Multi-modal customer behavior model construction method and system

Abstract

The invention relates to the technical field of power grid customer behavior analysis, and discloses a multi-modal customer behavior model construction method and system, and the method comprises the following steps: obtaining power utilization information data of power utilization customers, constructing a multi-modal customer behavior database, and constructing a power utilization information data analysis mode according to the analysis demands of customer behaviors; the power utilization information data analysis mode comprises hierarchical progressive analysis, the hierarchical progressive analysis comprises a first analysis layer and a second analysis layer, and the second analysis layer is a sub-analysis layer of the first analysis layer. According to the method, before the model is constructed, data in the multi-modal customer behavior database is analyzed, and the periodicity of a data sequence is judged, so that the data with periodicity is screened out and divided into period sections, and subsequent model construction is focused on the data with clear characteristics; the waste of computing resources on a large amount of irregular or non-satisfactory data is reduced, the pertinence of data processing is improved, and then the model training speed is increased.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid customer behavior analysis, and particularly relates to a method and system for constructing a multi-modal customer behavior model. Background Art

[0002] In the current era of the rapid development of smart grids, it is crucial to accurately analyze the behavior of power grid customers. With the gradual opening of the power market and the increasing diversification of user demands, power grid operators need to deeply understand the electricity consumption behavior patterns of customers in order to achieve efficient power resource allocation, reasonable electricity price formulation, and high-quality customer service. However, when constructing a customer behavior model, there is often a lack of effective analysis of multi-modal customer behavior data. In the data processing link, when researchers try to use traditional methods to construct a model for a large amount of electricity consumption data, they lack detailed analysis of the data and cannot fully utilize the potential laws in the data, making it difficult for the model to achieve the optimal fitting effect and reducing the prediction accuracy and generalization ability of the model. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for constructing a multi-modal customer behavior model to solve at least one of the above-mentioned prior art problems.

[0004] In a first aspect, the present invention provides a method for constructing a multi-modal customer behavior model, including the following steps: Obtain the electricity consumption information data of electricity customers, construct a multi-modal customer behavior database, and construct an analysis mode for electricity consumption information data according to the analysis requirements of customer behavior. The analysis mode for electricity consumption information data includes hierarchical progressive analysis, where the hierarchical progressive analysis includes a first analysis layer and a second analysis layer, and the second analysis layer is a sub-analysis layer of the first analysis layer.

[0005] As a further solution of the present invention: the specific process of constructing the analysis mode for electricity consumption information data is as follows: The first analysis layer includes a periodic analysis layer, and the second analysis layer includes a similarity analysis layer for data between periods. Obtain different data sequences in the multi-modal customer behavior database and mark them as data sequences to be screened. Perform periodic analysis on each data sequence to be screened to determine whether it has periodicity. If it has periodicity, perform similarity analysis on the data between periods to generate a similarity signal or a non-similarity signal between period segments.

[0006] As a further solution of the present invention: the process of performing periodic analysis and judgment is as follows: Use the method of calculating the autocorrelation coefficient to determine whether the data sequence to be screened has periodicity: Determine the range of lag orders when calculating the autocorrelation coefficient according to the length of the data sequence and the analysis purpose, and calculate the autocorrelation coefficient using the formula; Use the autocorrelation of the Ljung-Box test. The null hypothesis is that the data has no autocorrelation, and calculate the Ljung-Box statistic; After obtaining the statistic, look up the chi-square distribution table according to the degrees of freedom and significance level to obtain the critical value. If the Ljung-Box statistic is greater than the critical value, it indicates that the data sequence to be screened has periodicity.

[0007] As a further solution of the present invention: The specific process of performing similarity analysis on the data between periods is as follows: Obtain the period interval and divide the data sequence to be screened into several complete period segments; Analyze and calculate the data in each period segment to obtain the period segment data characterization value, and analyze and calculate the period segment data characterization value to obtain the period segment data characterization discrete value; If the period segment data characterization discrete value is less than the period segment data characterization discrete threshold, generate a similar signal between period segments. If the period segment data characterization discrete value is greater than or equal to the period segment data characterization discrete threshold, generate a non-similar signal between period segments.

[0008] As a further solution of the present invention: The process of obtaining the period interval is as follows: Fit the autocorrelation coefficient change curve from the lag stage; Mark the peak values of the autocorrelation coefficient in the autocorrelation coefficient change curve, and the lag orders corresponding to these peak values are the period intervals.

