Power consumption side fee settlement optimization method based on causal conduction model

Abstract

The invention discloses an electricity consumption measurement fee settlement optimization method based on a causal conduction model. The method comprises the following steps: S1, collecting user electricity consumption behavior data and policy intervention data to form a data set; s2, key user power consumption behavior characteristics are extracted from the user data set through principal component analysis, a clustering algorithm and a random forest algorithm are used for analyzing response modes of users to different policies, and user data sets and policy intervention data sets of different user category matrixes are formed; s3, analyzing the causal relationship between the power consumption behavior of the user and the policy intervention change through Granger causal verification and a structural equation model; and S4, simulating user power consumption behavior responses in different policy intervention scenes by adopting a reinforcement learning algorithm, simulating influences of different policy combinations by adopting a Monte Carlo method, and determining an optimal policy intervention strategy. According to the invention, the problems of complex non-linear relationship and difficult policy intervention effect prediction in the power consumption side fee settlement process under the power marketization background are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system optimization, and in particular relates to an electricity consumption fee settlement optimization method based on a causal conduction model. Background Art

[0002] With the deepening reform of the power market and the increasing demand for power consumption management, the complexity of electricity bill settlement is increasing. Existing methods for bill settlement mainly rely on fixed rules or traditional time series analysis, ignoring the complex interactions between various external factors and user behavior, and lacking causal analysis of policy responses and changes in behavior patterns, resulting in insufficient generalization capabilities in a market environment and inability to cover all application scenarios. In this context, how to effectively optimize the bill settlement process and improve the accuracy and flexibility of settlement has become a problem that the power industry needs to solve urgently.

[0003] The complexity of quantity-fee settlement is reflected in many aspects: (1) The complexity of the market-based pricing mechanism: With the gradual opening of the electricity market, electricity prices are affected by many factors, including policy adjustments, changes in supply and demand, and socio-economic conditions. This has resulted in the inability of the traditional fixed electricity price model to adapt to the changing market environment. (2) Diversification of user behavior patterns: The electricity consumption behaviors of different user groups show different regularities. Traditional methods are difficult to capture the potential driving factors of these behavioral changes, especially the impact of external factors such as policy changes, economic fluctuations, and social communication on user electricity consumption patterns. (3) Improved settlement accuracy requirements: Grid companies and market participants expect more accurate quantity-fee forecasts to improve the efficiency of electricity trading and settlement, but the existing settlement model cannot provide effective means of dynamic adjustment of quantity and price.

[0004] The Chinese patent with publication number CN116703437A discloses a method for analyzing the impact of time-of-use electricity price policy formulation on power supply security; it includes: step 1, adjusting the user peak and valley time-of-use electricity price mechanism according to the original time-of-use electricity price policy; step 2, determining the demand change according to the adjusted time-of-use electricity price policy; step 3, measuring and analyzing the impact of the adjusted time-of-use electricity price policy; step 4, combining the impact of the winter time-of-use electricity price policy on the characteristics of electricity load, and analyzing the power supply security situation in summer. The method described in this invention lacks a dynamic tracking and feedback mechanism after the implementation of the electricity price policy, and fails to adjust the policy in time to cope with changes in actual conditions. Summary of the invention

[0005] The present invention provides an optimization method for electricity consumption fee settlement based on a causal conduction model, aiming to solve the problems of complex nonlinear relationships and difficult-to-predict policy intervention effects in the electricity consumption fee settlement process under the background of power marketization.

[0006] In order to solve the above technical problems, the present invention provides a method for optimizing electricity consumption fee settlement based on a causal conduction model, comprising the following steps:

[0007] S1: Collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form user data sets and policy intervention data sets.

[0008] S2: Extract key user electricity consumption behavior features from the user data set through principal component analysis, use clustering algorithm and random forest algorithm to analyze the user response patterns to different policies, divide the user response patterns into different categories, and form user data sets and policy intervention data sets for different user category matrices.

[0009] S3: The datasets of different user category matrices and policy intervention datasets are converted into time series datasets. The impact of different policy interventions on user electricity consumption behavior is verified through Granger causality, and the structural equation model is further used to analyze the causal relationship between user electricity consumption behavior and policy intervention changes.

