regional power customer similarity set, m1 is the number of customers in the regional power customer similarity set, X j is the distance between the power consumption location of the jth customer in the regional power customer similarity set and the center point, m2 is the number of similar indicators in the regional power customer similarity set, β b is the weight of the bth similarity index, C b is the bth similarity index, is the initial regional similarity coefficient, It represents the initial regional similarity coefficient obtained by mapping the average influence of similar indicators, and k is a preset constant.

[0024] In some embodiments of the present application, the influence relationship between the external power data and the internal power data is analyzed, including:

[0025] Calculate the correlation between external power data and internal power data, filter external power data based on the correlation, and analyze the impact relationship between external power data and internal power data through a regression model.

[0026] In some embodiments of the present application, the division ratio of the data set is formulated by influencing the relationship and the regional similarity coefficient, including:

[0027] The number of types of external power data in the influence relationship between external power data and internal power data is counted, and the information entropy of the influence relationship between external power data and internal power data is calculated. The model complexity is evaluated according to the size and number of regional similarity coefficients and the number of types of external power data and the information entropy of the influence relationship. The division ratio of the data set is determined by the model complexity.

[0028] In some embodiments of the present application, the model complexity is evaluated according to the size and number of regional similarity coefficients and the number of types and information entropy of the influence relationship of external power data, including:

[0029]

[0030] Among them, S is the model complexity, ω is the complexity coefficient of the regional similarity coefficient, d is the number of regional similarity coefficients, is the size of the v-th regional similarity coefficient, is the complexity coefficient determined by the number of types of power external data, and g is the information entropy that affects the relationship.

[0031] Correspondingly, the present application also provides a system for constructing a multi-dimensional portrait of power customer orders based on big data, including:

[0032] A construction module is used to obtain the internal power data of the power customer, calculate the power data characteristic value according to the internal power data of the power customer, construct the power consumption pattern of the customer through the power data characteristic value, and evaluate the power consumption volatility level according to the power consumption pattern of the customer;

[0033] A determination module is used to obtain the electricity consumption location information of all electricity customers, determine similar electricity customers in a similar area by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set;

[0034] A partitioning module is used to obtain the external power data of power customers, analyze the influence relationship between the external power data and the internal power data, formulate the partition ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of power customers, and divide the feature sets into training sets and test sets according to the partition ratio of the data set;

[0035] The portrait module is used to train the customer order portrait model through training sets and test sets, and provide customized electricity services to customers based on the customer order portrait model.

[0036] By applying the above technical solutions, the customer's electricity consumption pattern is constructed through the characteristic values ​​of electricity data, and the electricity consumption volatility level is evaluated based on the customer's electricity consumption pattern, so as to accurately capture the customer's electricity consumption pattern and the volatility of the pattern, and provide a basis for subsequent regional similarity. Similar electricity customers in similar areas are determined by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, so as to screen out similar and similar customers. The division ratio of the data set is determined by the influence relationship and regional similarity coefficient, and the customer order portrait model is trained through the training set and the test set. In the process of dividing the data set, multiple customer dimensions, regional dimensions and external data dimensions are considered, which improves the adaptability and accuracy of the customer electricity order portrait construction and ensures the matching of customized electricity services. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a process for constructing a multi-dimensional portrait of power customer orders based on big data proposed by the present invention;

[0038] Figure 2 This is a structural schematic diagram of a method for constructing a multi-dimensional portrait of electricity customer orders based on big data proposed in the present invention. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0040] Reference Figure 1 ,A method for constructing a multi-dimensional portrait of power customer orders based on big data includes the following steps:

[0041] Step S101, obtaining the internal power data of the power customer, and calculating the power data characteristic value according to the internal power data of the power customer, constructing the power consumption pattern of the customer through the power data characteristic value, and evaluating the power consumption volatility level according to the power consumption pattern of the customer.

