Marketing customer label management system and management method based on electric power field

Abstract

The invention discloses a marketing customer label management system and management method based on the electric power field, and relates to the technical field of electric power marketing customer label management, and the system comprises a data collection module, a data storage module, a label generation module, a customer grouping module, a marketing strategy generation module and an interface module. According to the invention, through multi-dimensional data real-time acquisition, dynamic label management, demand elasticity modeling and physical layer accurate intervention, the problems of single data dimension, insufficient real-time performance, unquantified demand elasticity and the like of an existing static label system are solved, synchronous updating of a marketing strategy and user behaviors is realized, the customer satisfaction is improved, power grid resource allocation is optimized, and the user experience is improved. According to the method, the policy compliance is automatically checked, the management risk is reduced, data fusion without additional hardware is realized through power line carrier communication, and a full-closed-loop precision solution from data acquisition to policy intervention is provided for power marketing.

Description

Technical Field

[0001] This invention relates to the field of electricity marketing customer tag management technology, specifically to a marketing customer tag management system and management method based on the electricity sector. Background Technology

[0002] In the field of electricity marketing, customer tagging systems are the core foundation for achieving precise services and differentiated marketing strategies. Currently, the construction of customer tags mainly relies on historical electricity consumption data (such as static indicators like electricity consumption, payment records, and load curves) to form fixed tag classifications (such as "high-energy-consuming users" and "stable-paying users"). However, electricity consumption scenarios exhibit significant volatility and demand elasticity characteristics:

[0003] The impact of time-of-use (TOU) pricing policies: For example, after Yunnan Province adjusted its TOU pricing in 2025, peak hours were changed from 9:00-12:00 and 17:00-22:00 to 7:00-9:00 and 18:00-24:00, while off-peak hours were divided into midday and evening periods. This policy change will lead to a structural shift in users' electricity consumption behavior, such as industrial users adjusting their production shifts to match the midday off-peak pricing, and residential users adjusting their charging times.

[0004] Dynamic adjustments to policy subsidies: For example, environmental protection companies may receive subsidies due to policies, and their electricity consumption patterns may shift from high-energy consumption to green electricity consumption, but the existing labeling system cannot capture such changes in real time.

[0005] Observations of actual marketing scenarios revealed significant shortcomings in the static tagging system in the following situations:

[0006] Lagging marketing strategies: For example, if the power company continues to send peak electricity consumption reminders to users with high load at noon after the time-of-use electricity price adjustment, it may lead to a 15% increase in customer complaint rate, reflecting that the tagging system has failed to keep up with changes in user behavior.

[0007] Demand elasticity is not quantified: For example, after receiving policy subsidies, users in industrial parks have a 30% shift rate in electricity usage periods, but the existing system only records historical load data and cannot quantify the demand elasticity coefficient, resulting in the failure of marketing strategies.

[0008] Limitations of existing technology:

[0009] The inherent flaw of static tagging systems:

[0010] The data dimensions are limited: existing tags are mostly built based on static indicators such as electricity consumption and payment records, lacking quantitative analysis of dynamic behaviors (such as time period shifts and policy responses). For example, although the Inner Mongolia Power Group's Marketing 2.0 system has built a customer tag system, it has not achieved real-time dynamic updates of the tags.

[0011] Lack of cross-domain data integration: Electricity marketing needs to integrate grid operation data, policy variables (such as time-of-use pricing), and user behavior data, but existing systems are mostly deployed in isolation and cannot achieve real-time linkage of cross-domain data.

[0012] The lack of a real-time dynamic adjustment mechanism:

[0013] Insufficient data processing capabilities: Existing systems mostly rely on offline batch processing (such as T+1 daily updates), which cannot meet the needs of minute-level tag updates. For example, although the State Grid Tianjin Electric Power Real-time Marketing System reduces data latency to 5 minutes, it is not coupled with the demand elasticity model and cannot trigger dynamic tag correction.

[0014] Lack of demand elasticity modeling: Existing demand response research (such as Q-learning algorithm) mainly focuses on load transfer and does not deeply integrate elasticity coefficient model into the labeling system, resulting in a disconnect between label updates and changes in user behavior.

[0015] Barriers to interdisciplinary knowledge:

[0016] Difficulties in transferring economic models: The application of demand elasticity theory (such as the calculation of price elasticity coefficient) in the power sector needs to be combined with industry characteristics (such as the design of peak-valley electricity price difference), but existing technologies lack a methodology for deeply integrating economic models with power data.