[0010] S4: Use reinforcement learning algorithm to simulate user electricity consumption behavior responses under different policy intervention scenarios, build a comprehensive causal conduction model based on causal relationship to design the reward function of reinforcement learning, obtain the optimal policy intervention strategy combination, and use the Monte Carlo method to simulate the impact of different policy intervention strategy combinations on user electricity consumption behavior. Determine the optimal policy intervention strategy based on the simulation results of the reinforcement learning method and the Monte Carlo method.

[0011] Preferably, the step S1 specifically comprises:

[0012] Collect user electricity usage behavior data including historical electricity consumption, electricity charges, customer service records, and payment channels, as well as policy intervention data including time-of-use electricity prices, peak and valley periods, tiered electricity prices, demand response policies, and green electricity grid-connected policies.

[0013] The collected data are preprocessed using data cleaning techniques such as outlier detection and missing value filling to form user data sets and policy intervention data sets.

[0014] Preferably, the step S2 is specifically:

[0015] The principal component analysis is used to extract the key user electricity consumption behavior features X from the user data set. Reduced =WX, where X Reduced is the data matrix after dimensionality reduction, R n*d is the user's electricity consumption behavior data, n represents the number of users, and d represents the number of features for each user; W∈R k*d is the dimension reduction matrix, k is the reduced feature dimension, and X is the original user dataset.

[0016] A clustering algorithm is used to classify the user electricity usage behavior features extracted by principal component analysis to obtain the user electricity usage behavior classification results.

[0017] Based on the classification results of user electricity consumption behavior and combined with the policy intervention data set, the random forest algorithm is used to analyze the user's response patterns to different electricity price policies, and the user response patterns of different categories are obtained. The patterns are matched with the clustered user electricity consumption behavior classification results to form user data sets and policy intervention data sets of different user category matrices.

[0018] Preferably, in step S3, the effects of different policy interventions on the user's electricity consumption behavior are verified by Granger causality as follows:

[0019] The user data set is converted into a time series y(t), and the policy intervention data set is converted into a time series x(t). The single causal transmission model is expressed as:

[0020] y(t)=αy(t-1)+βx(t-1)+∈(t)

[0021] Where α is the autoregressive influence coefficient of the user's electricity consumption behavior at time point t-1 on the user's electricity consumption behavior y(t) at the current time point; β is the influence coefficient of the policy intervention variable at time point t-1 on the electricity consumption behavior y(t) at the current time point; ∈(t) is the error term; and t is time.

[0022] According to Granger causality test, if x(t) can significantly predict y(t), then x(t) has a causal relationship with y(t):

[0023] H 0 :β=0,H 1 :β≠0

[0024] In the formula, H 0 Assume that policy intervention is not a causal factor in users’ electricity consumption behavior; H 1 Assume that policy intervention is the causal factor of users' electricity consumption behavior; test H 0 Whether it was rejected.

[0025] Preferably, the causal relationship between the user's electricity consumption behavior and the same policy is analyzed by using a single causal conduction model constructed by the structural equation model:

[0026] y=Λx+∈

[0027] Where y is the user's electricity consumption behavior data vector; x is the policy intervention data variable; Λ is the causal weight matrix, which is estimated by the least squares method.

[0028] Preferably, the step S4 is specifically:

[0029] A comprehensive causal transmission model is constructed based on causal relationships, which can be expressed as:

[0030] y t =f(x(t),y(t-1),z(t),∈(t))

[0031] Where y(t) is the user's electricity consumption behavior at time t, x(t) is the policy intervention, z(t) is the user's environmental factor, and ∈(t) is the error term.

[0032] The Q-learning algorithm in reinforcement learning is used to simulate the user's electricity consumption behavior under different policy intervention scenarios, and the optimal policy intervention strategy combination is output. The state space S is the user's current electricity consumption behavior and policy context, the action space A is the policy adjustment measures, and the reward function R is the change in electricity consumption after the policy is implemented. The reward function of reinforcement learning is designed based on the comprehensive causal conduction model:

[0033] Q(s t ,a t )=Q(s t ,a t )+α[r t +γmaxQ(s t+1 ,a)-Q(s t ,a t )]

[0034] Among them, Q(s t ,a t ) is the state s at time t t ∈S, take action a t ∈A’s value function; r t ∈R is the immediate reward; a is the learning rate; γ is the discount factor; Q(s t+1 ,a t ) is the state s at time t+1 t+1 ∈S, take action a t ∈A’s value function.

