An interactive method and system for power marketing

Abstract

The application provides an interactive method and system for power marketing, the method comprising: collecting power consumption data of users through an Internet of Things device and preprocessing the data; constructing a power consumption behavior prediction model according to the preprocessed data to predict the future power demand of the users and respond to sudden changes; constructing a dynamic power pricing model according to the prediction results, adjusting the power pricing according to the actual demand of the users and the change of power grid load; monitoring the market condition in real time, combining the trained power consumption behavior prediction model and the dynamic power pricing model to timely find potential abnormalities; and recommending the optimal power package for each user through an optimization algorithm and making real-time adjustments. The application not only improves the accuracy and effect of power company marketing, but also enhances the interactivity and energy-saving awareness of users through a social mechanism, and promotes the transformation of the power industry to intelligence, individualization and socialization.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of power marketing, and particularly relates to an interactive method and system for power marketing. BACKGROUND

[0002] As a key component in the energy industry, power marketing has traditionally relied on static pricing models and periodic promotional activities to attract consumers. However, with the increasing competition in the power market and the increasing diversification of consumer demand, traditional power marketing methods have gradually exposed their limitations. First, power demand itself has significant seasonal and temporal characteristics, for example, electricity consumption is higher in winter and summer, and power demand in the evening and morning is often higher than in the daytime. In addition, due to factors such as consumer lifestyle, number of family members, and residential area, power consumption behavior often shows individualization and dynamic change. Therefore, a single power pricing model and unified marketing activities are difficult to respond quickly to these changes, resulting in poor marketing effectiveness, decreased customer satisfaction, and potential loss of profits for power companies.

[0003] On the other hand, with the rapid development of smart home and Internet of Things (IoT) technologies, consumers' power consumption behavior has gradually become more intelligent and traceable. Traditional power marketing models cannot fully utilize the large amount of real-time data provided by these smart devices, missing the opportunity for personalized recommendations and dynamic pricing. Although some power companies have tried to use big data analysis and machine learning-based technologies to predict users' electricity consumption behavior, these technologies still face some challenges. For example, although existing smart meters and home appliances can provide accurate electricity consumption data, there is a lack of effective ways to convert these data into business intelligence that can drive accurate recommendations of power products and services. In addition, although personalized recommendations have become an important means for many industries to improve user experience, personalized recommendations in power marketing still lack in-depth analysis of consumers' electricity consumption behavior, resulting in recommended content that is usually based on rough speculation of historical consumption data, lacking real-time and accuracy. Therefore, existing power marketing systems are still insufficient in responding to user demand and providing accurate marketing solutions.

[0004] Meanwhile, the rise of social media and social platforms also brings new challenges and opportunities to the power marketing model. Many industries have promoted product sales through social recommendations and user interactions, while the power industry is also exploring this direction, but most attempts still remain at the level of simple promotion and interaction, failing to truly combine social recommendation mechanisms with personalized needs of power consumption behavior. The consumption behavior, energy-saving experience and product recommendations shared by users on social platforms have a certain reference value for other users, but the existing power marketing systems mostly fail to effectively integrate such social data with personalized recommendation algorithms, resulting in insignificant social marketing effects, low user engagement, and the inability of power product recommendations and promotions to fully leverage the advantages of socialization.

[0005] Therefore, the existing power marketing technology faces many problems, mainly reflected in the following aspects: first, the traditional pricing and recommendation methods cannot be personalized according to the real-time power consumption behavior and time patterns of users; second, the data generated by smart devices cannot be effectively utilized to provide personalized and dynamic marketing recommendations for power companies; third, the application of social recommendation mechanisms in power marketing is still immature, with low user engagement and interactivity, and social recommendations cannot effectively promote product sales. Therefore, how to combine advanced technologies such as smart Internet of Things, big data, and machine learning to provide accurate marketing recommendations for personalized power needs, while encouraging user participation and interaction through social platforms, has become a technical problem that needs to be solved in the field of power marketing. SUMMARY

[0006] The purpose of the present application is to provide an interactive method and system for power marketing by collecting real-time power consumption data generated by smart home devices and combining user time perception behavior to predict future power needs of users through advanced machine learning algorithms, thereby providing personalized power product recommendations for users. This technical breakthrough can effectively solve the problems of insufficient personalized recommendations and untimely dynamic response in existing power marketing systems.

[0007] To achieve the above purpose, in the first aspect of the present application, an interactive method for power marketing is provided, the method comprising:

[0008] S1, collecting power consumption data of a user through an Internet of Things device, and preprocessing the data to obtain preprocessed power consumption data, and introducing user social behavior as user behavior characteristics, and collecting the diversity of home devices and changes in environmental factors as external environmental factors; wherein the power consumption data includes power consumption, device status and timestamp;

[0009] S2, constructing a power consumption behavior prediction model based on the preprocessed data to predict the future power demand of the user and respond to sudden changes; wherein the power consumption behavior prediction model adopts a time series convolutional neural network, extracts high-order features from time series data of power consumption through a convolutional layer, identifies long-term dependencies, and in the training process, the convolutional layer and the pooling layer work together to extract the potential law of user power consumption, and the fully connected layer fuses the time series features with user behavior features and external environmental factors to predict power demand;

[0010] Wherein, the power consumption behavior prediction model designed based on the time series convolutional neural network is a multi-task learning framework, which optimizes two targets: power demand prediction and user behavior prediction, wherein the loss function of the power consumption behavior prediction model is designed as a weighted combination of the mean square error loss of power demand prediction and the cross-entropy loss of user behavior classification;

[0011] S3, constructing a dynamic power pricing model according to the prediction results, and adjusting the power pricing according to the actual demand of the user and the load change of the power grid; wherein the dynamic power pricing model is constructed as follows:

[0012] Design a dynamic power pricing mechanism combining power demand prediction and user behavior prediction, and flexibly adjust power supply according to real-time demand and price;

[0013] Design multiple constraint conditions, including supply and demand balance constraint and fairness constraint;

[0014] Use a deep Q learning algorithm based on reinforcement learning to dynamically adjust the price strategy; wherein the deep Q learning algorithm is represented as:

[0015] Q(s t , a t )←Q(s t , a t )+α[r t+1 +γmax a′ Q(s t+1 , a′)-Q(s t , a t )]

[0016] Wherein, s t is the current state, a t is the selected pricing strategy, r t+1 is the actual income, and γ is the discount factor;

[0017] S4, real-time data flow monitors market conditions, combines the trained power consumption behavior prediction model and dynamic power pricing model, and timely discovers potential abnormalities, and if abnormalities are found, intelligent early warning is performed;

[0018] S5, combine historical data and real-time data of power marketing, recommend the optimal power package for each user through optimization algorithm, and make real-time adjustment to ensure that the most suitable power scheme is provided for the user under different conditions.