[0017] Real-time computing technology challenges: Dynamic demand elasticity models need to process high-frequency data (such as minute-level electricity price fluctuations) and adjust labels in real time, which places extremely high demands on computing architecture and algorithm efficiency, and no mature solutions have been found in existing power systems.

[0018] The necessity of solving this problem:

[0019] Improve marketing precision: Dynamic tag updates can synchronize marketing strategies with changes in user behavior. For example, after the adjustment of time-of-use electricity prices, the system can automatically identify users' time-of-use migration behavior and push personalized electricity consumption suggestions, which can improve customer satisfaction.

[0020] Optimize resource allocation: By quantifying demand elasticity, power companies can dynamically adjust the allocation of grid resources. For example, they can guide high-elasticity users to increase electricity consumption during midday off-peak hours, promote the consumption of new energy sources, and reduce the curtailment rate of solar power.

[0021] Compliance and Risk Management: Dynamic policy changes may cause historical labels to become invalid. A dynamic labeling system can automatically verify compliance and avoid policy risks caused by label lag.

[0022] In summary, existing technologies cannot meet the needs of dynamic demand elasticity quantification and real-time tag updates in electricity marketing scenarios. Solving this problem has significant practical implications for improving the marketing efficiency of power companies, optimizing resource allocation, and strengthening compliance management.

[0023] In view of this, a marketing customer tag management system and management method based on the power industry are provided to overcome the above problems. Summary of the Invention

[0024] The purpose of this invention is to provide a marketing customer tag management system and method based on the power industry, so as to solve the problems mentioned in the background art.

[0025] To address the aforementioned technical problems, this invention provides a marketing customer tagging management system based on the power industry, comprising a data acquisition module, a data storage module, a tag generation module, a customer segmentation module, a marketing strategy generation module, and an interface module.

[0026] The data acquisition module is used to collect basic customer information, electricity consumption data, payment records and request information, and to clean and preprocess the data.

[0027] The tag generation module extracts features from preprocessed data based on a preset algorithm, constructs a multi-dimensional tag system that includes basic customer attributes, electricity usage behavior, payment behavior, and complaint behavior, and dynamically updates the tag weights according to changes in customer behavior.

[0028] The customer segmentation module uses a clustering algorithm to group labeled customers and generate customer clusters with different characteristics.

[0029] The marketing strategy generation module generates personalized marketing strategies based on customer cluster characteristics and preset business rules;

[0030] The interface module is used to interact with external systems for data exchange and for sharing and calling tag services.

[0031] Furthermore, the tag generation module integrates multi-source data through knowledge graph technology and combines statistical modeling and machine learning algorithms to generate fact tags, behavioral tags, and predictive tags.

[0032] Furthermore, machine learning algorithms, including at least one of Kmeans++, random forest, and logistic regression, are used to uncover potential customer needs and risk levels.

[0033] Furthermore, the conditions for dynamically updating the label weights include changes in customer electricity load, adjustments to payment cycles, and changes in the type of request.

[0034] Furthermore, the customer segmentation module combines cosine similarity algorithm to analyze the electricity consumption dependence of different customer groups and establish a dynamic electricity consumption pattern model library.

[0035] Furthermore, the marketing strategy generation module evaluates tag quality based on customer segmentation results using a weighted integral algorithm and adjusts strategy content in conjunction with customer service quality sensitivity analysis.

[0036] A marketing customer tag management method based on the power industry includes the following steps:

[0037] Data Acquisition and Preprocessing: Acquire and clean customer data from multiple sources to form a standardized dataset;

[0038] Tag system construction: Data is integrated through knowledge graphs, and multi-dimensional customer tags are generated by combining statistical and machine learning algorithms;

[0039] Dynamic tag management: Adjust tag weights in real time based on changes in customer behavior and update the tag library;

[0040] Customer segmentation: Clustering algorithms are used to segment customers and generate customer clusters with different characteristics;

[0041] Marketing strategy generation: Based on customer cluster characteristics and business rules, generate personalized marketing strategies and push them to the corresponding customers.

[0042] Furthermore, in the dynamic tag management process, changes in customer electricity consumption data, payment records, and request information are monitored to trigger a recalculation of tag weights.

[0043] Furthermore, in the marketing strategy generation process, based on the customer credit rating assessment results, risk warnings and payment reminders are prioritized for high-risk customers.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] I. Improve the precision and real-time nature of marketing:

[0046] Dynamic tag update mechanism: By monitoring changes in electricity load, payment cycle adjustments, and changes in complaint type, the system dynamically triggers tag weight updates, achieving minute-level closed-loop management of "data collection - feature calculation - tag update." This solves the problem of delayed response to changes in user behavior in traditional static tag systems. For example, after adjustments to time-of-use pricing, the system can automatically identify users' time-of-use migration behavior and push personalized suggestions, avoiding an increase in customer complaint rates due to tag lag.