[0035] Based on the optimal policy intervention strategy combination output by reinforcement learning, different policy intervention strategy combinations are defined. The Monte Carlo method is used to conduct multiple random experiments under different policy intervention strategy combinations to simulate the effects of different policy intervention strategy combinations, evaluate their long-term impact on the overall user electricity consumption behavior, and determine the optimal policy intervention strategy.

[0036] On the other hand, the present invention provides an electricity consumption fee settlement optimization system based on a causal conduction model, including a data collection and processing module, a user electricity consumption behavior identification module, a causal relationship analysis module and a policy intervention strategy generation module.

[0037] The data collection and processing module is used to collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form user data sets and policy intervention data sets.

[0038] The user electricity consumption behavior identification module is used to extract key user electricity consumption behavior features from the user data set through principal component analysis, use clustering algorithm and random forest algorithm to analyze the user's response pattern to different policies, divide the user response patterns into different categories, and form user data sets and policy intervention data sets for different user category matrices.

[0039] The causal relationship analysis module is used to convert the data sets of different user category matrices and policy intervention data sets into time series data sets, verify the impact of different policy interventions on user electricity consumption behavior through Granger causality, and further use the structural equation model to analyze the causal relationship between user electricity consumption behavior and policy intervention changes.

[0040] The policy intervention strategy generation module is used to use the reinforcement learning algorithm to simulate the user's electricity consumption behavior response under different policy intervention scenarios, build a comprehensive causal conduction model based on causal relationships to design the reinforcement learning reward function, obtain the optimal policy intervention strategy combination, and simulate the impact of different policy intervention strategy combinations on user electricity consumption behavior through the Monte Carlo method. The optimal policy intervention strategy is determined based on the simulation results of the reinforcement learning method and the Monte Carlo method.

[0041] On the other hand, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for optimizing electricity consumption fee settlement based on the causal conduction model as described in any embodiment of the present invention when executing the computer program.

[0042] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for optimizing electricity consumption fee settlement based on a causal conduction model as described in any embodiment of the present invention.

[0043] Compared with the prior art, the present invention has the following technical effects:

[0044] 1. Improved accuracy: Based on the causal transmission model, the present invention can deeply analyze the complex relationship between user behavior and policy factors, capture nonlinear relationships and multi-factor interactions, thereby improving the accuracy of quantity and fee settlement, reducing settlement errors, and ensuring fairness in electricity charges for users.

[0045] 2. Dynamic adaptability: By real-time monitoring of user power consumption behavior and policy changes, the present invention has dynamic adaptability. Compared with the traditional static settlement model, this method can update the causal transmission model in real time according to the new power consumption pattern and policy adjustment, so as to more effectively respond to changes in power policies, optimize the settlement process, and improve the flexibility and response speed of the system.

[0046] 3. Intelligent audit process: The present invention realizes the intelligentization of quantity and fee settlement through the causal transmission model, simplifies the traditional manual audit process, enables the State Grid to automatically identify abnormal electricity consumption behavior, adjust settlement strategies in time, reduce labor costs and error rates, thereby improving overall audit efficiency and accuracy, and promoting the digital transformation of the power industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is an overall flow chart of the electricity consumption fee settlement optimization method based on the causal conduction model described in the present invention. DETAILED DESCRIPTION

[0048] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0049] Embodiment 1

[0050] This embodiment provides an optimization method for electricity consumption-side quantity and price settlement based on a causal conduction model, which can analyze the causal relationship between user electricity consumption behavior and market-based policy response, optimize the electricity quantity and price settlement scheme, solve the complex problem of quantity and price settlement in market-based electricity transactions, and improve the accuracy of quantity and price prediction. Figure 1 As shown, the following steps are included:

[0051] S1: Collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form user data sets and policy intervention data sets.

[0052] As a preferred implementation of this embodiment, step S1 is specifically as follows:

[0053] The collection includes electricity consumption data, including historical electricity consumption, electricity charges, customer service records, payment channels and other user electricity behavior data, as well as policy intervention data including time-of-use electricity prices, peak and valley periods, tiered electricity prices, demand response policies, and green power grid connection policies. Specifically, the user's electricity behavior data such as historical electricity consumption, market electricity prices, and policy responses can be collected through infrastructure such as the Internet of Things technology and smart meters, and data integration can be performed in combination with customer service records, payment channels and other information in the power marketing system. On this basis, distributed storage systems such as Hadoop and NoSQL are used for efficient data management.