[0009] As a further solution of the present invention: The process of obtaining the discrete characterization value is as follows: Calculate the period segment data characterization mean value of all period segment data characterization values, calculate the period segment data characterization amplitude value, and calculate the adjacent change mean value; Perform a ratio calculation on the period segment data characterization amplitude value and the period segment data characterization mean value, and then perform a calculation with the adjacent change mean value to obtain the period segment data characterization discrete value.

[0010] As a further solution of the present invention: The process of obtaining the adjacent change mean value: For each period segment data characterization value, sum and average all the period segment data characterization values adjacent to it to obtain the adjacent period segment mean value; Perform a difference calculation between the period segment data characterization value and the adjacent period segment mean value, and take the absolute value to obtain the adjacent change value, and calculate the adjacent change mean value of the calculated adjacent change values.

[0011] As a further solution of the present invention: The process of obtaining the period segment data characterization value is as follows: Calculate the mean value of the data within each period segment; Arrange all the data in descending order, obtain the quartiles, calculate the difference between the third quartile and the first quartile, and label it as the data fluctuation value of the period segment; Perform weighted calculation on the mean value of the period segment data and the data fluctuation value of the period segment to obtain the data characterization value of the period segment.

[0012] As a further solution of the present invention: specifically, it further includes the following steps: Based on the non-similar signals between period segments, obtain the adjacent change values of the data characterization values of each period segment; Based on any one of the adjacent change values, compare it with the adjacent change threshold in turn. If it is less than the adjacent change threshold, merge the period segment corresponding to the adjacent change value with the adjacent period segment to obtain a merged period segment group. Otherwise, start a new round of merging with the period segment corresponding to the adjacent change value as the starting period segment until all the adjacent change values of all the period segments have been compared with the adjacent change threshold; Obtain all the merged period segment groups and construct models respectively.

[0013] In a second aspect, the present invention provides a multi-modal customer behavior model construction system, which includes: Data acquisition module: acquire the electricity consumption information data of electricity customers and construct a multi-modal customer behavior database; Data analysis module: construct an electricity consumption information data analysis mode according to the analysis requirements of customer behavior.

[0014] Advantages of the present invention: 1. Before constructing the model, the present invention first analyzes the data in the multi-modal customer behavior database. By judging the periodicity of the data sequence, the data with periodicity is screened out and divided into period segments, so that the subsequent model construction focuses on the data with clear features, reduces the waste of computing resources on a large amount of irregular or non-compliant data, improves the pertinence of data processing, and then improves the model training speed; 2. The present invention provides a quantitative basis for model parameter adjustment based on various statistical values calculated from the period segment data, such as the mean value and discrete value of the period segment data characterization; 3. The present invention deeply analyzes the periodicity of the data, finds the period interval, and thus discovers the potential time series law in the data, which is convenient for analyzing the change mode of the customer's electricity consumption behavior in different periods, provides more comprehensive and accurate information for the subsequent model analysis, and enables the model to better capture the data features; 4. The present invention analyzes the similarity degree of data in different cycle segments, enabling analysts to adopt different analysis strategies and models for data of different patterns. For data with similar signals, a model that can capture periodic and similarity features can be selected; for data with dissimilar signals, a more flexible and adaptable model is chosen to improve the accuracy and effectiveness of data analysis, making the model more in line with the actual situation of the data and enhancing the reliability of the analysis results. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0016] Figure 1 is a flowchart of a method for constructing a multi-modal customer behavior model of the present invention; Figure 2 is a schematic structural diagram of a multi-modal customer behavior model construction system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] To enable those skilled in the art to better understand the solution of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Embodiment 1 Figure 1 is a flowchart of a method for constructing a multi-modal customer behavior model provided in Embodiment 1 of the present invention. The embodiments of the present invention are applicable to the situation of analyzing different types of data before constructing a model and then selecting a corresponding suitable model. This multi-modal customer behavior model construction method can be executed by a multi-modal customer behavior model construction system, which can be implemented by software and / or hardware and can be configured in a multi-modal customer behavior model construction device. Optionally, a multi-modal customer behavior model construction device can be an electronic device, which can be a notebook, a desktop computer, a smart tablet, etc., and the embodiments of the present invention do not limit this.