[0047] Multi-dimensional tagging system: Based on knowledge graphs, it integrates multi-source data and combines machine learning algorithms to generate factual tags, behavioral tags, and predictive tags, covering multiple dimensions such as basic customer attributes, electricity consumption behavior, payment behavior, and complaint behavior. It breaks through the limitations of existing technologies that rely solely on static indicators such as electricity consumption and payment records, and improves the comprehensiveness and dynamic adaptability of the tagging system.

[0048] II. Quantifying Demand Elasticity and Optimizing Resource Allocation:

[0049] Real-time calculation of demand elasticity coefficient: By constructing a demand elasticity coefficient model that includes parameters such as harmonic distortion rate, electricity price, load, and voltage, this technology combines physical parameters with economic elasticity theory for the first time, quantifying users' sensitivity to electricity price fluctuations (accuracy ±0.05), thus solving the problem of unquantified demand elasticity in existing technologies. For example, it can dynamically guide highly elastic users to increase electricity consumption during off-peak hours at midday, promoting the consumption of new energy sources and reducing the curtailment rate of solar power.

[0050] Physical layer data fusion and policy intervention: Utilizing power line carrier communication (PLC) technology, real-time cross-domain linkage of harmonics, voltage fluctuations, and policy data is achieved (latency ≤200ms). This is then precisely targeted to users through physical means such as voltage regulation (e.g., OLTC tap changer adjustment) and phase angle adjustment (e.g., SVG compensation). For example, the migration rate of highly flexible users during peak hours is increased to 45% (15% higher than traditional methods), and the regional wind curtailment rate is reduced from 15% to 9%, effectively optimizing grid resource allocation.

[0051] III. Strengthen compliance and risk management:

[0052] Dynamic label verification mechanism: By monitoring policy changes in real time (such as peak-valley time adjustments and subsidy policies), the system automatically verifies label compliance and avoids policy risks caused by the expiration of historical labels. For example, when environmental protection companies adjust their electricity usage patterns due to policy changes, the system can capture the characteristic of a ≥30% decrease in harmonic distortion rate in real time, triggering the update of the "policy subsidy response user" label, ensuring that the label is dynamically synchronized with the policy.

[0053] Highly efficient label update: The improved DBSCAN algorithm combined with a dynamic clustering method based on spatiotemporal weight factors results in a label update latency of ≤3 minutes, which is 480 times more efficient than the traditional T+1 day update. The measured accuracy reaches 92%, significantly reducing the management risks caused by label lag.

[0054] IV. Engineering Application Value:

[0055] A high-precision current transformer and an FFT processing chip are integrated into the smart meter to collect harmonic data in real time (sampling rate 12.8kHz, THD calculation accuracy 0.1%). Multi-source data encoding and transmission are achieved through power line impedance characteristics, eliminating the need for an additional communication network and solving the problems of data isolation and insufficient real-time computing capabilities in existing systems.

[0056] Differentiated strategy support: A dynamic electricity consumption pattern model library is established using the cosine similarity algorithm, and the quality of tags is evaluated using a weighted integral algorithm. Differentiated marketing strategies are generated for different customer clusters (such as high-elasticity migration customers and medium-elasticity response customers). For example, risk warnings are prioritized for high-risk customers, improving service quality sensitivity and strategy effectiveness.

[0057] In summary, this invention optimizes the entire process from data collection to strategy intervention, providing a precise and real-time solution for power marketing, and has significant engineering application value. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a marketing customer tag management system and management method based on the power industry according to the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Please see Figure 1 The present invention provides a technical solution:

[0061] See Figure 1 As shown, an embodiment of a marketing customer tagging management system and method based on the power industry is presented:

[0062] I. System Implementation Example:

[0063] (I) System Architecture and Core Modules:

[0064] 1. Power Grid Harmonic Behavior Sensing Module:

[0065] 1.1 Hardware Architecture: A 0.2S-level precision current transformer (meeting GB / T17215.323-2008 standard) and an ADSP-21489 FFT processing chip are integrated into the SDM630 smart meter to achieve real-time acquisition of the 3rd (150Hz) and 5th (250Hz) harmonic components within the 10-5000Hz frequency band. Specific parameters are as follows:

[0066] Sampling rate: 12.8kHz (satisfies the Nyquist sampling theorem, ensuring no aliasing for signals below 2500Hz);

[0067] Harmonic distortion rate (THD) calculation accuracy: 0.1% (based on spectral analysis of 1024 sampled data using the FFT algorithm, the error is better than the requirements of GB / T14549-1993 standard);

[0068] 1.2 Mapping Library Construction: Based on historical data from over 2000 users in the existing industrial park (collection period of 1 year, containing over 100,000 harmonic samples), a "harmonic-equipment type" mapping library was established using Support Vector Machine (SVM) training. Typical equipment characteristics are shown in Table 1 below:

[0069] Table 1:

[0070] Equipment type Third harmonic distortion rate (THD3) 5th harmonic distortion rate (THD5) Three-phase asynchronous motor ≤5％ 12％-15％ Switching power supply equipment 8％-12％ ≤6％

[0071] 1.3 Policy Synchronization Mechanism: Power line carrier communication (PLC) technology is adopted, and policy data (such as peak-valley time adjustment and subsidy policies) are encoded into a 10kHz carrier signal through OFDM modulation. Specific implementation steps are as follows:

[0072] 1.3.1 Encryption: Data is encrypted using the SM4 national cryptographic algorithm (128-bit key length, conforming to GM / T0002-2012 standard).

[0073] 1.3.2 Coupling: The signal is injected into the power line through a 0.1μF coupling capacitor (meeting IEC60664-1 insulation class).

[0074] 1.3.3 Transmission: Achieve real-time synchronization with government policy platforms and enterprise ERP systems, with a measured latency of 180ms (≤200ms).

[0075] 2. Physical Layer Data Fusion Engine:

[0076] 2.1 Three-element fusion architecture: Establish a three-dimensional feature space of "harmonics-voltage fluctuations-policy", and map the following data to the same coordinate system:

[0077] Harmonic data: 3rd harmonic amplitude U3 (unit: V);

[0078] Voltage fluctuation data: ΔV = V - Vn (V is the real-time voltage, Vn is the rated voltage of 220V, unit: V);

[0079] Policy parameters: Peak-valley price difference ΔP = P_peak - P_valley (P_peak is the electricity price during peak hours, P_valley is the electricity price during off-peak hours, unit: yuan / kWh);

[0080] 2.2 Cross-domain coding method: Utilizing the electric field line impedance characteristic Z = R + jX (R is DC resistance, X is reactance, unit: Ω), different types of data are encoded into sine waves of a specific frequency:

[0081] 2.2.1 Harmonic data: 5kHz (corresponding to the characteristic frequency of the 5th harmonic, 250Hz × 20 times the harmonic frequency, to avoid power frequency interference)

[0082] 2.2.3 Voltage fluctuation: 8kHz (corresponding to the characteristic frequency of power grid fluctuation, the main energy frequency band is determined by Fourier analysis)

[0083] 2.2.3 Policy Data: 12kHz (Dedicated communication frequency band, avoiding the main harmonic distribution range)

[0084] 2.3 Signal Processing Flow:

[0085] 2.3.1 Transmitter: The digital signal is converted into an analog signal by a 16-bit D / A converter (model: AD5628) and superimposed onto the power line (amplitude ≤10V, compliant with IEC61000-3-2 electromagnetic compatibility standard).

[0086] 2.3.2 Receiver: The signal is separated by a bandpass filter (passband error ±0.5kHz, model: MAX274), and the original data is restored by FFT transformation.

[0087] 2.3.3 Correlation Analysis Model:

[0088]

[0089] Parameter explanation:

[0090] M: User migration rate (dimensionless, representing the probability of users migrating from peak hours to off-peak hours);

[0091] ΔV: The difference between the real-time voltage and the rated voltage (V);

[0092] V n Rated voltage (220V);

[0093] ΔP: Peak-valley electricity price difference (yuan / kWh);

[0094] P n Benchmark electricity price (based on the average electricity price of the previous 30 days, unit: yuan / kWh);

[0095] 3. Elasticity coefficient physical calculation module:

[0096] Dynamic model construction: Based on the coupling relationship between harmonic distortion rate (H), electricity price (P), load (L), and voltage (V), a formula for calculating the demand elasticity coefficient is established.