[0054] The collected data are preprocessed using data cleaning techniques such as outlier detection and missing value filling to form user data sets and policy intervention data sets to ensure data consistency and integrity.

[0055] S2: Extract key user electricity consumption behavior features from the user data set through principal component analysis, use clustering algorithm and random forest algorithm to analyze the user response patterns to different policies, divide the user response patterns into different categories, and form user data sets and policy intervention data sets for different user category matrices.

[0056] As a preferred implementation of this embodiment, step S2 is specifically as follows:

[0057] The principal component analysis is used to extract key user electricity consumption behavior characteristics (such as electricity consumption, electricity consumption time, consumption frequency, etc.) from the user data set. Reduced =WX, where X Reduced is the data matrix after dimensionality reduction, R n*d is the user's electricity consumption behavior data, n represents the number of users, and d represents the number of features for each user; W∈R k*d is the dimension reduction matrix, k is the reduced feature dimension, and X is the original user dataset.

[0058] A clustering algorithm is used to classify the extracted user electricity usage behavior features to obtain the user electricity usage behavior classification results.

[0059] Specifically, the clustering algorithms include K-means clustering, hierarchical clustering algorithms, etc. The classification results of the clustering algorithms are electricity consumption behavior patterns of different users (such as peak electricity consumption, valley electricity consumption, etc.).

[0060] Furthermore, this embodiment takes K-means clustering as an example, and the formula for clustering user electricity consumption behavior patterns through K-means clustering is:

[0061]

[0062] Where I(·) is the indicator function, which means user x i Belongs to cluster C j , μ j is the cluster center.

[0063] The K-means clustering pseudo-algorithm is:

[0064]

[0065]

[0066] Based on the classification results of user electricity consumption behavior and combined with the policy intervention data set, the random forest algorithm is used to analyze the user's response pattern to different electricity price policies, and the user response patterns of different categories are obtained. They are matched with the clustered user electricity consumption behavior classification results to form user data sets and policy intervention data sets of different user category matrices. The random forest algorithm analyzes the relationship between different user electricity consumption behavior patterns and policy changes, extracts potential influencing factors, and outputs the user's electricity consumption behavior pattern categories, laying the foundation for the establishment of a causal conduction model.

[0067] Furthermore, the random forest algorithm analyzed users' electricity consumption behavior data (such as historical electricity consumption, electricity price fluctuations, electricity consumption time, etc.) and policy intervention data (such as electricity price adjustments, peak and valley electricity prices, demand response policies, etc.), and divided users into three categories: "high sensitivity", "medium sensitivity" and "low sensitivity".

[0068] The classification results of random forests directly provide input information for causal analysis. The classified user response patterns are used for the next step of causal inference to clarify how different sensitive types of users are affected by policy interventions.

[0069] S3: The datasets of different user category matrices and policy intervention datasets are converted into time series datasets. The impact of different policy interventions on user electricity consumption behavior is verified through Granger causality, and the structural equation model is further used to analyze the causal relationship between user electricity consumption behavior and policy intervention changes.

[0070] As a preferred implementation of this embodiment, the influence of different policy interventions on the user's electricity consumption behavior is specifically verified by Granger causality in step S3 as follows:

[0071] The user data set is converted into a time series y(t), and the policy intervention data set is converted into a time series x(t). The single causal transmission model is expressed as:

[0072] y(t)=αy(t-1)+βx(t-1)+∈(t)

[0073] Where α is the autoregressive influence coefficient of the user's electricity consumption behavior at time point t-1 on the user's electricity consumption behavior y(t) at the current time point; β is the influence coefficient of the policy intervention variable at time point t-1 on the electricity consumption behavior y(t) at the current time point; ∈(t) is the error term; and t is time.

[0074] According to Granger causality test, if x(t) can significantly predict y(t), then x(t) has a causal relationship with y(t):

[0075] H 0 :β=0,H 1 :β≠0

[0076] In the formula, H 0 Assume that policy intervention is not a causal factor in users’ electricity consumption behavior; H 1 Assume that policy intervention is the causal factor of users' electricity consumption behavior; test H 0 Whether it was rejected.

[0077] As a preferred implementation of this embodiment, the causal relationship between the user's electricity consumption behavior and the same policy is analyzed by constructing a single causal conduction model through a structural equation model:

[0078] y=Λx+∈

[0079] Where y is the user's electricity consumption behavior data vector; x is the policy intervention data variable; Λ is the causal weight matrix, which is estimated by the least squares method.