[0019] Further, the preprocessing includes:

[0020] Interpolation based on user historical consumption patterns: assuming that the power consumption pattern of the user in a certain time period has certain regularity, the missing values are predicted through historical data;

[0021] Standardization method based on device category: in the standardization process, the mean and standard deviation of each device are calculated according to the historical power consumption data of each device, and then the power consumption data at the current time is standardized according to the following formula;

[0022] Multi-time scale processing: for the intra-day period, the intra-week period and the holiday, the algorithm based on time sequence feature nested weighted average is adopted, the influence of historical data on the current moment is adjusted through different weights, and the preprocessed power consumption data is obtained;

[0023] Data feature expansion is introduced, user social behavior is collected as user behavior feature, and the diversity of household devices and the change of environmental factors are collected as external environmental factors.

[0024] Further, for the power demand prediction regression task, the loss function is:

[0025]

[0026] wherein, is the actual power demand, is the predicted power demand;

[0027] For the user behavior classification task, the loss function is:

[0028]

[0029] wherein, y i is the actual label of user behavior, is the predicted behavior probability;

[0030] Finally, the loss function of the power consumption behavior prediction model is:

[0031]

[0032] wherein, λ1 and λ2 are the weight coefficients between tasks.

[0033] Further, to improve the generalization ability of the power consumption behavior prediction model and avoid overfitting, an adaptive regularization mechanism is introduced, which dynamically adjusts the strength of the regularization term based on the changes in user behavior data, avoiding imposing the greatest constraint on users with the greatest changes in behavior patterns, represented as:

[0034]

[0035] where, is the L2 norm of user behavior features, and α t is the regularization strength at the current time. β1 and β2 are adjustment parameters;

[0036] The Adam optimizer is used to train the model, combined with the backpropagation algorithm to minimize the loss function , and a learning rate decay strategy is used to improve the stability of the training process;

[0037] After training, the model will output the user's future power demand prediction and the probability of user behavior change, and then use the RMSE and AUC two indicators to evaluate the accuracy of power demand prediction and the effectiveness of user behavior prediction, respectively.

[0038] For power demand prediction, the RMSE calculation formula is:

[0039]

[0040] where, is the user's future power demand prediction, is the user's actual power demand, N is the total number of samples, and i is the current user i;

[0041] For user behavior prediction, the AUC calculation formula is:

[0042]

[0043] where, TPR is the true positive rate, FPR is the false positive rate; the AUC value closest to 1 indicates that the model performs best in the classification task.

[0044] Further, the dynamic power pricing mechanism includes a basic power price and a dynamic adjustment factor, the basic power price is a basic pricing based on the current market power cost and long-term trend prediction, and the dynamic adjustment factor is a price fluctuation factor adjusted based on power demand prediction and user behavior prediction, the dynamic power pricing mechanism is represented as:

[0045] P t = P base + ΔP t

[0046] where, Pt Pmarket is the market electricity price base Pbase is the base electricity price, ΔP t E is the dynamic adjustment part based on electricity demand prediction user behavior prediction, which is calculated as follows:

[0047]

[0048] wherein γ1 and γ2 are adjustment coefficients, E is the user behavior prediction, avg Pbase is the average electricity demand in a certain period of time;

[0049] The supply-demand balance constraint is used to maintain market supply-demand balance, and is expressed as:

[0050]

[0051] wherein S t Pmarket is the electricity supply at time t, ensuring that the pricing matches the demand; T is the total duration;

[0052] The fairness constraint is used to set the price fluctuation range to not exceed a certain proportion, and is expressed as:

[0053] ΔP t ≤α·P base

[0054] wherein α is a preset maximum price fluctuation proportion, used to prevent excessive price fluctuations from affecting low-income users.

[0055] Further, the S3 further comprises:

[0056] Two types of evaluation indexes are used to evaluate the effect of the dynamic electricity pricing model, including supply-demand matching degree and user satisfaction; the supply-demand matching degree Mismatch t is the difference between electricity demand and supply at each time, expressed as:

[0057]

[0058] The user satisfaction is based on the user's response behavior, calculates the total consumption of the user under the adjusted pricing, and analyzes the acceptance of the pricing strategy.

[0059] Further, the S4 specifically comprises:

[0060] S401, obtaining operation data from each power user and power grid through a real-time data interface; wherein the operation data includes real-time electricity demand, electricity supply, electricity market price and historical user behavior data;

[0061] S402, monitor the current power demand and supply and compare, identify potential abnormalities;

[0062] S403, based on real-time feedback, set optimization target according to power demand and supply, optimize pricing strategy, adjust market electricity price, realize supply and demand balance; wherein, the objective function of the optimization target is expressed as:

[0063]

[0064] Wherein, E t and S t are power demand and supply, P t is market electricity price, is the electricity price predicted by the model, and λ is the weight factor of balancing supply and demand and electricity price fluctuation.

[0065] Further, the S402 specifically includes:

[0066] Abnormalities are identified by calculating the deviation between the current actual demand and the predicted demand, which is expressed as:

[0067]

[0068] Wherein, E t represents the actual power demand, is the expected demand based on historical data and prediction model. If the deviation ΔE t exceeds the set threshold, it is judged as demand abnormality;

[0069] Similarly, if the supply amount (S t ) and the model predicted supply capacity ( ) difference exceeds the threshold, the supply abnormality alarm will be triggered:

[0070]

[0071] When the deviation value ΔE t or ΔS t exceeds the preset threshold, the intelligent early warning is automatically triggered.

[0072] Further, the S5 specifically includes:

[0073] S501, design a personalized electricity package recommendation model, expressed as:

[0074]

[0075] Wherein, E t is the actual power demand of the user in period t, is the user power demand prediction based on the prediction model, P rec(t) is the recommended electricity package price, P t is the actual electricity market price, A i (t) is the behavior pattern of the user at time period t, A rec (t) is the recommended user behavior pattern, and γ and λ are adjustment coefficients for controlling the balance between electricity price fluctuations and behavior patterns;

[0076] S502, when the user accepts the recommended package, the electricity consumption behavior prediction model is adjusted to optimize the electricity package by monitoring the user feedback in real time, so that the recommended package is more adaptive; wherein the objective function of the electricity consumption behavior prediction model for optimizing the electricity package is represented as:

[0077]

[0078] Wherein, U t is the actual electricity consumption behavior history data of the user at time period t, U rec is the user electricity consumption history recommended by the system based on the prediction result, and δ, μ and θ are dynamic adjustment coefficients for balancing different optimization objectives;

[0079] S503, when the user uses the electricity package, the electricity consumption behavior will be collected in real time through feedback, and the effect of the recommended package is automatically evaluated, and the subsequent recommendation strategy is adjusted accordingly; wherein the effect of the recommended package is automatically evaluated by the reinforcement learning model, the recommendation process is optimized based on the historical feedback data, and the objective function is:

[0080]

[0081] Wherein, α, β and γ are dynamic adjustment coefficients for controlling the effect of feedback results in the optimization process.