[0042] In this embodiment, the internal power data includes basic power consumption data (electricity quantity data, voltage and current data, power factor data), power equipment data, etc. The power data characteristic values ​​(maximum value, minimum value and average value, etc.) are calculated, and the customer's power consumption pattern is mined, and the changes in the power consumption pattern (volatility level) are analyzed.

[0043] In some embodiments of the present application, the power data characteristic value is calculated based on the power internal data of the power customer, and the power consumption pattern of the customer is constructed through the power data characteristic value, including:

[0044] Calculate the characteristic value of each type of power data, which includes the maximum value, the minimum value, the mode value and the average value, construct a curve of the change of each type of power data over time, randomly select a number of marking points on the power data change curve, and mark the points corresponding to the characteristic values ​​on the power data change curve, and record the two types of marking points as the first marking points and the second marking points respectively;

[0045] Determine the number of splits according to the difference between the mode value and the maximum value and the minimum value, split the power data change curve into a corresponding number of sub-curves according to the number of splits, recalculate the average value and standard deviation of each sub-curve, compare the average value and standard deviation of each sub-curve to obtain the change stability of the sub-curve, and predict the second mark point at the future time node according to the change stability of the sub-curve and the second mark point;

[0046] The size of random noise is determined by the standard deviation of the sub-curve, and the first marked point in the sub-curve is combined to predict the first marked point at the future time node. The interpolation method is used to smoothly and continuously connect the first marked point and the second marked point at the future time node to obtain the future prediction curve of each customer's power data, thereby constructing the customer's power consumption pattern.

[0047] In this embodiment, the number of splits is determined according to the difference between the mode value and the maximum value and the minimum value (the difference between the mode value and the maximum value can determine the range of the power data value, so as to help split the curve), and the difference between the mode value and the two maximum values ​​is calculated respectively. A comprehensive difference is determined by combining the two differences. Different comprehensive differences correspond to different numbers of splits. The change stability of the sub-curve is obtained by comparing the average value and standard deviation of each sub-curve, and the change stability of the sub-curve is determined according to the change of the average value and the standard deviation on multiple sub-curves. The second marked point at the future time node is predicted according to the change stability of the sub-curve and the second marked point. Based on the change stability of the sub-curve and the historical data of the second marked point, a prediction model is established to complete the second marked point at the future time node. The first marked point represents the random fluctuation on the curve data. The random noise size is determined by the standard deviation of the sub-curve. Different standard deviations correspond to different random noises. On the basis of the prediction, a certain amount of random noise can be added to simulate this fluctuation. Finally, in order to ensure that the constructed curve is smooth and continuous, we can use interpolation or other smoothing techniques to connect the predicted feature points (second marked points) and random fluctuation points (first marked points).

[0048] In some embodiments of the present application, the power consumption volatility level is evaluated based on the power consumption pattern of the customer, including:

[0049] Determine the standard deviation, mean value, peak-to-valley difference and fluctuation frequency in the future prediction curve of each type of customer's electricity data, and determine the electricity consumption volatility level based on the standard deviation, mean value, peak-to-valley difference and fluctuation frequency;

[0050]

[0051] Among them, L is the level of electricity consumption volatility, α1 is the conversion coefficient of the coefficient of variation, Q1 is the standard deviation, Q2 is the average value, α2 is the conversion coefficient of the peak-to-valley difference, Q3 is the peak-to-valley difference, n is the number of time periods of the future forecast curve, and f ti is the fluctuation frequency in the ti-th period, a1 and a2 are the first constant and the second constant respectively, and [] is the rounding symbol.

[0052] In this embodiment, the peak-to-valley difference is the difference between the peak value and the valley value in the power consumption curve, reflecting the amplitude of the change in power consumption, and the fluctuation frequency is the number of times the power consumption changes in a short period of time (such as the change exceeds a threshold within one hour). It represents the correction of the fluctuation frequency to the comprehensive fluctuation of the coefficient of variation and the peak-to-valley difference.

[0053] Step S102, obtain the electricity usage location information of all electricity customers, determine similar electricity customers in similar areas by comparing the similarities between electricity usage patterns and the differences in electricity usage volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set.