[0080] The pseudo-algorithm of the Granger causality test is specifically:

[0081]

[0082]

[0083] S4: Use reinforcement learning algorithm to simulate user electricity consumption behavior under different policy intervention scenarios, build a comprehensive causal transmission model based on causal relationship to design the reward function of reinforcement learning, and use Monte Carlo method to simulate the impact of different policy combinations on user electricity consumption behavior. Determine the optimal policy intervention strategy based on the simulation results of reinforcement learning method and Monte Carlo method.

[0084] As a preferred implementation of this embodiment, step S4 is specifically as follows:

[0085] A comprehensive causal transmission model is constructed based on causal relationships, which can be expressed as:

[0086] y t =f(x(t),y(t-1),z(t),∈(t))

[0087] Where y(t) is the user's electricity consumption behavior at time t, x(t) is the policy intervention, z(t) is the user's environmental factor, and ∈(t) is the error term.

[0088] The Q-learning algorithm in reinforcement learning is used to simulate the user's electricity consumption behavior under different policy intervention scenarios, and the optimal policy intervention strategy combination is output. The state space S is the user's current electricity consumption behavior and policy context, the action space A is the policy adjustment measures, and the reward function R is the change in electricity consumption after the policy is implemented. The reward function of reinforcement learning is designed based on the comprehensive causal conduction model:

[0089] Q(s t ,a t )=Q(s t ,a t )+α[r t +γmaxQ(s tt1 ,a)-Q(s t ,a t )]

[0090] Among them, Q(s t ,a t ) is the state s at time t t ∈S, take action a t ∈A’s value function; r t ∈R is the immediate reward; a is the learning rate; γ is the discount factor; Q(s t+1 ,a t ) is the state s at time t+1 t+1 ∈S, take action a t ∈A’s value function.

[0091] Based on the optimal policy intervention strategy combination output by reinforcement learning, different policy intervention strategy combinations are defined, and a large amount of random user electricity consumption behavior simulation data is generated through the Monte Carlo method. Multiple random experiments are conducted under different policy intervention strategies to simulate the effects of different policy intervention strategy combinations, evaluate their long-term impact on the overall user electricity consumption behavior, and determine the optimal policy intervention strategy.

[0092] In the specific implementation and application, first, based on the optimal policy intervention strategy combination output by reinforcement learning, different policy combinations are defined (such as peak-valley electricity price adjustment, time-of-day electricity price change, etc.); secondly, the Monte Carlo method is used to generate a large amount of random user electricity consumption behavior simulation data, and multiple random experiments are conducted under different policy intervention strategies; finally, through statistical analysis of these simulation results, the long-term impact of each policy combination is evaluated, the stability and effectiveness of the policy intervention strategy are further verified, and the optimal policy intervention strategy is confirmed.

[0093] As a preferred implementation mode of this embodiment, the method also includes predicting the effect of policy intervention on user electricity consumption behavior through a simulation platform, optimizing the policy intervention strategy using an experimental design method (DOE), and tuning the hyperparameters of the causal transmission model through Bayesian optimization or grid search methods to improve the accuracy and optimization effect of the policy strategy.

[0094] As a preferred implementation of this embodiment, the method can integrate the causal conduction model with the existing power settlement system, realize real-time call and automated data analysis of the model through the API interface, and use container technologies such as Docker to encapsulate and deploy the model to ensure its cross-platform adaptability in different hardware and software environments. Subsequently, the monitoring and feedback mechanism (such as A / B testing) is used to track the effect of policy implementation, and the model is continuously optimized through reinforcement learning and feedback optimization algorithms to ensure its efficiency and accuracy under policy changes.

[0095] In order to verify the effectiveness and superiority of the method provided in this embodiment, some specific cases are provided below:

[0096] Taking Fujian Province as an example, the optimization method of electricity consumption quantity and fee settlement based on the causal conduction model is applied to analyze the relationship between user electricity consumption behavior and electricity price policy intervention in Fujian Province, and optimize the quantity and price settlement strategy. Fujian Province has currently implemented a number of policy intervention measures, such as time-of-use electricity prices and peak-valley period adjustments, to encourage residents to reduce peak loads and fill valleys and reduce grid load.