[0019] A method for constructing a multi-modal customer behavior model provided in the embodiments of the present invention specifically includes the following steps: Step 1: Obtain the power consumption information data of electricity customers from the smart meter system and the power billing database, and construct a multi-modal customer behavior database; In some embodiments, obtain the power consumption information data of electricity customers, where the power consumption information data includes but is not limited to: basic information data, power consumption data, power load data, power consumption cost data, and power quality data; Furthermore, the basic information data includes but is not limited to: the customer number, power consumption address, and power consumption category of the customer, which are used for basic classification and positioning of the customer; The power consumption data includes but is not limited to: the power consumption in different time periods, such as daily power consumption, monthly power consumption, and annual power consumption, which is used to reflect the power consumption habits of customers in different power consumption periods; The power load data includes but is not limited to: the maximum load, minimum load, and average load, which are used to analyze the change of the customer's power consumption power and understand the usage scale and pattern of the customer's electrical equipment; The power consumption cost data includes but is not limited to: the electricity bill amount, payment record, and arrears situation, which are used to reflect the customer's power consumption behavior and consumption ability; The power quality data includes but is not limited to: voltage deviation, frequency deviation, and harmonic content, which are used to evaluate the impact of the customer's power consumption on the power grid quality and the possible power quality problems that the customer may suffer; The power consumption information data of electricity customers is obtained through the smart meter and the power technology fee database; The smart meter can collect and record the customer's power consumption data in real time, transmit the data to the main station system of the power company through the power communication network, and directly obtain the real-time and historical power consumption data from the database of the main station system; Exemplarily, through query statements and interfaces, extract the various power consumption information of the corresponding customers according to the customer number or other identification fields; The power billing database stores the technical parameters and cost information related to the customer's power consumption, and can be associated with the data of the smart meter system; Exemplarily, using the unique identifier of the customer (such as the customer number) as the associated field, obtain the relevant cost data and other technical parameters from the power technology fee database to further enrich the content of the power consumption information data; Other technical parameters include but are not limited to: power factor; Step 2: Based on the multi-modal customer behavior database, construct a power consumption information data parsing mode according to the analysis requirements of customer behavior; In some optional implementation manners of some embodiments, the analysis requirement is specifically the periodic requirement in the time series; It should be noted that the reasons for selecting periodic analysis requirements are as follows: First, in the power generation, transmission, distribution, and other links of the power grid, reasonable resource allocation needs to be carried out according to electricity demand. The periodic analysis of electricity consumption behavior can help grid operators understand the peak and trough of electricity consumption in different time periods in advance. Second, if the load of the power grid suddenly fluctuates greatly, it may damage the power grid equipment and even cause power grid failures. Through the periodic analysis of electricity consumption data, grid dispatching personnel can predict the periodic change trend of the load and take corresponding measures in advance. Third, the periodic analysis of electricity consumption behavior can provide a basis for formulating electricity prices. Fourth, periodic analysis can help power grid enterprises evaluate the development needs of the power grid at different stages; Obtain different data sequences in the multi-modal customer behavior database, and mark different types of data sequences as data sequences to be screened respectively; Based on any data sequence to be screened, use the method of calculating the autocorrelation coefficient to judge whether the data sequence has periodicity. The specific process is as follows: According to the length of the data sequence and the analysis purpose, determine the range of the lag order when calculating the autocorrelation coefficient. Generally speaking, the maximum lag order can be set to 1 / 4 - 1 / 3 of the data sequence length; Exemplarily, if there is electricity load data at 360 time points, the maximum lag order can be set to 90 - 120; Use the formula autocorrelation coefficient , and the calculation formula is: , where, represents the t-th data in the data sequence to be screened, is the mean value of the data in the data sequence to be screened, n represents the total number of data, and k represents the lag order; Use the Ljung-Box test for autocorrelation. Its null hypothesis is that the data has no autocorrelation (that is, the data is random and has no periodicity). Calculate the Ljung-Box statistic Q, and the calculation formula is: , where, n represents the total number of data, h represents the set maximum lag stage, represents the autocorrelation coefficient at lag k; After obtaining the statistic Q, look up the chi-square ( ) distribution table according to the degrees of freedom h and the significance level to obtain the critical value , if , then reject the null hypothesis, indicating that the data sequence to be screened has periodicity, otherwise, it indicates that the data sequence to be screened does not have periodicity. If the data sequence to be screened does not have periodicity, other analyses will be performed on the data sequence to be screened, such as trend analysis, etc.; Based on the fact that the data sequence to be screened has periodicity, with the lag stage as the X-axis and the autocorrelation coefficient as the Y-axis, fit the autocorrelation coefficient change curve; Mark the peaks of the autocorrelation coefficient in the autocorrelation coefficient change curve. Then, the lag orders corresponding to these peaks are the period intervals. Exemplarily, if the Ljung-Box test result of the data sequence to be screened shows autocorrelation, and the autocorrelation plot has an obvious peak at the 24th lag order (assuming the data recording interval is 1 hour), it indicates that the data sequence has a periodic change with a period interval of 24 hours. Based on the obtained period interval, divide the data sequence to be screened into several complete period segments. It should be noted that if the length of the sequence to be screened is not an integer multiple of the period interval, the remaining data less than one period is removed. Based on the data within any one period segment. Sum all the data and take the average to obtain the average value of the period segment data. Arrange all the data in descending order and divide it into four equal parts. The values at the splitting points are the quartiles. Among them, the