[0082] In another aspect of the application, an interactive system for electricity marketing is provided, which comprises:

[0083] The electricity consumption data acquisition unit is used to collect the electricity consumption data of the user through the Internet of Things device, and to preprocess the data to obtain the preprocessed electricity consumption data, and to introduce the user social behavior as the user behavior characteristics, and to collect the diversity of the household equipment and the change of the environmental factors as the external environmental factors; wherein the electricity consumption data includes the electricity consumption, the equipment state and the time stamp;

[0084] The power demand prediction unit is configured to predict the future power demand of the user according to the preprocessed data and to respond to sudden changes, wherein the power consumption behavior prediction model adopts a time series convolutional neural network, extracts high-order features from time series data of power consumption through a convolutional layer, identifies long-term dependencies, and in the training process, the convolutional layer and the pooling layer jointly act to extract the potential law of the user's power consumption, and the fully connected layer fuses the time series features with the user behavior features and external environmental factors to perform power demand prediction.

[0085] The power consumption behavior prediction model is designed based on the time series convolutional neural network as a multi-task learning framework, which optimizes two targets: power demand prediction and user behavior prediction, wherein the loss function of the power consumption behavior prediction model is designed as a weighted combination of the mean square error loss of power demand prediction and the cross-entropy loss of user behavior classification.

[0086] The dynamic pricing unit is configured to construct a dynamic power pricing model according to the prediction results, and to adjust the power pricing according to the actual demand of the user and the load variation of the power grid, wherein the dynamic power pricing model is constructed as follows:

[0087] The dynamic power pricing mechanism is designed in combination with power demand prediction and user behavior prediction, and the power supply is flexibly adjusted according to real-time demand and price;

[0088] A plurality of constraint conditions are designed, including supply and demand balance constraint and fairness constraint;

[0089] A deep Q learning algorithm based on reinforcement learning is used to dynamically adjust the price strategy, wherein the deep Q learning algorithm is represented as:

[0090] Q(s t , a t )←Q(s t , a t )+α[r t+1 +γmax a′ Q(s t+1 , a′)-Q(s t , a t )]

[0091] Wherein, s t is the current state, a t is the selected pricing strategy, r t+1 is the actual income, and γ is the discount factor.

[0092] The power consumption data anomaly alarm unit is configured to monitor market conditions in real time, combine the trained power consumption behavior prediction model and dynamic power pricing model, and timely discover potential abnormalities, and if abnormalities are found, intelligent early warning is performed.

[0093] The personalized recommendation unit is used to recommend the optimal power package for each user through an optimization algorithm combined with historical data and real-time data of power marketing, and to make real-time adjustments to ensure that the most suitable power solution is provided for the user under different conditions.

[0094] The beneficial technical effects of the present application are at least as follows:

[0095] The present application constructs a dynamic power consumption model by deeply analyzing the data collected by smart meters, smart home appliances, sensors and other Internet of Things devices, and the user's historical consumption behavior. Through the application of machine learning algorithms, the platform can analyze the user's power consumption data in real time, accurately predict future power demand, and customize personalized power product recommendations for different users. For example, when it is predicted that a user will use a large amount of power in the next few days, the system will push a time period discount electricity package in advance, or recommend energy-saving equipment to help the user reduce power consumption costs during peak periods. This effectively overcomes the limitations of static pricing and unified recommendations in traditional power marketing.

[0096] In combination with the user's power consumption time regularity, the present application proposes a time-aware pricing mechanism. The system can predict the user's future power demand based on their power consumption behavior at different times, and dynamically adjust the electricity pricing scheme. For example, for users who use more power during the evening peak period, the system will push a night valley electricity pricing scheme to encourage users to use more power during the power demand valley period, achieving the purpose of balancing the load. This time-based pricing strategy enables power companies to flexibly adjust electricity prices to reduce power market fluctuations, while providing more attractive power packages for users and improving their power consumption experience.

[0097] The present application also combines social recommendation technology to promote interaction between users and promotion of power products through social platforms. Users can not only share their power consumption data and energy-saving tips, but also choose power products based on recommendations from their social circle. This social recommendation not only increases user interaction and engagement, but also improves the credibility and marketing effectiveness of power product recommendations based on trust and evaluation mechanisms between users. Through the use of an integral system and social challenges, the platform can also encourage users to share their energy-saving experiences, thereby promoting the popularization of energy-saving behavior. This innovative point effectively makes up for the shortcomings of traditional power marketing in terms of user engagement and social recommendation, and improves the user's long-term participation and platform stickiness.

[0098] In summary, the application combines intelligent Internet of Things data, time-aware pricing and social recommendations to provide a new power consumption behavior prediction and marketing method, solving the problems of insufficient personalized recommendations, untimely response and low user participation in the prior art. This technical innovation not only improves the accuracy and effectiveness of power company marketing, but also enhances user interaction and energy-saving awareness through social mechanisms, promoting the transformation of the power industry towards intelligence, personalization and socialization. BRIEF DESCRIPTION OF DRAWINGS

[0099] The application is further described by means of the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the application, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of the following drawings.

[0100] Figure 1 An interaction method flowchart for power marketing according to an embodiment of the application.

[0101] Figure 2 An interaction system framework diagram for power marketing according to an embodiment of the application. DETAILED DESCRIPTION

[0102] The embodiments of the application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are only used to explain the application, and cannot be understood as a limitation on the application.

[0103] As shown in Figure 1 , an interaction method for power marketing according to an embodiment of the application, the method comprises the following steps:

[0104] S1, collecting power consumption data of users through Internet of Things devices, and preprocessing the data to obtain preprocessed power consumption data, and introducing user social behavior as user behavior characteristics, and the diversity of household devices and the change of environmental factors as external environmental factors; wherein the power consumption data includes power consumption, device status and time stamp.

[0105] Specifically, the collection of power consumption data is obtained in real time through multiple data sources such as smart meters, environmental sensors and device status sensors. Each data point is composed of the following core variables:

[0106] Power consumption (E t ): the total amount of user power consumption at time point t, unit: kilowatt-hour (kWh).

[0107] Device status (S t) : Record the on-off state of user's home appliances (such as air conditioner, refrigerator, washing machine, etc.) at time point t, binary (0 represents off, 1 represents on).

[0108] Timestamp (t) : Time point of collecting data, usually in minutes or hours.

[0109] Further, the collected data not only includes power consumption, but also includes appliance usage state, environmental variables (such as temperature, humidity) and other factors that affect power consumption. Through these data, the invention can fully understand the user's power consumption behavior.

[0110] Further, due to the inevitable existence of missing values and noise in the data collection process, the invention designs an adaptive interpolation method to handle missing data. Unlike the common mean or linear interpolation method, the invention uses an interpolation method based on user historical consumption patterns. The invention assumes that the user's power consumption pattern in a certain time period has certain regularity, so it can predict missing values through historical data.

[0111] For example, if the consumption data is missing at a certain time t, the invention will consider the weighted average of the last few historical data points (t-1, t-2, …):

[0112]

[0113] Where, α i is the weighted coefficient, and the specific calculation method takes into account the time difference and the change of user behavior pattern. The design of the weighted coefficient not only depends on the time proximity of historical data, but also combines with the device usage pattern, for example, the frequency of air conditioner use is higher at night, and the weight coefficient will be adjusted accordingly.