[0097] Step S1:

[0098] The data was collected by the State Grid Fujian Electric Power Company over the past three years (on a monthly basis, a total of 36 months). The data comes from data platforms such as Marketing 2.0, including monthly electricity consumption, electricity bill records, user electricity price response (such as participation in time-of-use electricity prices), payment time and channels, etc. The number of users is 100,000, and the data covers multiple categories of user groups.

[0099] Through market information, government websites and other channels, we collected data on electricity price policy interventions in the past three years, especially electricity price adjustment policies, peak and valley time division, time-of-use electricity price policies and demand response policies.

[0100] Through data cleaning techniques, such as outlier detection and missing value filling, the collected data is preprocessed to ensure the integrity and consistency of the data, forming user data sets and policy intervention data sets.

[0101] Step S2:

[0102] The principal component analysis is performed on the user data set panel data (100,000 users × 36 months) to extract key features. Specific features include users' monthly electricity consumption, payment time, payment frequency, etc.

[0103] Cluster analysis was performed on the user electricity consumption behavior data after dimensionality reduction, and the K-means clustering algorithm was used to divide the users into three categories: "peak electricity consumption type", "valley electricity consumption type" and "smooth electricity consumption type". Among them, "peak electricity consumption type" users consume more electricity during the peak electricity price period, while "valley electricity consumption type" users consume electricity mainly during the valley period.

[0104] Based on the classification results of users' electricity consumption behaviors and combined with the policy intervention data set, the random forest algorithm is used to analyze users' response patterns to different electricity price policies, extract potential influencing factors, and divide users into three categories: "high sensitivity", "medium sensitivity" and "low sensitivity". The users are then matched with the clustering results to form a 3*3 user category matrix.

[0105] Step S3:

[0106] The 36-month user electricity consumption behavior data and policy intervention data are converted into a time series dataset. The user's monthly electricity consumption is the electricity consumption time series, and the policy intervention (such as electricity price adjustment and peak and valley time) is the policy time series.

[0107] Through the Granger causality test, the impact of different policy interventions on users' electricity consumption behavior was verified. For example, the results showed that the peak-valley electricity price adjustment significantly affected the electricity consumption of "peak electricity consumption-high sensitivity" users, and the causal relationship was significant at a confidence level of 95%.

[0108] The structural equation model is further used to analyze the causal relationship between user electricity consumption behavior and policy changes, revealing the long-term impact of electricity price adjustments on different user groups. The results show that time-of-use electricity prices have a significant impact on the long-term electricity consumption patterns of "peak electricity consumption-high sensitivity" users, while the impact on "stable electricity consumption-low sensitivity" users is relatively small.

[0109] Step S4:

[0110] Based on the user's electricity consumption behavior and policy data, the Q-learning algorithm is used to simulate the user's electricity consumption behavior response under different policy scenarios. The state space includes the user's current electricity consumption behavior and policy scenarios (such as peak and valley electricity prices), and the action space is the adjustment of the electricity price policy (such as adjusting the peak and valley time periods). The reward function is the change in user electricity consumption after the policy is implemented.

[0111] The Monte Carlo method was used to simulate the long-term impact of different policy combinations on user electricity consumption behavior. More than 100,000 electricity consumption scenarios with different policy combinations were simulated to generate random user electricity consumption behavior data. The results show that the combination of "peak-valley electricity price + time adjustment" can effectively reduce the peak-time electricity consumption of "peak electricity consumption type-high sensitivity type" users and optimize the power grid load.

[0112] Based on the simulation results of the reinforcement learning algorithm and the Monte Carlo method, the combination of "time-of-use electricity price adjustment + demand response" was finally determined as the optimal policy intervention strategy, which can reduce electricity consumption during peak hours while ensuring the balance of power grid load.

[0113] Embodiment 2

[0114] Accordingly, this embodiment provides an electricity consumption fee settlement optimization system based on a causal conduction model, including a data collection and processing module, a user electricity consumption behavior identification module, a causal relationship analysis module and a policy intervention strategy generation module.

[0115] The data collection and processing module is used to collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form a user data set and a policy intervention data set. This module is used to implement the function of step S1 in Example 1 and will not be repeated here.

[0116] The user electricity consumption behavior identification module is used to extract key user electricity consumption behavior features from the user data set through principal component analysis, use clustering algorithm and random forest algorithm to analyze the user's response pattern to different policies, divide the user response patterns into different categories, and form user data sets and policy intervention data sets of different user category matrices. This module is used to implement the function of step S2 in embodiment 1 and will not be repeated here.

