CN117132334A - Intelligent pricing method, system, electronic device and storage medium for electric power retail - Google Patents

Intelligent pricing method, system, electronic device and storage medium for electric power retail Download PDF

Info

Publication number
CN117132334A
CN117132334A CN202310838437.1A CN202310838437A CN117132334A CN 117132334 A CN117132334 A CN 117132334A CN 202310838437 A CN202310838437 A CN 202310838437A CN 117132334 A CN117132334 A CN 117132334A
Authority
CN
China
Prior art keywords
user
behavior
electricity
data
attention
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202310838437.1A
Other languages
Chinese (zh)
Inventor
阮嘉祺
白焰
赵俊华
梁高琪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Chinese University of Hong Kong Shenzhen
Original Assignee
Chinese University of Hong Kong Shenzhen
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Chinese University of Hong Kong Shenzhen filed Critical Chinese University of Hong Kong Shenzhen
Priority to CN202310838437.1A priority Critical patent/CN117132334A/en
Publication of CN117132334A publication Critical patent/CN117132334A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0283—Price estimation or determination
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • General Health & Medical Sciences (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Biophysics (AREA)
  • Data Mining & Analysis (AREA)
  • Marketing (AREA)
  • Software Systems (AREA)
  • Finance (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Accounting & Taxation (AREA)
  • Computational Linguistics (AREA)
  • Tourism & Hospitality (AREA)
  • Primary Health Care (AREA)
  • Human Resources & Organizations (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Water Supply & Treatment (AREA)
  • Public Health (AREA)
  • Game Theory and Decision Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application discloses an intelligent pricing method, system, electronic device and storage medium for electric power retail, wherein the method comprises the following steps: acquiring historical energy consumption behavior data of a target user; inputting the historical energy utilization behavior data into a pre-trained energy utilization behavior learning model of the user, and predicting the energy utilization behavior of the user; calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption requirement of the user according to the energy consumption behavior, and obtaining the requirement response potential of the energy consumption behavior of the user; inputting the demand response potential into a pre-constructed pricing model to obtain a pricing method; the application can be practically applied to personalized pricing of power retailers facing business users and resident users, and the time-varying demand response potential of the users can be developed to the greatest extent on the premise of controlling the power price fluctuation risk.