[0114] Further, due to the large difference in power consumption of different power devices (such as air conditioner, refrigerator, etc.), standardization is an important step to ensure data consistency and fairness. In order to avoid the disproportionate influence of high-power devices on the model, the invention standardizes the power consumption of various devices.

[0115] A standardization method based on device category is adopted. In the standardization process, the invention calculates the mean and standard deviation of each device's historical power consumption data, and then standardizes the current power consumption data according to the following formula:

[0116]

[0117] Where, μ Et is the mean of the historical power consumption data of device E t , σ Et is the standard deviation, is the standardized power consumption data. This standardization method can eliminate the influence of power differences between devices, so that the subsequent behavior prediction model can more fairly handle the power demand of different devices.

[0118] Further, the power consumption data has obvious time series, that is, the power consumption mode is significantly affected by time factors. In this step, the data is processed in multiple time scales to capture the power consumption changes at different levels such as within a day, within a week, and during holidays.

[0119] Within-day period (D t ): The power consumption behavior of each day is affected by the morning peak, evening peak, and other time periods.

[0120] Within-week period (W t ): Different working days and weekend power consumption modes may be different.

[0121] Holiday (H t ): Holidays or special events will affect the user's power consumption mode.

[0122] Further, to handle these time series changes, the present invention uses a "time series feature nested weighted average" based technique. This method takes into account the characteristics of each time period and adjusts the influence of historical data on the current time through different weights:

[0123]

[0124] where, is the time series power consumption data, w i is the weight adjusted according to the time period (such as working day, holiday, etc.), E t-i is the consumption data of the historical period.

[0125] This method can ensure that the time series of power consumption is fully considered in the model, especially the ability to identify special situations such as within-day peak, weekend, and holiday.

[0126] Further, in addition to standardization and time series processing, the present invention also designs the expansion of user behavior features. The introduction of these features can help the present invention to more accurately capture the user's power consumption rules and provide additional context information for subsequent demand prediction. These features include:

[0127] Diversity of household devices: The switching frequency and duration of different household devices (such as air conditioners, washing machines, etc.).

[0128] Changes in environmental factors: such as the influence of indoor temperature and humidity on the use of air conditioning and heating devices.

[0129] User social behavior: For example, the user's sharing of power consumption behavior on social platforms may reflect certain specific power consumption patterns.

[0130] These features, by fusing with basic power consumption data, generate a comprehensive feature set to provide more comprehensive information.

[0131] S2, according to the preprocessed data, a power consumption behavior prediction model is constructed to predict the future power demand of the user and respond to sudden changes.

[0132] Specifically, in order to capture the time sequence characteristics of the user's power consumption behavior, the present application designs a time sequence convolutional neural network (TCN), which extracts high-order features from the time sequence data of power consumption through convolutional layers, and identifies long-term dependencies. In this process, the convolutional layer and the pooling layer work together to extract the potential rules of the user's power consumption. The fully connected layer further fuses these time sequence features with user behavior features (B t ) and environmental factors (H t ) to make power demand prediction.

[0133] Input data: B t , H t

[0134] Output prediction: (Future power demand of the user)

[0135] The output of the network is the predicted value of the future power demand and the user behavior prediction related to the power demand, which will be further processed in the subsequent multi-task learning stage.

[0136] Further, the present model adopts a multi-task learning framework, which optimizes two targets: power demand prediction and user behavior prediction. The loss function is designed as a weighted combination of the mean square error (MSE) loss of power demand prediction and the cross-entropy loss of user behavior classification.

[0137] For the power demand prediction regression task, the loss function is:

[0138]

[0139] Where, is the actual power demand, is the predicted power demand.

[0140] For the user behavior classification task, the loss function is:

[0141]

[0142] Where, y iactual label of user behavior, predicted behavior probability.

[0143] The final loss function is:

[0144]

[0145] where λ1 and λ2 are the weight coefficients between tasks.

[0146] Further, in order to improve the generalization ability of the model and avoid overfitting, the present application introduces an adaptive regularization mechanism. This mechanism dynamically adjusts the strength of the regularization term based on the changes in user behavior data, avoiding excessive constraints on users with large changes in behavior patterns. In particular, the strength of the regularization term depends on the volatility of the behavior feature B t , using the following formula:

[0147]

[0148] where, is the L2 norm of the user behavior feature, and α t is the regularization strength at the current time. β1 and β2 are adjustment parameters. In this way, when the user behavior pattern changes dramatically, the model can adapt to these changes while avoiding over-regularization.

[0149] Further, the present application uses the Adam optimizer to train the model, combines the backpropagation algorithm to minimize the loss function , and adopts a learning rate decay strategy to improve the stability of the training process. To prevent overfitting, the present application combines cross-validation with early stopping techniques to dynamically adjust the model's hyperparameters, including convolution kernel size and pooling layer parameters. Through multiple rounds of training, the model will gradually converge to an optimal state.

[0150] Further, after training is complete, the model will output the user's future power demand prediction and the probability of user behavior change To evaluate the performance of the model, the present application uses RMSE (Root Mean Square Error) and AUC (Area Under the Curve) two indicators to evaluate the accuracy of power demand prediction and the effectiveness of user behavior prediction, respectively.

[0151] For power demand prediction, the RMSE calculation formula is:

[0152]

[0153] For user behavior prediction, the AUC calculation formula is:

[0154]

[0155] Where TPR is the true positive rate, and FPR is the false positive rate. The closer the AUC value is to 1, the better the model performs in the classification task.

[0156] Further, the innovative design of this step fully combines the time sequence convolutional neural network (TCN) and the multi-task learning framework, effectively solving the multi-objective optimization problem of power demand prediction and user behavior prediction. In particular, the adaptive regularization mechanism introduced enables the model to adjust the regularization strength according to the dynamic changes in user behavior, avoiding overfitting and improving the model's ability to adapt to new changes. This innovation effectively improves the prediction accuracy and generalization ability of the model, meeting the complex timing and behavior fluctuations in power consumption prediction. Through the above steps, the model not only accurately predicts power demand, but also provides reliable predictions under complex user behavior and external environmental changes, thereby providing strong support for power demand response and personalized power marketing strategies.

[0157] S3, constructing a dynamic power pricing model according to the prediction results, adjusting the power pricing according to the actual demand of the user and the change of the power grid load.

[0158] Specifically, to solve the imbalance between supply and demand in the power market, the present application designs a dynamic power pricing mechanism that combines the predicted demand and behavior characteristics of users, and can flexibly adjust power supply according to real-time demand and price. The power pricing is set as P t , which consists of two parts:

[0159] Basic power price P base : This is the basic pricing based on the current market power cost and long-term trend prediction.

[0160] Dynamic adjustment factor ΔP t : Based on the user's power demand prediction and behavior prediction adjusted price fluctuation factor.