[0117] The causal relationship analysis module is used to convert the data sets of different user category matrices and the policy intervention data sets into time series data sets, verify the impact of different policy interventions on user electricity consumption behavior through Granger causality, and further use the structural equation model to analyze the causal relationship between user electricity consumption behavior and policy intervention changes. This module is used to implement the function of step S3 in Example 1 and will not be repeated here.

[0118] The policy intervention strategy generation module is used to use the reinforcement learning algorithm to simulate the user's electricity consumption behavior response under different policy intervention scenarios, build a comprehensive causal conduction model based on causal relationships to design the reinforcement learning reward function, and simulate the impact of different policy combinations on user electricity consumption behavior through the Monte Carlo method. The optimal policy intervention strategy is determined based on the simulation results of the reinforcement learning method and the Monte Carlo method. This module is used to implement the function of step S4 in Example 1 and will not be repeated here.

[0119] Embodiment 3

[0120] This embodiment provides an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for optimizing electricity consumption fee settlement based on a causal conduction model as described in any embodiment of the present invention is implemented.

[0121] Embodiment 4

[0122] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the method for optimizing electricity consumption fee settlement based on a causal conduction model as described in any embodiment of the present invention is implemented.

[0123] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0124] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0125] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0126] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.

[0127] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for optimizing electricity consumption fee settlement based on a causal conduction model, characterized in that: The following steps are involved: S1: Collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form user data sets and policy intervention data sets; S2: Extract key user electricity consumption behavior features from user data sets through principal component analysis, use clustering algorithm and random forest algorithm to analyze user response patterns to different policies, divide user response patterns into different categories, and form user data sets and policy intervention data sets for different user category matrices; S3: Convert the datasets of different user category matrices and policy intervention datasets into time series datasets, verify the impact of different policy interventions on user electricity consumption behavior through Granger causality, and further use the structural equation model to analyze the causal relationship between user electricity consumption behavior and policy intervention changes; S4: Use reinforcement learning algorithm to simulate user electricity consumption behavior responses under different policy intervention scenarios, build a comprehensive causal conduction model based on causal relationship to design the reward function of reinforcement learning, obtain the optimal policy intervention strategy combination, and use the Monte Carlo method to simulate the impact of different policy intervention strategy combinations on user electricity consumption behavior. Determine the optimal policy intervention strategy based on the simulation results of the reinforcement learning method and the Monte Carlo method.

2. The electricity consumption fee settlement optimization method based on the causal conduction model according to claim 1 is characterized in that: The step S1 is specifically as follows: Collect user electricity behavior data including historical electricity consumption, electricity charges, customer service records, and payment channels, as well as policy intervention data including time-of-use electricity prices, peak and valley periods, tiered electricity prices, demand response policies, and green electricity grid connection policies; The collected data are preprocessed using data cleaning techniques such as outlier detection and missing value filling to form user data sets and policy intervention data sets.

3. The electricity consumption fee settlement optimization method based on the causal conduction model according to claim 1 is characterized in that: The step S2 is specifically as follows: The principal component analysis is used to extract the key user electricity consumption behavior features X from the user data set. Reduced =WX, where X Reduced is the data matrix after dimensionality reduction, R n*d is the user's electricity consumption behavior data, n represents the number of users, and d represents the number of features for each user; W∈R k*d is the dimension reduction matrix, k is the reduced feature dimension, and X is the original user data set; Use clustering algorithm to classify the user electricity consumption behavior features extracted by principal component analysis to obtain the user electricity consumption behavior classification results; Based on the classification results of user electricity consumption behavior and combined with the policy intervention data set, the random forest algorithm is used to analyze the user's response patterns to different electricity price policies, and the user response patterns of different categories are obtained. The patterns are matched with the clustered user electricity consumption behavior classification results to form user data sets and policy intervention data sets of different user category matrices.

4. The electricity consumption fee settlement optimization method based on the causal conduction model according to claim 1 is characterized in that: In step S3, the effects of different policy interventions on the user's electricity consumption behavior are verified by Granger causality as follows: The user data set is converted into a time series y(t), and the policy intervention data set is converted into a time series x(t). The single causal transmission model is expressed as: y(t)=αy(t-1)+βx(t-1)+∈(t) In the formula, α is the autoregressive influence coefficient of the user's electricity consumption behavior at time point t-1 on the user's electricity consumption behavior y(t) at the current time point; β is the influence coefficient of the policy intervention variable at time point t-1 on the electricity consumption behavior y(t) at the current time point; ∈(t) is the error term; t is time; According to Granger causality test, if x(t) can significantly predict y(t), then x(t) has a causal relationship with y(t): H0:β=0,H1:β≠0 In the formula, H0 assumes that policy intervention is not a causal factor in users' electricity consumption behavior; H1 assumes that policy intervention is a causal factor in users' electricity consumption behavior; regression analysis is used to test whether H0 is rejected.