value at the 25% position is called the first quartile, the value at the 50% position is the median, and the value at the 75% position is the third quartile. It should be noted that if the splitting point of the four equal parts is not an integer, round up. For example, if there are 210 data in the period segment, the 25% position is the 52.5th value. At this time, take the 53rd value as the first quartile. Calculate the difference between the third quartile and the first quartile, and mark it as the fluctuation value of the period segment data. Perform a weighted sum calculation on the average value of the period segment data and the fluctuation value of the period segment data to obtain the characterization value of the period segment data. Obtain the characterization values of all period segments, sum all the characterization values of the period segments and take the average to obtain the average characterization value of the period segments. Extract the maximum and minimum values from the characterization values of all period segments, and perform a difference calculation to obtain the amplitude value of the period segment data characterization. Based on the characterization value of any one period segment data, sum and take the average of all the characterization values of the adjacent period segments to obtain the average value of the adjacent period segments. It should be noted that among all the characterization values of the period segment data, there is only the characterization value of the second period segment adjacent to the characterization value of one period segment, and the characterization values of the first and third period segments are adjacent to the characterization value of the second period segment. Perform a difference calculation on the characterization value of this period segment and the average value of the adjacent period segments, take the absolute value of the difference to obtain the adjacent change value, and sum and take the average of all the adjacent change values corresponding to the characterization values of the period segment data to obtain the average adjacent change value. Calculate the ratio of the amplitude value represented by the periodic segment data to the mean value represented by the periodic segment data, and then calculate the product with the adjacent change mean value to obtain the discrete value represented by the periodic segment data; Set the discrete threshold represented by the periodic segment data, where the discrete threshold represented by the periodic segment data is set by those skilled in the art according to experience; If the discrete value represented by the periodic segment data is less than the discrete threshold represented by the periodic segment data, it indicates that the data size changes of the periodic segments are similar, and a similar signal between periodic segments is generated; If the discrete value represented by the periodic segment data is greater than or equal to the discrete threshold represented by the periodic segment data, it indicates that the data size changes of the periodic segments are not similar, and a non - similar signal between periodic segments is generated; Step three: Based on the constructed power consumption information data analysis model, select the corresponding model for construction; In some embodiments, under the condition that the data sequence to be screened has periodicity; Based on the generated similar signal between periodic segments, usually select some models that can capture periodic and similarity features, including but not limited to: Recurrent Neural Network (RNN) and its variants Long Short - Term Memory Network (LSTM), Gated Recurrent Unit (GRU); These models can well handle the long - term dependencies in sequence data. For periodic signals with similar patterns, they can learn the rules and perform effective modeling and prediction. In addition, time - series decomposition models such as Seasonal Decomposition (STL) are also suitable. It can decompose the time series into components such as trend, seasonality, and residuals, which helps to better understand and process similar periodic signals; Reason for selection: Similar signals mean that there is a certain degree of repeatability and regularity in the data in different periodic segments. The above - mentioned models can utilize this rule, through memorizing and learning the information of previous periods, to accurately predict and analyze the current and future periods; And the model parameters can be optimized according to the characteristics of similar periodic segments. For example, the order, learning rate, etc. of the model can be adjusted according to characteristics such as the period length and fluctuation amplitude of similar periodic segments, so that the model can better fit the change patterns of these similar data; Based on the generated non - similar signal between periodic segments, usually select more flexible and adaptable models, including but not limited to: Convolutional Neural Network (CNN) can be used to capture local features and patterns in the signal. Even if the signal is not similar in different periodic segments, there may be some meaningful features locally, and CNN can effectively extract these features; Generative models such as variational autoencoders (VAE) and generative adversarial networks (GAN) can learn the potential distribution of data and have certain advantages in generating and processing non-similar periodic signals. Tree-based models such as decision trees and random forests can handle complex nonlinear relationships and screen and model various features in non-similar signals. Reason for selection: The characteristics of non-similar signals are that the differences between the periodic segments are large and there is no obvious repetitive pattern. The above model can learn and understand the characteristics of the signal from a more complex and diverse perspective, does not rely on fixed periodicity and similarity, and can better adapt to the changes and uncertainties of non-similar signals; The technical solution of this embodiment is: various types of electricity consumption information data of electricity users are obtained through the smart meter system and the electricity billing database, and a multimodal customer behavior database is constructed. Then, based on the database and the periodic demand for customer behavior, the data sequence is analyzed periodically by calculating the autocorrelation coefficient and the Ljung-Box test, etc., the periodic interval is determined and divided into periodic segments, and the periodic segment data characterization value and discrete value are calculated, and then similar or non-similar signals between periodic segments are generated. Finally, according to the generated signal characteristics, the corresponding model is selected for construction; This enables a more precise judgment of the periodicity of the data, accurate division of the periodic segments, and a more detailed analysis of the similarity of data between periods, providing a more accurate data basis for model construction and improving the model's prediction accuracy. Based on the characteristics of similar or dissimilar signals between periodic segments, appropriate models can be selected in a targeted manner. For example, for similar signals, models that are good at capturing periodic and similarity characteristics can be selected, and for dissimilar signals, models that are adaptable to complex changes can be selected. This allows the model to better fit the data and improves the model's adaptability and generalization capabilities.