[0161] The specific dynamic power pricing formula is:

[0162] P t = P base + ΔP t

[0163] Where ΔP t is the dynamic adjustment part based on demand and behavior prediction, and its calculation method is as follows:

[0164]

[0165] Where γ1 and γ2 are adjustment coefficients, and E avg is the average power demand in a certain period.

[0166] where, is the predicted power demand of the user at future time, derived from the prediction results in step 2. avg is the average power demand of the entire user group, used to determine whether the current demand is beyond normal. is the user behavior prediction, representing the user's sensitivity or response to price changes, used to dynamically adjust pricing strategies. γ1 and γ2 are adjustment coefficients, controlling the influence of demand and behavior on pricing.

[0167] Further, to ensure that power pricing not only considers user demand, but also takes into account system stability and fairness, the model designs multiple constraints, including:

[0168] Supply-demand balance constraint: To avoid excessive fluctuations in power supply and demand, power pricing needs to maintain market supply and demand balance, that is:

[0169]

[0170] where, S t is the power supply at time t, ensuring that pricing matches demand.

[0171] Fairness constraint: To ensure fairness among users, set the price fluctuation range not to exceed a certain proportion:

[0172] ΔP t ≤ α·P base

[0173] where, α is the preset maximum price fluctuation ratio, used to prevent excessive price fluctuations affecting low-income users.

[0174] Further, the model uses a reinforcement learning-based optimization algorithm to dynamically adjust the price strategy. Specifically, a deep Q-learning (DQN) algorithm based on time series is designed, which can optimize the power pricing strategy under the constraints of power supply and demand and user behavior. Through this algorithm, the model can adjust the power price in real time at each time step, ensuring that users maximize utility while the overall efficiency of the power system is also optimized. The Q-value function update is:

[0175]

[0176] where, s t is the current state, a t is the selected pricing strategy, r t+1 is the actual revenue (such as power demand matching degree), and γ is the discount factor.

[0177] Further, during the training process, the model adjusts the adjustment coefficients such as γ1 and γ2 based on historical power demand data, user behavior data, and supply-demand balance constraints through reinforcement learning algorithms to achieve the goal of optimizing power pricing. Each time the model is updated, it evaluates the impact of power price adjustments on market stability, user satisfaction, and other aspects, and adjusts the strategy in real time to ensure that the model converges to the optimal pricing scheme.

[0178] Further, to evaluate the effectiveness of the power pricing model, two types of evaluation indicators are used:

[0179] Supply-demand matching degree: Calculate the difference between power demand and supply at each time:

[0180]

[0181] User satisfaction: Based on the user's response behavior, calculate the total consumption of users under the adjusted pricing, and analyze their acceptance of the pricing strategy.

[0182] Through comprehensive evaluation of these indicators, the model will adjust the pricing strategy based on feedback to make it more in line with actual demand.

[0183] Further, the innovative design of this step solves the pricing challenges brought by power demand prediction and user behavior uncertainty through the combination of dynamic power pricing mechanism and reinforcement learning. In particular, the adaptive pricing factor ΔP t can dynamically adjust the price according to real-time power demand and user behavior, avoiding the lagging response to demand fluctuations in traditional pricing methods. In addition, the design of constraints and optimization objectives ensures that the pricing mechanism can effectively balance market supply and demand while considering system fairness.

[0184] This dynamic power pricing model not only accurately matches user demand, but also improves the overall economic efficiency of the power system, reducing the risk of power supply surplus or shortage, while providing flexible and adjustable pricing strategies for power suppliers.

[0185] S4, real-time data flow monitoring market conditions, combined with the trained power consumption behavior prediction model and dynamic power pricing model, timely discovery of potential abnormalities, if found, intelligent early warning.

[0186] Specifically, in the practical application of the power pricing model, real-time monitoring and intelligent early warning system plays a crucial role, especially when facing uncertain factors such as power demand fluctuations, supply fluctuations, etc., it can effectively control and respond to risks. This step aims to monitor market conditions through real-time data flow, combined with the trained power consumption behavior prediction model and dynamic pricing model, to timely discover potential abnormalities (such as power demand surge, supply shortage, market price volatility, etc.), and perform intelligent early warning.

[0187] Further, at the beginning of this step, the system first acquires important operational data from various power users and the power grid through real-time data interfaces, including but not limited to:

[0188] Real-time power demand (E t ): represents the current power consumption, including power consumption at the user end (home, industry, etc.).

[0189] Power supply (S t ): represents the current supply capacity of the power grid, including power generation capacity of power plants and available power transmission capacity.

[0190] Power market price (P t ): reflects the current price fluctuations in the power market, dynamically adjusted based on supply and demand relationships.

[0191] Historical user behavior data (U t ): includes periodic changes and seasonal fluctuations in power consumption, used in conjunction with demand forecasting.

[0192] After cleaning and standardizing the real-time data stream, the system inputs these data into subsequent models to ensure that the models can make intelligent decisions based on the latest information.

[0193] Further, based on the output results of the power consumption behavior prediction model established in the previous step, the invention needs to compare the current power demand and supply to identify potential abnormalities. The system identifies abnormalities by calculating the deviation between the actual demand and the predicted demand:

[0194]

[0195] where E t represents the actual power demand, is the expected demand based on historical data and prediction models. If the deviation ΔE t exceeds the set threshold (such as a multiple of the standard deviation), the system determines that there is a demand anomaly.

[0196] Similarly, monitoring the power supply is also crucial. If the supply (S t ) differs significantly from the model-predicted supply capacity ( ), the system will trigger a supply anomaly alert:

[0197]

[0198] By calculating these deviations in real time, the system can promptly detect deviations in power supply and demand, and then initiate a warning mechanism.

[0199] It can be understood that when the deviation value ΔE t or ΔS t exceeds the preset threshold, the system will automatically trigger an intelligent early warning. To more accurately assess the early warning trigger condition, a weighted threshold parameter can be introduced, taking into account the sensitivity of different time periods and market volatility. For example, the system can dynamically adjust the thresholds δ E and δ S according to the volatility of historical data to improve the accuracy of early warning.

[0200] When generating an early warning, the system's logic is as follows:

[0201] Power demand anomaly early warning: when |ΔE t |>δ E , indicating that the power demand deviates significantly from the expected value, the system triggers a demand early warning.

[0202] Power supply anomaly early warning: when |ΔS t |>δ S , indicating that the power supply is significantly lower than expected, the system triggers a supply early warning.

[0203] where δ E and δ S are adaptive thresholds learned based on historical data, which can be dynamically adjusted according to different situations.

[0204] Further, to respond to changes in power supply and demand, the real-time monitoring system will dynamically adjust the pricing of the power market. Based on real-time feedback, the system will optimize pricing strategies according to power demand and supply conditions, adjusting market electricity prices (P t ) to achieve supply and demand balance. At this time, the optimization goal not only includes error minimization, but also needs to consider the impact of price fluctuations, and design a dynamic pricing model to adjust electricity prices.