[0097]

[0098] Parameter explanation:

[0099] E: Demand elasticity coefficient (dimensionless, representing the sensitivity of users to electricity price fluctuations);

[0100] ΔH: The difference (%) between the current harmonic distortion rate and the reference value;

[0101] H n : Baseline harmonic distortion rate (average of the previous 7 days, %);

[0102] ΔP: The difference between the current electricity price and the benchmark electricity price (yuan / kWh);

[0103] P n Benchmark electricity price (average of the previous 7 days, yuan / kWh);

[0104] ΔL: The difference between the current load and the reference value (kW);

[0105] L n Baseline load (average of the previous 7 days, kW);

[0106] ΔV: The difference (V) between the current voltage and the rated voltage;

[0107] V n Rated voltage (220V);

[0108] Edge computing implementation: Deploy a random forest algorithm (containing 50 decision trees) on an edge node of model UNO-2483, and input grid parameters in real time.

[0109] Current I (accuracy ±0.5%, from smart meter);

[0110] Voltage V (accuracy ±0.2%, from voltage transformer);

[0111] The frequency f (accuracy ±0.01Hz, calculated via zero-crossing detection algorithm) outputs model parameter weights (update period 1 minute) to adapt to different device characteristics.

[0112] For motor users: harmonic distortion rate weight 0.4, voltage fluctuation weight 0.3;

[0113] For charging station users: electricity price parameter weight 0.5, load change weight 0.25;

[0114] 4. Tag Physical Trigger Update Module:

[0115] 4.1 Trigger Rule Definition:

[0116] 4.1.1 Highly flexible users migrating during midday:

[0117] Time period: Midday off-peak period (12:00-14:00, which can be dynamically adjusted through the policy synchronization module)

[0118] Conditions: ΔV ≥ +5%Vn (i.e., ΔV ≥ 11V) and H ≤ 80%Hn (H is the real-time harmonic distortion rate, H n (Base value)

[0119] 4.1.2. Policy subsidy response users:

[0120] Triggering condition: The harmonic distortion rate H decreases by ≥30% compared to the baseline value before the policy took effect.

[0121] (Right now );

[0122] The policy subsidy is effective within the 72-hour validity period (starting from the time the policy data is synchronized to the system);

[0123] 4.2 Dynamic Clustering Algorithm: An improved DBSCAN algorithm is adopted, introducing a spatiotemporal weight factor to construct a distance metric formula, thereby achieving multi-dimensional clustering of user features.

[0124] d=α×d 谐波 +β·d 电压 +γ·d 时空 ;

[0125] Parameter explanation:

[0126] Weighting coefficients: α = 0.5, β = 0.3, γ = 0.2 (determined by the Analytic Hierarchy Process (AHP) to reflect the priority of the influence of harmonic characteristics, voltage fluctuations, and spatiotemporal attributes on user classification).

[0127] Harmonic characteristic Euclidean distance (d 谐波 ):

[0128] The Euclidean distance between the user's harmonic eigenvectors [THD3, THD5] is calculated using the following formula:

[0129]

[0130] Where THD3 is the third harmonic distortion rate (%), and THD5 is the fifth harmonic distortion rate (%).

[0131] Voltage fluctuation Euclidean distance d 电压 :

[0132] Calculate the real-time voltage fluctuation ΔV = VV n Euclidean distance (V) n =220V is the rated voltage), the formula is:

[0133] d 电压 =|ΔV i -ΔV j |;

[0134] Manhattan Distance in Spacetime d 时空 :

[0135] Distance metric integrating geographic location and time-of-day differences: - Geographic location: Using Gaussian projected coordinates (x, y), calculate the sum of the absolute values ​​of the coordinate differences |x i -x j |+|y i -y j |;

[0136] Time difference: Convert the user's electricity consumption time period to minutes (e.g., 12:00 corresponds to 720 minutes, 14:00 corresponds to 840 minutes), and calculate the absolute value of the time difference |t i -t j |;

[0137] Comprehensive formula: (Unit: km + hour);

[0138] Update mechanism: Triggering condition: When the fluctuation of power grid parameters exceeds the threshold (e.g., |ΔV|>3%V) n When ΔV > 6.6V or ΔV > -6.6V, the edge node sends a tag update command via the PLC.

[0139] Performance metrics:

[0140] The measured tag update latency is 180 seconds (≤3 minutes), which meets the real-time requirements;

[0141] By quantifying the coupling relationship between spatiotemporal characteristics and physical parameters, the problem of lagging response of traditional tagging systems to dynamic changes in user behavior is solved, realizing minute-level closed-loop management of "data acquisition-feature calculation-tag update", and providing precise technical support for power demand-side management.