[0020] Embodiment 2

[0021] Based on the above embodiment 1, Figure 2 As shown, a method for constructing a multimodal customer behavior model provided by an embodiment of the present invention specifically includes the following steps: Based on the non-similar signals between the periodic segments, the adjacent change values ​​of the data representation value of each periodic segment are obtained; Set an adjacent change threshold, and compare any adjacent change value with the adjacent change threshold in turn. If it is less than the adjacent change threshold, merge the cycle segment corresponding to the adjacent change value with the adjacent cycle segment to obtain a merged cycle segment group. Otherwise, start a new round of merging with the cycle segment corresponding to the adjacent change value as the starting cycle segment until the adjacent change values ​​of all cycle segments are compared with the adjacent change threshold. Get all the merged period segments and build models for each; The merged cycle segment groups are formed by merging based on the comparison of adjacent change values with thresholds. The data characteristics within the groups are more convergent, which enables the constructed model to capture the electricity consumption behavior patterns of customers during specific periods more accurately. The merged cycle segment groups can reduce the dispersion degree of data and enable the model to focus on key behavior characteristics; Taking residential electricity consumption as an example, by merging cycle segments with similar electricity consumption patterns and constructing a model, it can be clearly found that customers have high electricity consumption behavior during fixed periods in the evenings on weekdays, accurately grasping the electricity consumption rules of customers and providing strong support for subsequent precise services.

[0022] Embodiment III

[0023] Based on the above Embodiment I and Embodiment II, as Figure 2 shown, a multi-modal customer behavior model construction system provided by an embodiment of the present invention specifically includes: Data acquisition module: Obtain the electricity consumption information data of electricity customers through the smart meter system and the electricity billing database, and construct a multi-modal customer behavior database; Data parsing module: Based on the multi-modal customer behavior database, construct an electricity consumption information data parsing mode according to the analysis requirements of customer behavior; Model construction module: Based on the constructed electricity consumption information data parsing mode, select the corresponding model for construction.

[0024] In the above embodiments, the descriptions of each embodiment have their own emphases. For parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0025] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0026] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.

Claims

1. A method for constructing a multimodal customer behavior model, characterized in that: The following steps are involved: Obtain electricity consumption information data of electricity users, build a multimodal customer behavior database, and build an electricity consumption information data analysis model based on the analysis needs of customer behavior; The power consumption information data parsing mode includes hierarchical progressive analysis, wherein the hierarchical progressive analysis includes a first analysis layer and a second analysis layer, and the second analysis layer is a sub-analysis layer of the first analysis layer.