[0205] The invention handles the task of electricity price adjustment and supply and demand balance through weighted optimization of the objective function. The form of the objective function is as follows:

[0206]

[0207] where E t and S t are power demand and supply, P t is the market electricity price, is the electricity price predicted by the model, and λ is the weight factor balancing supply and demand and price fluctuations. The goal of optimization is to minimize the difference between actual power demand and supply, and limit price fluctuations within an acceptable range.

[0208] Further, after the early warning system issues an alert, the feedback mechanism plays a crucial role. The system responds quickly based on real-time data and makes real-time adjustments to the price, supply, and demand of the electricity market. After each feedback, the predicted values of electricity demand and supply and are dynamically updated based on new data. This real-time feedback mechanism ensures that the model can adapt to changing market conditions, avoiding power supply and demand imbalances or excessive market fluctuations.

[0209] Through this closed-loop adjustment mechanism, the pricing of the electricity market can more flexibly respond to emergencies and abnormal fluctuations, ensuring stable operation of the system.

[0210] This step is closely related to the aforementioned prediction model, which provides accurate demand forecasts, and the dynamic pricing model, which optimizes electricity pricing based on changes in demand and supply. This dynamic adjustment mechanism is the core of intelligent dispatching of the electricity market, effectively addressing the limitations of traditional electricity pricing models in real-time response.

[0211] S5, combined with historical data and real-time data of electricity marketing, the optimal electricity package for each user is recommended through optimization algorithm, and real-time adjustment is made to ensure that the most suitable electricity scheme is provided for users under different conditions.

[0212] Specifically, personalized electricity package recommendation and real-time optimization are crucial in the system. By combining the electricity demand prediction, market electricity price fluctuations, and user behavior patterns from the previous steps, this step uses optimization algorithms to recommend the optimal electricity package for each user and makes real-time adjustments to ensure that the most suitable electricity scheme is provided for users under different conditions. The core goal of this step is to optimize the allocation of electricity resources, reduce user electricity costs, and improve the efficiency and response speed of the power grid.

[0213] Further, the input of this step comes from the output of the previous steps, especially the electricity demand prediction ( ), real-time electricity demand (E t ), electricity market price (P t ), and user electricity usage behavior patterns (A t ) from the dynamic electricity pricing model. Specifically, it includes:

[0214] User electricity consumption history (U t ): Historical electricity consumption data of users, reflecting the periodicity, seasonality, and other characteristics of user electricity consumption.

[0215] Real-time electricity demand (E t ): User current electricity demand data collected from the monitoring system.

[0216] Electricity market price (Pt ) Real-time electricity price fluctuation data obtained from the market.

[0217] Electricity package price prediction (P rec (t)) Recommended electricity price given by the system according to real-time demand and electricity price prediction results.

[0218] User behavior pattern analysis (A t ) Extract user behavior patterns (e.g., whether to prefer peak or off-peak consumption) through analysis of user historical electricity consumption data.

[0219] Further, in order to optimize the recommendation of electricity packages, the goal is to design an optimal electricity package recommendation model according to the user's electricity demand, market electricity price fluctuations, and behavior patterns. To this end, the present application first quantifies the effect of the recommended package by the following optimization objective function:

[0220]

[0221] Where E t is the actual electricity demand of the user at time period t, is the user electricity demand prediction based on the prediction model, P rec (t) is the recommended electricity package price (recommended by the system according to real-time demand and electricity price fluctuations), P t is the actual electricity market price, A i (t) is the user's behavior pattern at time period t (e.g., preferring to use electricity during peak or off-peak hours), A rec (t) is the recommended user behavior pattern, and γ and λ are adjustment coefficients used to control the balance between electricity price fluctuations and behavior patterns. Through the above optimization function, the present application makes the recommended electricity package as close as possible to the actual demand of the user, while considering electricity price fluctuations and user behavior preferences, thereby achieving personalized recommendation.

[0222] Further, after the user accepts the recommended package, the system still needs to dynamically adjust the electricity package to respond to real-time changes in electricity demand, market electricity prices, and user behavior patterns. At this time, the present application optimizes the electricity package through real-time monitoring and prediction models, with the goal of making the recommended package more adaptable.

[0223] Considering electricity price fluctuations and user demand changes, the present application introduces a dynamic adjustment objective function:

[0224]

[0225] Where U t is the actual electricity consumption behavior history data of the user at time period t, U recThe user's electricity consumption history based on the prediction results (i.e., the recommended power usage pattern) is recommended by the system. δ, μ, and θ are dynamic adjustment coefficients used to balance different optimization objectives. The additional term θ in this objective function introduces the influence of the user's historical electricity consumption data, aiming to adjust the recommended scheme to better adapt to the user's electricity habits. Through this adjustment, the system can more accurately respond to market changes and user demand fluctuations, ensuring that users always enjoy the most suitable power package at different time periods.

[0226] Further, to ensure the continuous optimization and adaptive ability of the power package recommendation system, the present application adds an intelligent feedback mechanism. When the user actually uses the power package, their electricity consumption behavior will be collected in real time through the feedback system, and the system will automatically evaluate the effectiveness of the recommended package and adjust the subsequent recommendation strategy accordingly.

[0227] This feedback mechanism adjusts itself through a reinforcement learning model, optimizing the recommendation process based on historical feedback data. The optimization objective function is:

[0228]

[0229] Where α, β, and γ are dynamic adjustment coefficients used to control the effect of feedback results in the optimization process.

[0230] The core of this feedback mechanism is to fine-tune the power package through real-time user feedback, thereby achieving continuous optimization of personalized packages and the adaptive ability of the system.

[0231] Further, to enable users to better understand and manage their power packages, the system provides an easy-to-interact control interface. Users can view their power packages, real-time electricity demand, current electricity prices, and estimated costs through the interface. If the user believes that the existing package is not suitable, they can also manually adjust their power package settings. User behavior feedback will directly affect the system's optimization process, helping the system more accurately recommend suitable power packages.

[0232] Through personalized recommendation and real-time optimization, the system provides users with an economic and flexible electricity solution. These optimization measures can effectively reduce power waste and improve user satisfaction for power suppliers, while helping power companies achieve more refined power management.

[0233] As Figure 2 shown, in another embodiment of the present application, an interactive system for power marketing is provided, which includes:

[0234] The power consumption data acquisition unit 301 is configured to acquire power consumption data of a user through an Internet of Things device, pre-process the data, obtain pre-processed power consumption data, introduce user social behavior as a user behavior feature, and collect the diversity of household devices and changes in environmental factors as external environmental factors; wherein the power consumption data includes power consumption, device status, and a timestamp;

[0235] The power demand prediction unit 302 is configured to construct a power consumption behavior prediction model according to the pre-processed data to predict the future power demand of the user and respond to sudden changes; wherein the power consumption behavior prediction model adopts a time series convolutional neural network, extracts high-order features from time series data of power consumption through a convolutional layer, identifies long-term dependencies, and in the training process, the convolutional layer and the pooling layer jointly act to extract the potential law of user power consumption, and the fully connected layer fuses the time series features with the user behavior features and the external environmental factors to perform power demand prediction.