[0054] In this embodiment, different customers in adjacent areas often have similar electricity usage situations, and similar situations are captured to obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set, which describes the similarity of electricity usage situations of different customers in a certain area. The larger the coefficient, the more similar it is.

[0055] In some embodiments of the present application, similar power customers in a nearby area are determined by comparing the similarity between power consumption patterns and the difference in power consumption volatility levels, and a similar set of regional power customers is obtained, including:

[0056] Adjacent customers are divided into adjacent customer sets according to their electricity usage location information. The similarities of future prediction curves of electricity data between customers are compared in the adjacent customer sets. The differences in electricity usage volatility levels between customers are compared in the adjacent customer sets. Similarity indicators are defined based on the similarities of future prediction curves and the differences in electricity usage volatility levels. Similar electricity customers in similar areas are determined based on the similarity indicators, and regional electricity customer similarity sets are constructed based on similar electricity customers in similar areas.

[0057] In this embodiment, the similarity of future prediction curves between different customers is calculated (Euclidean distance: this is the most direct distance measurement method, which calculates the square root of the sum of the squares of the differences between two curves at corresponding points. The smaller the Euclidean distance, the more similar the curves are) and the difference in power consumption volatility levels between different customers are calculated. The similarity index is jointly determined by combining the similarity and the difference in power consumption volatility levels, and the customers with higher similarity indexes and in adjacent areas are used to construct a regional power customer similarity set.

[0058] In some embodiments of the present application, the regional similarity coefficient of the regional power customer similarity set is determined by relying on the regional power customer similarity set, including:

[0059] According to the union of the electricity usage location information of each customer in the regional electricity customer similarity set, the regional electricity customer similarity area is planned, and the center point of the regional electricity customer similarity area is determined, and the distance between the electricity usage location of each customer in each regional electricity customer similarity set and the center point is calculated, and the regional similarity coefficient of the regional electricity customer similarity set is determined in combination with the similarity index;

[0060]

[0061] in, is the regional similarity coefficient of the regional power customer similarity set, m1 is the number of customers in the regional power customer similarity set, X j is the distance between the power consumption location of the jth customer in the regional power customer similarity set and the center point, m2 is the number of similar indicators in the regional power customer similarity set, β b is the weight of the bth similarity index, C b is the bth similarity index, is the initial regional similarity coefficient, It represents the initial regional similarity coefficient obtained by mapping the average influence of similar indicators, and k is a preset constant.

[0062] In this embodiment, the number of customers, that is, the number of distances between the electricity consumption locations and the central point, is combined with the average value of the distances and the similarity index to jointly determine the regional similarity coefficient.

[0063] Step S103, obtain the external power data of the power customer, analyze the influence relationship between the external power data and the internal power data, formulate the division ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of the power customer, and divide the feature sets into training sets and test sets according to the division ratio of the data set.

[0064] In this embodiment, the external power data includes the temperature, humidity, rainfall, economic level and other external data that may affect the power consumption in the area. The complexity of the model is determined by the influence relationship and the regional similarity coefficient, and the division ratio of the data set is determined by the model complexity. The division ratio of the data set here refers to the division ratio of the training set and the test set, such as 7:3. For example, if the regional similarity coefficient is high, it means that the customer's power consumption behavior is relatively similar, and the model may not need to be too complex to capture this commonality; conversely, if the regional similarity coefficient is low, it means that the customer's power consumption behavior is quite different, and the model may need a more complex structure to capture this difference. Similarly, the influence relationship between the external power data and the internal power data may also affect the complexity of the model. If external factors have a significant impact on power consumption, the model may need to consider these factors, thereby increasing the complexity of the model.

[0065] In some embodiments of the present application, the influence relationship between the external power data and the internal power data is analyzed, including:

[0066] Calculate the correlation between external power data and internal power data, filter external power data based on the correlation, and analyze the impact relationship between external power data and internal power data through a regression model.