[0142] 5. Physical Strategy Intervention Module:

[0143] 5.1 Voltage Regulation Strategy:

[0144] 5.1.1 Highly elastic users (E≥0.8):

[0145] Regulation method: The voltage is increased by 3% during the midday off-peak period (220V→226.6V, within the ±10% allowable range specified in IEC60038) by adjusting the tap of the on-load tap changing transformer (OLTC, model: UTK-100).

[0146] Adjustment accuracy: 1.25% (each adjustment step is 0.55V, meeting GB / T1094.11-2021 standard);

[0147] 5.1.2 Medium-elasticity users (0.5≤E<0.8):

[0148] Intervention method: Push customized messages via the electricity marketing APP, including:

[0149] Price incentives: "Midday electricity price discount of 0.2 yuan / kWh (off-peak price 0.35 yuan / kWh → 0.15 yuan / kWh)";

[0150] Environmental information: "Currently, the renewable energy power supply rate is 85% (60% photovoltaic and 25% wind power)";

[0151] 5.2 Phase Angle Adjustment: For users connected to distributed photovoltaic systems, the voltage phase angle Δθ on the user side is adjusted to 15° (adjustment accuracy ±2°) using a static var generator (SVG, model: ANAPF-400). Specific implementation details are as follows:

[0152] 5.2.1 Real-time monitoring of the output power factor of the photovoltaic inverter (target ≥ 0.95);

[0153] 5.2.2. The SVG compensation current is dynamically adjusted through a PID control algorithm to achieve precise phase angle adjustment;

[0154] Effectiveness evaluation: Data is collected every 15 minutes to verify the effectiveness of the strategy. Judgment criteria:

[0155] The harmonic distortion rate H decreased by ≥10% (compared to before adjustment);

[0156] The voltage fluctuation ΔV returns to within ±2%Vn (i.e., ΔV∈[-4.4V,+4.4V]);

[0157] II. Implementation Examples of Management Methods:

[0158] (I) Methodology and Core Steps:

[0159] 1. Harmonic and Policy Synchronous Data Acquisition Steps:

[0160] Data collection:

[0161] The smart meter triggers sampling every 10 seconds, and the FFT chip calculates THD3 and THD5 in real time with an error of ≤0.5% (chip drift is corrected through a temperature compensation algorithm);

[0162] Receive policy data via PLC and parse examples:

[0163] Peak and off-peak hours: After adjustments in 2025, Yunnan Province's peak hours are 7:00-9:00 and 18:00-24:00 (off-peak hours are 11:00-17:00);

[0164] Subsidy Policy: Environmental protection enterprises will receive a subsidy of 0.15 yuan / kWh during off-peak hours (valid from January 1, 2025 to December 31, 2025).

[0165] Anomaly detection: When THD5 exceeds 15% for three consecutive cycles (30 seconds) (e.g., THD5 = 16%, 17%, 16% for a certain motor user), the "Equipment Overload Warning" label is triggered. This label is used to assist in the calculation of the elasticity coefficient and does not directly affect the marketing strategy.

[0166] Breaking through the limitations of traditional electricity consumption statistics, it can perceive the operating status of equipment in real time through harmonic characteristics (such as improving the accuracy of motor start-stop identification);

[0167] Data acquisition latency has been reduced from T+1 day to 10 seconds, meeting the requirements of real-time demand response (DR) scenarios;

[0168] A new dimension for device operating status has been added to address the issue of limited data dimensions.

[0169] 2. Physical layer data fusion steps:

[0170] Feature extraction: Taking existing industrial users as an example, construct feature vectors:

[0171] THD3 = 10%, THD5 = 14% (normal operating characteristics of a three-phase asynchronous motor);

[0172] ΔV = +5V (real-time voltage 225V, rated voltage 220V);

[0173] ΔP = 0.5 yuan / kWh (0.85 yuan / kWh during peak hours and 0.35 yuan / kWh during off-peak hours);

[0174] Cross-domain coding and transmission: Modulating the feature vector using OFDM to:

[0175] 5kHz (THD3 encoded signal);

[0176] 8kHz (ΔV encoded signal);

[0177] A 12kHz (ΔP encoded signal) is superimposed onto a power line (amplitude 8V). After transmission over 500m, the FFT parsing error at the receiving end is ≤1% (achieved through synchronous clock calibration technology).

[0178] Association analysis: Through training with historical data (n = 100,000 records), we found that:

[0179] Voltage fluctuations are strongly correlated with time-of-use migration (R0). 2 =0.89);

[0180] For every 1% increase in ΔV (2.2V), the proportion of electricity consumption by industrial users during midday increases by 2.3% (95% confidence level).