[0236] The power consumption behavior prediction model is designed based on the time series convolutional neural network as a multi-task learning framework, which optimizes two targets: power demand prediction and user behavior prediction, wherein the loss function of the power consumption behavior prediction model is designed as a weighted combination of the mean square error loss of power demand prediction and the cross-entropy loss of user behavior classification.

[0237] The dynamic pricing unit 303 is configured to construct a dynamic power pricing model according to the prediction results, and adjust the power pricing according to the actual demand of the user and the change of the power grid load; wherein the dynamic power pricing model is constructed as follows:

[0238] The dynamic power pricing mechanism is designed in combination with power demand prediction and user behavior prediction, and the power supply is flexibly adjusted according to real-time demand and price;

[0239] A plurality of constraint conditions are designed, including supply and demand balance constraints and fairness constraints.

[0240] A deep Q learning algorithm based on reinforcement learning is used to dynamically adjust the pricing strategy; wherein the deep Q learning algorithm is represented as:

[0241]

[0242] wherein s t is the current state, a t is the selected pricing strategy, r t+1 is the actual return, and γ is the discount factor.

[0243] The power consumption data anomaly alarm unit 304 is used for monitoring market conditions in real time data flow, combining the trained power consumption behavior prediction model and the dynamic power pricing model, and discovering potential anomalies in time, and if an anomaly is found, intelligent early warning is performed.

[0244] The personalized recommendation unit 305 is used for combining historical data and real-time data of power marketing, recommending the optimal power package for each user through an optimization algorithm, and performing real-time adjustment, so as to ensure that the most suitable power scheme is provided for the user under different conditions.

[0245] It should be noted that the above-described workflow is merely illustrative and does not limit the protection scope of the present application. In actual application, a person skilled in the art can select part or all of the above-described workflow to achieve the purpose of the present embodiment, and the present application is not limited in this regard.

[0246] In addition, technical details not described in detail in the present embodiment can be found in the parameter operation method provided by any embodiment of the present application, and will not be described here.

[0247] It should be noted that in this document, the terms "comprise", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or system including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or system. Without more limitations, the element defined by the statement "comprises a" does not exclude the presence of another identical element in the process, method, article or system including the element.

[0248] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages or disadvantages of the embodiments.

[0249] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be realized by means of software and necessary general hardware platforms, of course, they can also be realized by hardware, but in many cases, the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, an optical disk), and includes a plurality of instructions for making a terminal device (which can be a mobile phone, a computer, a server, an air conditioner, or a network device) execute the methods described in the various embodiments of the present application.

[0250] The above merely describes the preferred embodiments of the present application, and is not intended to limit the patent scope of the present application, and any equivalent structure or equivalent process conversion, or direct or indirect application in other related technical fields, which are made by using the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. An interactive method for power marketing, characterized by, The method comprises: S1, collecting the power consumption data of the user through the Internet of Things device, preprocessing the data to obtain preprocessed power consumption data, introducing the user's social behavior as the user behavior feature, collecting the diversity of the home device and the change of the environmental factor as the external environment factor; wherein the power consumption data includes power consumption, device status and timestamp; S2, constructing a power consumption behavior prediction model according to the preprocessed data to predict the future power demand of the user and respond to sudden changes; wherein the power consumption behavior prediction model adopts a time series convolutional neural network, extracts high-order features from the time series data of power consumption through a convolutional layer, identifies long-term dependencies, and in the training process, the convolutional layer and the pooling layer jointly act to extract the potential law of the user's power consumption, and the fully connected layer fuses the time series features with the user behavior features and the external environment factors to predict the power demand; Wherein, the power consumption behavior prediction model based on the time series convolutional neural network is designed as a multi-task learning framework, which optimizes two targets: power demand prediction and user behavior prediction, wherein the loss function of the power consumption behavior prediction model is designed as a weighted combination of the mean square error loss of power demand prediction and the cross-entropy loss of user behavior classification; S3, constructing a dynamic power pricing model according to the prediction result, adjusting the power pricing according to the actual demand of the user and the change of the power grid load; wherein the dynamic power pricing model is constructed as follows: Design a dynamic power pricing mechanism combining power demand prediction and user behavior prediction, and adjust the power supply flexibly according to real-time demand and price; Design multiple constraint conditions, including supply and demand balance constraint and fairness constraint; Use a deep Q learning algorithm based on reinforcement learning to dynamically adjust the price strategy; wherein the deep Q learning algorithm is represented as: Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )] where s t is the current state, a t is the chosen pricing strategy, r t+1 is the actual reward, and γ is the discount factor. S4, real-time data flow monitors market conditions, combines the trained power consumption behavior prediction model and dynamic power pricing model, and timely discovers potential abnormalities, and if abnormalities are found, intelligent early warning is performed; S5, combining the historical data and real-time data of power marketing, recommending the optimal power package for each user through an optimization algorithm, and adjusting it in real time to ensure that the most suitable power scheme is provided for the user under different conditions.

2. The interactive method for power marketing according to claim 1, wherein, The preprocessing comprises: Interpolation based on user historical consumption patterns: assuming that the user's power consumption pattern in a certain time period has certain regularity, the missing values are predicted through historical data; Standardization method based on device category: in the standardization process, the mean and standard deviation of each device are calculated according to its historical power consumption data, and then the power consumption data at the current time is standardized; Multi-time scale processing: for intra-day period, intra-week period and holidays, the algorithm based on time series feature nested weighted average is used to adjust the influence of historical data on the current time through different weights to obtain preprocessed power consumption data; Introduce data feature expansion, collect user social behavior as user behavior feature, and collect the diversity of home device and the change of environmental factor as external environment factor.

3. The interactive method for power marketing according to claim 1, wherein, For the power demand prediction regression task, the loss function is: wherein, is the actual power demand at time t+Δt, is the predicted power demand at time t+Δt, N is the total number of samples, and i is the current user i. For the user behavior classification task, the loss function is: where y i is the actual label of the user behavior of user i, is the predicted probability of the behavior of user i; Finally, the loss function of the power consumption behavior prediction model is: where λ1 and λ2 are the weight coefficients between tasks.

4. The interactive method for power marketing according to claim 3, wherein, In order to improve the generalization ability of the power consumption behavior prediction model and avoid overfitting, an adaptive regularization mechanism is introduced, which dynamically adjusts the strength of the regularization term based on the changes in user behavior data, avoiding imposing the greatest constraint on users with the greatest change in behavior pattern, denoted as: wherein, is the L2 norm of the user behavior feature, and a t is the regularization strength at the current time; and β1 and β2 are adjustment parameters. The model is trained by using an Adam optimizer and a back propagation algorithm to minimize a loss function And a learning rate decay strategy is used to improve the stability of the training process. After training, the model will output the user's future power demand prediction and the probability of user behavior change, and then use RMSE and AUC two indicators to evaluate the accuracy of power demand prediction and the effectiveness of user behavior prediction respectively; For power demand prediction, the RMSE calculation formula is: wherein, is the actual power demand at time t+Δt, is the predicted power demand at time t+Δt, N is the total number of samples, and i is the current user i; For user behavior prediction, the AUC calculation formula is: Where TPR is the true positive rate, FPR is the false positive rate; The AUC value closest to 1 indicates that the model performs best in the classification task.