[0067] In this embodiment, correlation analysis can help us quantify the degree of association between two variables, which is usually measured using a correlation coefficient. If the correlation coefficient is large and significant, it means that there is a strong association between the two variables.

[0068] However, correlation analysis can only reveal the degree of association between variables, but cannot determine the causal relationship. Therefore, we also need to infer the causal relationship. This can be achieved by building regression models and other methods. Regression models can help us determine the direction and magnitude of the impact of external power data on internal power data.

[0069] In some embodiments of the present application, the division ratio of the data set is formulated by influencing the relationship and the regional similarity coefficient, including:

[0070] The number of types of external power data in the influence relationship between external power data and internal power data is counted, and the information entropy of the influence relationship between external power data and internal power data is calculated. The model complexity is evaluated according to the size and number of regional similarity coefficients and the number of types of external power data and the information entropy of the influence relationship. The division ratio of the data set is determined by the model complexity.

[0071] In this embodiment, the number of types of external power data in the influence relationship between external power data and internal power data (the number of types after screening) and the information entropy of the influence relationship between external power data and internal power data are counted. Information entropy is a measure of information uncertainty. From the perspective of information theory, the influence relationship between external power data and internal power data can be regarded as a process of information transfer. Calculating information entropy can evaluate the degree of uncertainty of information in this information transfer process, thereby indirectly reflecting the complexity of the influence relationship.

[0072] In some embodiments of the present application, the model complexity is evaluated according to the size and number of regional similarity coefficients and the number of types and information entropy of the influence relationship of external power data, including:

[0073]

[0074] Among them, S is the model complexity, ω is the complexity coefficient of the regional similarity coefficient, d is the number of regional similarity coefficients, is the size of the v-th regional similarity coefficient, is the complexity coefficient determined by the number of types of power external data, and g is the information entropy that affects the relationship.

[0075] In this embodiment, the model complexity is evaluated by combining the regional similarity coefficient and the information entropy of the influence relationship.

[0076] It should be noted that when the model complexity is low, the ratio of training set to test set can be relatively balanced, because the model is not highly dependent on data, and the performance will not change significantly due to the amount of training data. When the model complexity is high, in order to prevent overfitting, we need more training data to "teach" the model so that it can better generalize to new data. Therefore, when the model complexity is high, the ratio of training set is usually relatively large. Different model complexities correspond to different division ratios.

[0077] Step S104, training the customer order portrait model through the training set and the test set, and providing customized electricity services to customers based on the customer order portrait model.

[0078] In this embodiment, the customized power service includes the following:

[0079] 1. Customized service

[0080] Personalized recommendations: Recommend suitable electricity packages, energy-saving products, etc. based on customer portraits.

[0081] Precision marketing: Develop differentiated marketing strategies for different types of customers to improve marketing effectiveness.

[0082] 2. Risk Management

[0083] Credit assessment: Use the customer's credit record and historical payment history to assess the customer's credit risk.

[0084] Anomaly detection: Monitor customers’ electricity usage behavior in real time, detect abnormal electricity usage, and take timely measures to prevent losses.

[0085] 3. Optimize operations

[0086] Resource allocation: Rationally allocate grid resources based on customer profiles and power consumption forecasts to ensure the stability and reliability of power supply.

[0087] Service optimization: Based on customer feedback and portrait analysis results, continuously optimize service processes to improve customer satisfaction and loyalty.

[0088] By applying the above technical solutions, the customer's electricity consumption pattern is constructed through the characteristic values ​​of electricity data, and the electricity consumption volatility level is evaluated based on the customer's electricity consumption pattern, so as to accurately capture the customer's electricity consumption pattern and the volatility of the pattern, and provide a basis for subsequent regional similarity. Similar electricity customers in similar areas are determined by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, so as to screen out similar and similar customers. The division ratio of the data set is determined by the influence relationship and regional similarity coefficient, and the customer order portrait model is trained through the training set and the test set. In the process of dividing the data set, multiple customer dimensions, regional dimensions and external data dimensions are considered, which improves the adaptability and accuracy of the customer electricity order portrait construction and ensures the matching of customized electricity services.