5. The electricity consumption fee settlement optimization method based on the causal conduction model according to claim 4 is characterized in that: The single causal transmission model constructed by the structural equation model analyzes the causal relationship between user electricity consumption behavior and the same policy as follows: y=Λx+∈ Where y is the user's electricity consumption behavior data vector; x is the policy intervention data variable; Λ is the causal weight matrix, which is estimated by the least squares method.

6. The electricity consumption fee settlement optimization method based on the causal conduction model according to claim 1 is characterized in that: The step S4 is specifically as follows: A comprehensive causal transmission model is constructed based on causal relationships, which can be expressed as: y t =f(x(t),y(t-1),z(t),∈(t)) In the formula, y(r) is the user's electricity consumption behavior at time t, x(t) is the policy intervention, z(t) is the user's environmental factor, and ∈(r) is the error term; The Q-learning algorithm in reinforcement learning is used to simulate the user's electricity consumption behavior under different policy intervention scenarios, and the optimal policy intervention strategy combination is output. The state space S is the user's current electricity consumption behavior and policy context, the action space A is the policy adjustment measures, and the reward function R is the change in electricity consumption after the policy is implemented. The reward function of reinforcement learning is designed based on the comprehensive causal conduction model: Q(s t ,a t )=Q(s t ,a t )+α[r t +γmaxQ(s t+1 ,a)-Q(s t ,a t )] Among them, Q(s t ,a t ) is the state s at time t t ∈S, take action a t ∈A’s value function; r t ∈R is the immediate reward; a is the learning rate; γ is the discount factor; Q(s t+1 ,a t ) is the state s at time t+1 t+1 ∈S, take action a t ∈A's value function; Based on the optimal policy intervention strategy combination output by reinforcement learning, different policy intervention strategy combinations are defined. The Monte Carlo method is used to conduct multiple random experiments under different policy intervention strategy combinations to simulate the effects of different policy intervention strategy combinations, evaluate their long-term impact on the overall user electricity consumption behavior, and determine the optimal policy intervention strategy.

7. An electricity consumption fee settlement optimization system based on a causal conduction model, characterized in that: The system realizes optimization of electricity consumption side fee settlement based on the electricity consumption side fee settlement optimization method based on the causal conduction model according to any one of claims 1 to 6, including a data collection and processing module, a user electricity consumption behavior identification module, a causal relationship analysis module and a policy intervention strategy generation module; The data collection and processing module is used to collect user electricity consumption behavior data and policy intervention data, and pre-process the data through data cleaning technology to form user data sets and policy intervention data sets; The user electricity consumption behavior identification module is used to extract key user electricity consumption behavior features from the user data set through principal component analysis, and use clustering algorithm and random forest algorithm to analyze the user response patterns to different policies, classify the user response patterns into different categories, and form user data sets and policy intervention data sets of different user category matrices; The causal relationship analysis module is used to convert the data sets of different user category matrices and policy intervention data sets into time series data sets, verify the impact of different policy interventions on user electricity consumption behavior through Granger causality, and further use the structural equation model to analyze the causal relationship between user electricity consumption behavior and policy intervention changes; The policy intervention strategy generation module is used to use the reinforcement learning algorithm to simulate the user's electricity consumption behavior response under different policy intervention scenarios, build a comprehensive causal conduction model based on causal relationships to design the reinforcement learning reward function, obtain the optimal policy intervention strategy combination, and simulate the impact of different policy intervention strategy combinations on user electricity consumption behavior through the Monte Carlo method. The optimal policy intervention strategy is determined based on the simulation results of the reinforcement learning method and the Monte Carlo method.

8. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the electricity consumption fee settlement optimization method based on the causal conduction model as described in any one of claims 1 to 6 when executing the computer program.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for optimizing electricity consumption fee settlement based on a causal conduction model as described in any one of claims 1 to 6 is implemented.

Images