5. The interactive method for power marketing according to claim 1, wherein, The dynamic power pricing mechanism includes a basic power price and a dynamic adjustment factor, the basic power price is a basic pricing based on current market power cost and long-term trend prediction, the dynamic adjustment factor is a price fluctuation factor adjusted based on power demand prediction and user behavior prediction, the dynamic power pricing mechanism is denoted as: P t = P base + ΔP t where P t is the market electricity price at time t, P base is the base electricity price, ΔP t is the dynamic adjustment part of the user behavior prediction based on the electricity demand forecast at time t, which is calculated as follows: wherein, is the predicted power demand at time t+Δt, γ1 and γ2 are adjustment coefficients, is the predicted behavior probability, E avg is the average power demand over a certain period of time; The supply and demand balance constraint is used to maintain market supply and demand balance, denoted as: where S t is the amount of power supplied at time t, ensuring that pricing matches demand; T is the total duration; The fairness constraint is used to set the price fluctuation range not to exceed a certain proportion, denoted as: wherein, is a preset maximum price fluctuation ratio for preventing excessive price fluctuations from affecting low-income users.

6. The interactive method for power marketing according to claim 5, wherein, The S3 also includes: The effect of the dynamic electricity pricing model is evaluated using two types of evaluation indexes, including supply-demand matching degree and user satisfaction; the supply-demand matching degree Mismatch t The difference between the electricity demand and supply at each time t is represented as: The user satisfaction is based on the user's response behavior, calculating the total consumption of the user under the adjusted pricing, and analyzing its acceptance of the pricing strategy.

7. The interactive method for power marketing according to claim 1, wherein, The S4 specifically includes: S401, obtaining operation data from each power user and power grid through a real-time data interface; wherein, the operation data includes real-time power demand, power supply, power market price and historical user behavior data; S402, monitoring the current power demand and supply and comparing, identifying potential abnormalities; S403, based on real-time feedback, setting optimization target optimization pricing strategy according to power demand and supply, adjusting market price, realizing supply and demand balance; wherein, the objective function of the optimization target is denoted as: where E t and S t are the power demand and supply at time t, respectively, P t is the market price at time t, is the price predicted by the model at time t, and λ is a weight factor balancing the supply-demand and price fluctuations.

8. The interactive method for power marketing according to claim 7, wherein, The S402 specifically includes: Abnormalities are identified by calculating the deviation between the current actual demand and the predicted demand, denoted as: where E t represents the actual power demand at time t, is the expected demand at time t based on historical data and a prediction model; if the deviation ΔE t exceeds a set threshold, the demand is determined to be abnormal. Likewise, if the supply S at time t t The supply capacity at time t If the difference exceeds a threshold, a supply anomaly alert will be triggered: When the deviation value ΔE t or ΔS t exceeds a preset threshold, an intelligent early warning is automatically triggered.

9. The interactive method for power marketing according to claim 1, wherein, The S5 specifically includes: S501, designing a personalized power package recommendation model, denoted as: where E t is the actual power demand of the user at time t, is the expected demand at time t based on historical data and prediction models, P rec (t) is the recommended electricity package price at time t, t is the market electricity price at time t, A i (t) is the behavior pattern of the user at time t, A rec (t) is the recommended user behavior pattern, γ and λ are adjustment coefficients for controlling the balance between electricity price fluctuations and behavior patterns; N is the total number of samples, and i is the current user i. S502, when the user accepts the recommended package, the power consumption behavior prediction model is adjusted to optimize the power package, making the recommended package more adaptive; wherein, the objective function of adjusting the power consumption behavior prediction model to optimize the power package is denoted as: where U t is the actual user consumption history at time t, U rec is the user consumption history recommended by the system based on the prediction result, and δ, μ and θ are dynamic adjustment coefficients used to balance different optimization objectives. S503, when the user actually uses the power package, the electricity consumption behavior will be collected in real time through feedback, and the effect of the recommended package is automatically evaluated, and the subsequent recommendation strategy is adjusted accordingly; wherein, the effect of the recommended package is automatically evaluated through a reinforcement learning model, the recommendation process is optimized based on historical feedback data, and the optimization objective function is: Wherein, ω, β and γ are dynamic adjustment coefficients, used to control the effect of feedback results in the optimization process.

10. An interactive system for power marketing, characterized by, The system comprises: An electricity consumption data acquisition unit for collecting electricity consumption data of users through Internet of Things devices, pre-processing the data to obtain pre-processed electricity consumption data, introducing user social behavior as user behavior characteristics, collecting the diversity of household devices and changes in environmental factors as external environmental factors; wherein the electricity consumption data includes electricity consumption, device status and timestamp; An electricity demand prediction unit for predicting future electricity demand of users according to the pre-processed data, and responding to sudden changes; wherein the electricity consumption behavior prediction model uses a time series convolutional neural network to extract high-order features from time series data of electricity consumption through a convolutional layer, identify long-term dependencies, and in the training process, the convolutional layer and the pooling layer work together to extract the potential law of user electricity consumption, and the fully connected layer fuses the time series features with user behavior characteristics and external environmental factors for electricity demand prediction; Wherein, the electricity consumption behavior prediction model is designed based on the time series convolutional neural network as a multi-task learning framework, which optimizes two targets: electricity demand prediction and user behavior prediction, wherein the loss function of the electricity consumption behavior prediction model is designed as a weighted combination of the mean square error loss of electricity demand prediction and the cross-entropy loss of user behavior classification; A dynamic pricing unit for constructing a dynamic electricity pricing model according to the prediction results, adjusting electricity pricing according to actual user demand and grid load changes; wherein the dynamic electricity pricing model is constructed as follows: A dynamic electricity pricing mechanism is designed in combination with electricity demand prediction and user behavior prediction, and electricity supply is flexibly adjusted according to real-time demand and price; Multiple constraint conditions are designed, including supply and demand balance constraints and fairness constraints; A deep Q-learning algorithm based on reinforcement learning is used to dynamically adjust the price strategy; wherein the deep Q-learning algorithm is represented as: Q(s t ,a t )←Q(s t ,a t )+α[r t+1 +γmax a′ Q(s t+1 ,a′)-Q(s t ,a t )] where s t is the current state, a t is the chosen pricing policy, r t+1 is the actual reward, and γ is the discount factor. An electricity consumption data anomaly alarm unit for real-time data stream monitoring market conditions, combining the trained electricity consumption behavior prediction model and dynamic electricity pricing model to timely discover potential anomalies, and if anomalies are found, intelligent early warning is performed; An individualized recommendation unit for combining historical data and real-time data of electricity marketing, recommending the optimal electricity package for each user through an optimization algorithm, and making real-time adjustments to ensure that the most suitable electricity scheme is provided for users under different conditions.

Images