2. A method for constructing a multimodal customer behavior model according to claim 1, characterized in that: The specific process of constructing the power consumption information data analysis model is as follows: The first analysis layer includes a periodicity analysis layer, and the second analysis layer includes an inter-periodic data similarity analysis layer; Obtain different data sequences in the multimodal customer behavior database and mark them as data sequences to be screened; A periodic analysis is performed on each data sequence to be screened to determine whether it has periodicity. If it has periodicity, a similarity analysis is performed on the data between cycles to generate similar signals between cycle segments or non-similar signals between cycle segments.

3. A method for constructing a multimodal customer behavior model according to claim 2, characterized in that: The process of performing periodic analysis and judgment is as follows: Use the method of calculating the autocorrelation coefficient to determine whether the data sequence to be screened is periodic: According to the length of the data series and the purpose of analysis, determine the range of lag orders when calculating the autocorrelation coefficient, and use the formula to calculate the autocorrelation coefficient; Use the Ljung-Box test for autocorrelation. The null hypothesis is that there is no autocorrelation in the data and calculate the Ljung-Box statistic. After obtaining the statistics, the chi-square distribution table is searched according to the degrees of freedom and significance level to obtain the critical value. If the Ljung-Box statistic is greater than the critical value, it means that the data sequence to be screened has periodicity.

4. A method for constructing a multimodal customer behavior model according to claim 2, characterized in that: The specific process of performing similar analysis on the periodic data is as follows: Obtain the period interval and divide the data sequence to be screened into several complete period segments; Analyze and calculate the data in each period segment to obtain the period segment data characterization value, and analyze and calculate the period segment data characterization value to obtain the period segment data characterization discrete value; If the discrete value representing the periodic segment data is less than the discrete threshold value representing the periodic segment data, a similar signal between periodic segments is generated; if the discrete value representing the periodic segment data is greater than or equal to the discrete threshold value representing the periodic segment data, a non-similar signal between periodic segments is generated.

5. A method for constructing a multimodal customer behavior model according to claim 4, characterized in that: The process of obtaining the periodic interval is as follows: Fitting the self-lag phase-correlation coefficient change curve; The peak value of the autocorrelation coefficient is marked in the autocorrelation coefficient change curve, and the lag order corresponding to the peak value is the cycle interval.

6. A method for constructing a multimodal customer behavior model according to claim 4, characterized in that: The process of obtaining the discrete characterization value is as follows: Calculate the periodic segment data representation mean of all periodic segment data representation values, calculate the periodic segment data representation amplitude value, and calculate the adjacent change mean; The amplitude value representing the periodic segment data is calculated by ratio with the mean value representing the periodic segment data, and then the ratio is calculated with the adjacent change mean to obtain the discrete value representing the periodic segment data.

7. A method for constructing a multimodal customer behavior model according to claim 6, characterized in that: The process of obtaining the adjacent change mean is as follows: For each periodic segment data characterization value, sum and average all the periodic segment data characterization values ​​adjacent to it to obtain the adjacent periodic segment mean; The difference between the data characterization value of the period segment and the mean value of the adjacent period segments is calculated, and the absolute value is taken to obtain the adjacent change value, and the adjacent change mean value of the calculated adjacent change value is obtained.

8. A method for constructing a multimodal customer behavior model according to claim 4, characterized in that: The process of obtaining the periodic segment data characterization value is as follows: Calculate the period data mean of the data in each period; Arrange all the data in descending order, obtain the quartiles, calculate the difference between the third quartile and the first quartile, and mark it as the period data fluctuation value; The mean value of the periodic segment data and the fluctuation value of the periodic segment data are weightedly calculated to obtain the characterization value of the periodic segment data.

9. A method for constructing a multimodal customer behavior model according to claim 6, characterized in that: The specific steps include: Based on the non-similar signals between the periodic segments, the adjacent change values ​​of the data representation value of each periodic segment are obtained; Based on any adjacent change value, it is compared with the adjacent change threshold in turn. If it is less than the adjacent change threshold, the period segment corresponding to the adjacent change value is merged with the adjacent period segment to obtain a merged period segment group. Otherwise, a new round of merging is performed with the period segment corresponding to the adjacent change value as the starting period segment until the adjacent change values ​​of all period segments are compared with the adjacent change threshold. Get all the merged period segment groups and build models for each.

10. A multimodal customer behavior model construction system, characterized in that: The system is used to execute the method described in any one of claims 1 to 9, and the system comprises: Data acquisition module: obtains electricity consumption information data of electricity users and builds a multi-modal customer behavior database; Data analysis module: Build an electricity consumption information data analysis model based on the analysis needs of customer behavior.

Images