[0181] Cross-domain data fusion can be achieved by leveraging the inherent physical characteristics of power lines, without the need to deploy additional communication networks;

[0182] Data linkage latency is ≤200ms, meeting the requirements for minute-level policy updates;

[0183] This addresses the problem of isolated data in existing systems, enabling real-time correlation of harmonic, voltage, and policy data.

[0184] 3. Steps for real-time calculation of elasticity coefficient:

[0185] Model Input and Calculation: Taking a user in an industrial park as an example (midday off-peak period):

[0186] Real-time parameters: L = 500kW (current load), V = 225V (real-time voltage), H = 12% (real-time THD5), P = 0.35 yuan / kWh (off-peak electricity price);

[0187] Reference parameter: L n =625kW (average load of the previous 7 days), V n =220V (rated voltage), H n =15% (average THD5 of the previous 7 days), P n =0.6 yuan / kWh (average electricity price of the previous 7 days) Substituted into the elasticity formula:

[0188]

[0189] The elasticity coefficient is taken as an absolute value, and in practical applications, attention is paid to the magnitude of the absolute value of the demand elasticity.

[0190] Model optimization: The random forest algorithm updates the feature weights every 10 minutes and found that the voltage phase angle of photovoltaic users has a weight of 0.32 on the elasticity coefficient, which is higher than the harmonic parameter weight of traditional industrial users. Therefore, the phase angle feature input is added to the model.

[0191] This is the first time that physical parameters such as harmonic distortion rate and voltage fluctuation have been combined with the economic elasticity theory.

[0192] The quantitative demand elasticity coefficient (accuracy ±0.05) addresses the lack of elasticity modeling in existing technologies.

[0193] This provides a scientific basis for differentiated strategies, such as prioritizing voltage regulation for highly resilient users.

[0194] 4. Tag Physical Trigger Update Steps:

[0195] Initial tag generation: A user's midday off-peak period:

[0196] ΔV = +6V (+2.7% Vn), H = 12% (compared to H) n =15% decrease (20% decrease);

[0197] If the "potential midday migration user" label is triggered and the conditions are met for three consecutive time periods (30 minutes), the user will be upgraded to an "actual midday migration user".

[0198] Dynamic clustering implementation: Clustering was performed on 100 users in the existing industrial park (ε = 0.2, MinPts = 5), and the results are shown in Table 2 below:

[0199] Table 2:

[0200]

[0201]

[0202] Real-time update verification: When the new peak and valley periods in Yunnan Province take effect in 2025, the system completes the update of 300,000 user tags within 3 minutes, which is 480 times more efficient than the traditional T+1 day update. The actual tag accuracy rate is 92% (verified by manual sampling).

[0203] Labels are triggered based on physical parameter thresholds to avoid classification bias caused by subjective judgment;

[0204] Dynamic classification is achieved by combining spatiotemporal clustering to adapt to the spatiotemporal differences in load distribution;

[0205] Tag update delay ≤ 3 minutes, solving the problem of lagging marketing strategies and reducing the predicted customer complaint rate.

[0206] 5. Physical strategy intervention steps:

[0207] Voltage regulation implementation: OLTC regulation is implemented for highly resilient users (E=0.85).

[0208] Before adjustment: Midday electricity consumption was 1000 kWh, with renewable energy accounting for 55% and THD5 = 15%;

[0209] After adjustment: Midday electricity consumption was 1280 kWh (an increase of 28%), the proportion of renewable energy was 75% (an increase of 20%), and THD5 was 13% (a decrease of 13.3%).

[0210] Phase angle adjustment implementation: After a wind farm connected to a user and adjusted the phase angle Δθ = 15°:

[0211] Midday wind power usage increased from 500 kWh to 700 kWh (an increase of 40%).

[0212] The regional wind curtailment rate decreased from 15% to 9% (monitored in real time via the SCADA system);

[0213] Strategy Iteration: If a user adjusts the THD5 and it rises to 16% (equipment overheating alarm), the system will automatically reduce the voltage adjustment range to 2% (224.4V) to balance power consumption and equipment safety. After adjustment, THD5 will drop back to 14%, and the equipment operating temperature will be normal.

[0214] Precise delivery is achieved through physical layer intervention (voltage and phase angle adjustment), avoiding the inefficiency of traditional SMS notifications;

[0215] The migration rate of highly flexible users during peak hours has increased to 45% (15% higher than traditional methods), and the amount of renewable energy consumed has increased.