[0089] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.

[0090] Correspondingly, this application also provides a system for constructing a multi-dimensional portrait of power customer orders based on big data, such as Figure 2 As shown, including,

[0091] A construction module is used to obtain the internal power data of the power customer, calculate the power data characteristic value according to the internal power data of the power customer, construct the power consumption pattern of the customer through the power data characteristic value, and evaluate the power consumption volatility level according to the power consumption pattern of the customer;

[0092] A determination module is used to obtain the electricity consumption location information of all electricity customers, determine similar electricity customers in a similar area by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set;

[0093] A partitioning module is used to obtain the external power data of power customers, analyze the influence relationship between the external power data and the internal power data, formulate the partition ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of power customers, and divide the feature sets into training sets and test sets according to the partition ratio of the data set;

[0094] The portrait module is used to train the customer order portrait model through training sets and test sets, and provide customized electricity services to customers based on the customer order portrait model.

[0095] Those skilled in the art will appreciate that the accompanying drawings are merely schematic diagrams of a preferred implementation scenario, and the modules or processes in the accompanying drawings are not necessarily required for implementing the present invention.

[0096] Those skilled in the art will appreciate that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.

[0097] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a multi-dimensional portrait of power customer orders based on big data, characterized in that: include, Obtaining the internal power data of the power customer, and calculating the power data characteristic value based on the power customer's internal power data, constructing the customer's power consumption pattern through the power data characteristic value, and evaluating the power consumption volatility level according to the customer's power consumption pattern; Obtain the electricity consumption location information of all electricity customers, determine similar electricity customers in similar areas by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set; Obtain the external power data of the power customer, analyze the influence relationship between the external power data and the internal power data, formulate the division ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of the power customer, and divide the feature sets into training sets and test sets according to the division ratio of the data set; The customer order profile model is trained through training sets and test sets, and customized electricity services are provided to customers based on the customer order profile model.

2. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 1 is characterized in that: The power data characteristic value is calculated based on the power customer's internal power data, and the customer's power consumption pattern is constructed through the power data characteristic value, including: Calculate the characteristic value of each type of power data, which includes the maximum value, the minimum value, the mode value and the average value, construct a curve of the change of each type of power data over time, randomly select a number of marking points on the power data change curve, and mark the points corresponding to the characteristic values ​​on the power data change curve, and record the two types of marking points as the first marking points and the second marking points respectively; Determine the number of splits according to the difference between the mode value and the maximum value and the minimum value, split the power data change curve into a corresponding number of sub-curves according to the number of splits, recalculate the average value and standard deviation of each sub-curve, compare the average value and standard deviation of each sub-curve to obtain the change stability of the sub-curve, and predict the second mark point at the future time node according to the change stability of the sub-curve and the second mark point; The size of random noise is determined by the standard deviation of the sub-curve, and the first marked point in the sub-curve is combined to predict the first marked point at the future time node. The interpolation method is used to smoothly and continuously connect the first marked point and the second marked point at the future time node to obtain the future prediction curve of each customer's power data, thereby constructing the customer's power consumption pattern.

3. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 2 is characterized in that: And evaluate the level of power consumption volatility based on the customer's power consumption pattern, including: Determine the standard deviation, mean value, peak-to-valley difference and fluctuation frequency in the future prediction curve of each type of customer's electricity data, and determine the electricity consumption volatility level based on the standard deviation, mean value, peak-to-valley difference and fluctuation frequency; Among them, L is the level of electricity consumption volatility, α1 is the conversion coefficient of the coefficient of variation, Q1 is the standard deviation, Q2 is the average value, α2 is the conversion coefficient of the peak-to-valley difference, Q3 is the peak-to-valley difference, n is the number of time periods of the future forecast curve, and f ti is the fluctuation frequency in the ti-th period, a1 and a2 are the first constant and the second constant respectively, and [] is the rounding symbol.

4. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 3 is characterized in that: By comparing the similarity between electricity consumption patterns and the differences in electricity consumption volatility levels, similar electricity customers in similar areas are determined, and a similar set of regional electricity customers is obtained, including: Adjacent customers are divided into adjacent customer sets according to their electricity usage location information. The similarities of future prediction curves of electricity data between customers are compared in the adjacent customer sets. The differences in electricity usage volatility levels between customers are compared in the adjacent customer sets. Similarity indicators are defined based on the similarities of future prediction curves and the differences in electricity usage volatility levels. Similar electricity customers in similar areas are determined based on the similarity indicators, and regional electricity customer similarity sets are constructed based on similar electricity customers in similar areas.

5. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 4 is characterized in that: The regional similarity coefficient of the regional power customer similarity set is determined by relying on the regional power customer similarity set, including: According to the union of the electricity usage location information of each customer in the regional electricity customer similarity set, the regional electricity customer similarity area is planned, and the center point of the regional electricity customer similarity area is determined, and the distance between the electricity usage location of each customer in each regional electricity customer similarity set and the center point is calculated, and the regional similarity coefficient of the regional electricity customer similarity set is determined in combination with the similarity index; Among them, θ is the regional similarity coefficient of the regional power customer similarity set, m1 is the number of customers in the regional power customer similarity set, X j is the distance between the power consumption location of the jth customer in the regional power customer similarity set and the center point, m2 is the number of similar indicators in the regional power customer similarity set, β b is the weight of the bth similarity index, C b is the bth similarity index, θ0 is the initial regional similarity coefficient, It represents the initial regional similarity coefficient obtained by mapping the average influence of similar indicators, and k is a preset constant.

6. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 1 is characterized in that: Analyze the impact relationship between external power data and internal power data, include, Calculate the correlation between external power data and internal power data, filter external power data based on the correlation, and analyze the impact relationship between external power data and internal power data through a regression model.

7. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 6 is characterized in that: The partitioning ratio of the data set is determined by influencing the relationship and geographical similarity coefficient, including, The number of types of external power data in the influence relationship between external power data and internal power data is counted, and the information entropy of the influence relationship between external power data and internal power data is calculated. The model complexity is evaluated according to the size and number of regional similarity coefficients and the number of types of external power data and the information entropy of the influence relationship. The division ratio of the data set is determined by the model complexity.

8. The method for constructing a multi-dimensional portrait of power customer orders based on big data according to claim 7 is characterized in that: The model complexity is evaluated based on the size and number of regional similarity coefficients, the number of types of external power data, and the information entropy of the influencing relationship, including: Among them, S is the model complexity, ω is the complexity coefficient of the regional similarity coefficient, d is the number of regional similarity coefficients, θ v is the size of the v-th regional similarity coefficient, is the complexity coefficient determined by the number of types of power external data, and g is the information entropy that affects the relationship.

9. A system for constructing a multi-dimensional portrait of power customer orders based on big data, characterized in that: include, A construction module is used to obtain the internal power data of the power customer, calculate the power data characteristic value according to the internal power data of the power customer, construct the power consumption pattern of the customer through the power data characteristic value, and evaluate the power consumption volatility level according to the power consumption pattern of the customer; A determination module is used to obtain the electricity consumption location information of all electricity customers, determine similar electricity customers in a similar area by comparing the similarity between electricity consumption patterns and the difference in electricity consumption volatility levels, obtain a regional electricity customer similarity set, and determine the regional similarity coefficient of the regional electricity customer similarity set based on the regional electricity customer similarity set; A partitioning module is used to obtain the external power data of power customers, analyze the influence relationship between the external power data and the internal power data, formulate the partition ratio of the data set according to the influence relationship and the regional similarity coefficient, extract the feature sets from the internal power data and the external power data of power customers, and divide the feature sets into training sets and test sets according to the partition ratio of the data set; The portrait module is used to train the customer order portrait model through training sets and test sets, and provide customized electricity services to customers based on the customer order portrait model.

Images