[0216] It addresses issues related to resource allocation optimization and compliance management, and the adjustment process complies with standards such as IEC60038 and GB / T12325.

[0217] Summarize:

[0218] Multi-dimensional data acquisition: Harmonic, voltage and policy data are perceived in real time through high-precision hardware, and the acquisition latency is reduced from T+1 day to 10 seconds, improving the accuracy of equipment status identification.

[0219] Cross-domain data fusion: Data fusion without additional hardware is achieved using power line carrier, with a linkage latency of ≤200ms, solving the problem of data isolation and constructing a three-dimensional feature space.

[0220] Precise elasticity modeling: Based on the elasticity coefficient formula of physical parameters and economic theory, with an accuracy of ±0.05, it supports differentiated responses of equipment, such as adaptive adjustment of the weights of motors and charging piles.

[0221] Dynamic label management: Spatiotemporal clustering is achieved by improving the DBSCAN algorithm, with label update latency ≤3 minutes and classification accuracy ≥92%, solving the problem of strategy lag.

[0222] Precise intervention at the physical layer: Combining voltage regulation with phase angle regulation, the time-period mobility is increased by 45%, the amount of new energy consumption is increased, and the equipment safety and compliance requirements are met at the same time.

[0223] In summary, this system and method realize a closed-loop management system from data collection to policy intervention, providing a precise solution for power marketing based on physical layer characteristics, and has significant engineering application value.

Claims

1. A marketing customer tagging management system based on the power industry, comprising a data acquisition module, a data storage module, a tag generation module, a customer segmentation module, a marketing strategy generation module, and an interface module, characterized in that: The data acquisition module is used to collect basic customer information, electricity consumption data, payment records and request information, and to clean and preprocess the data. The tag generation module extracts features from preprocessed data based on a preset algorithm, constructs a multi-dimensional tag system that includes basic customer attributes, electricity usage behavior, payment behavior, and complaint behavior, and dynamically updates the tag weights according to changes in customer behavior. The customer segmentation module uses a clustering algorithm to group labeled customers and generate customer clusters with different characteristics. The marketing strategy generation module generates personalized marketing strategies based on customer cluster characteristics and preset business rules; The interface module is used to interact with external systems for data exchange and for sharing and calling tag services.

2. The marketing customer tag management system based on the power sector as described in claim 1, characterized in that: The tag generation module integrates multi-source data through knowledge graph technology and combines statistical modeling and machine learning algorithms to generate fact tags, behavioral tags, and predictive tags.

3. The marketing customer tag management system and method based on the power sector as described in claim 2, characterized in that: Machine learning algorithms include at least one of Kmeans++, random forest, and logistic regression, used to uncover potential customer needs and risk levels.

4. The marketing customer tag management system and method based on the power sector as described in claim 1, characterized in that: The conditions for dynamically updating tag weights include changes in customer electricity load, adjustments to payment cycles, and changes in the type of request.

5. The marketing customer tag management system and method based on the power sector as described in claim 1, characterized in that: The customer segmentation module combines cosine similarity algorithm to analyze the electricity consumption dependence of different customer groups and establish a dynamic electricity consumption pattern model library.

6. The marketing customer tag management system and method based on the power industry as described in claim 1, characterized in that: The marketing strategy generation module evaluates tag quality based on customer segmentation results using a weighted integral algorithm and adjusts strategy content in conjunction with customer service quality sensitivity analysis.

7. A marketing customer tag management method based on the power sector, characterized in that, Includes the following steps: Data Acquisition and Preprocessing: Acquire and clean customer data from multiple sources to form a standardized dataset; Tag system construction: Data is integrated through knowledge graphs, and multi-dimensional customer tags are generated by combining statistical and machine learning algorithms; Dynamic tag management: Adjust tag weights in real time based on changes in customer behavior and update the tag library; Customer segmentation: Clustering algorithms are used to segment customers and generate customer clusters with different characteristics; Marketing strategy generation: Based on customer cluster characteristics and business rules, generate personalized marketing strategies and push them to the corresponding customers.

8. A marketing customer tag management method based on the power sector as described in claim 7, characterized in that: In the dynamic tag management process, changes in customer electricity consumption data, payment records, and request information are monitored to trigger a recalculation of tag weights.

9. A marketing customer tag management method based on the power sector as described in claim 7, characterized in that: In the marketing strategy generation process, based on the customer credit rating assessment results, risk warnings and payment reminders are prioritized for high-risk customers.

Images