Description

Intelligent pricing method, system, electronic device and storage medium for electric power retail
Technical Field
The application relates to the field of power, in particular to an intelligent pricing method, system, electronic device and storage medium for power retail.
Background
The power peak demand management (peak demand management, PDM) aims to reduce grid peak load, reduce grid operation safety risk caused by peak load and improve operation economy. As the overall power demand grows, a high proportion of renewable energy is accessed into the grid, the power system faces greater and more severe PDM demands.
Demand Response (DR) can be used to reduce system peak load levels, which is the basis of PDM. DR changes traditional "power generation tracking load" dispatch mode, guides user's rational electricity consumption through price response or excitation mechanism, initiatively changes the power consumption load profile, supports power supply and demand dynamic balance. The access of the distributed energy sources and the energy storage equipment at the user side enables the user side to have DR potential, and is a flexible resource for promoting PDM. In addition, under the new round of electricity change, the construction work of the national electric power market is steadily advanced, but the electric power retail market still belongs to the development stage, and the strengthening of the construction of the retail market is needed.
Disclosure of Invention
The application mainly aims to provide an intelligent pricing method, an intelligent pricing system, an electronic device and a storage medium for power retail, and aims to optimize pricing modes of power retail markets and strengthen power retail market construction.
To achieve the above object, a first aspect of the present application provides an intelligent pricing method for electric power retail, including: acquiring historical energy consumption behavior data of a target user; inputting the historical energy utilization behavior data into a pre-trained energy utilization behavior learning model of the user, and predicting the energy utilization behavior of the user; calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption requirement of the user according to the energy consumption behavior, and obtaining the requirement response potential of the energy consumption behavior of the user; and inputting the demand response potential into a pre-built pricing model to obtain a pricing method.
Further, the training method of the behavior learning model for the user comprises the following steps: pre-constructing a user energy consumption behavior learning model; selecting historical user behavior characteristics and inputting the user behavior learning model, wherein the behavior characteristics comprise: historical behavior, electricity price, temperature, humidity, wind power and time; performing feature processing on the user behavior feature by adopting a multi-class attention mechanism in the behavior learning model; and deep learning is carried out on the user behavior characteristics processed by the characteristics in the behavior learning model by adopting a stacked neural network, so that a trained user behavior learning model is obtained.
Further, the multiple types of attention mechanisms are time mode attention mechanisms based on long-short term memory neural network, and the method for extracting the historical behavior characteristics of the user comprises the following steps:
is provided withIs a hidden state matrix of the LSTM layer, wherein the dimension m is the feature quantity, the dimension u is the time step, and H LSTM Middle->The i-th hidden state of the LSTM layer;
hidden state matrix for previous u-1 stepsPerforming one-dimensional convolution processing to extract a time mode matrix H in a convolution kernel range C ：
Wherein:is H C The ith row and jth column elements in H represent the ith row vector and jth convolution kernel +.>A processed result value; />Is a matrix after convolution operation; k is the number of convolution kernels; h i，q For the i-th row and q-th column element in H, wherein q=u-T-1+l is calculated in the convolution field of view; t isConvolving the field of view; c j，l C is j The first element of (a);
a scoring mechanism s (·) is introduced to evaluate the hidden state h of the u-th step u With a convolution time model matrix H C Is a correlation of the row vectors of (a):
wherein:is H C The dimension K is the feature number of all convolution kernels after processing;mapping a matrix for attention in a scoring mechanism;
normalizing a scoring mechanism by using a Sigmoid (·) activation function to obtain a representative h u Andattention coefficient alpha of related information i ：
And according to the obtained attention coefficient, carrying out attention weighting and addition operation, and outputting the hidden state under the time mode attention mechanism:
wherein:and->Are all a learnable parameter matrix of the TPA layer; />The (u) th step hidden state after the LSTM layer and the TPA layer are treated contains m features.
Further, the multiple types of attention mechanisms are used for processing electricity price, temperature, humidity, wind power and time characteristics, and comprise the following steps:
processing electricity price characteristics: obtaining zero-one index of corresponding electricity price data through a time sequence learning layer and a conventional attention layer of historical behavior data; performing dot multiplication operation on the zero-one index and the electricity price data, and removing low-importance electricity price data, wherein the method specifically comprises the following steps of:
given a givenFor a price package containing N time periods of electricity price on day t, for L t Go about p t Processing to obtain the hidden state of electricity price related to the behavior of energy consumption>The timing learning layer is expressed as a function LSTM (.):
attention index related to electricity price is obtained through conventional attention layer
Then through dot product operationGet the weighted electricity price data characteristic hidden state +.>
The processing of the temperature, humidity and wind power characteristics comprises the following steps:
given a givenAnd->And->Andthe time series data of the outdoor dry bulb temperature, the outdoor wet bulb temperature, the relative humidity of the outdoor air, the outdoor air humidity ratio, the wind speed and the wind direction are the t day, and the dimension corresponds to the time of electricity price; obtaining the data characteristic hidden state of the outdoor dry bulb temperature and the outdoor wet bulb temperature after attention weighting through a time sequence learning layer, a conventional attention layer and dot multiplication operationAnd->And->And->
Further, the deep learning includes: the processed electricity price, temperature, humidity and wind power data are spliced through fusion layers, deep data mining is carried out through stacked full-connection layers, influences of different electricity prices, temperatures, humidity and wind power on energy utilization behaviors are learned, and output are carried outCo-dimensional mixed feature hidden state->Will->Superimposed on->And performing depth data mining again by stacking the full connection layers, and outputting user energy behaviors.
Further, the calculating method of the ratio comprises the following steps:
wherein beta is t，k Time-varying DR potential for the user at the kth period on the t-th day; d, d t，k And Δd t，k The electricity demand of the user at the kth period of the t-th day and the electricity demand that can be cut down/diverted, respectively.
The pricing model is further:
wherein,personalized retail electricity price packages representing the t-th day for decision variables; />For the user +.>Estimating the following energy utilization behavior; />The price of wholesale electricity generated for market clearing before the t day;for the user +.>Estimating the following energy utilization behavior; θ (·) is the energy consumption behavior learning model after training; />For the weather forecast collection on day t, obtained from the weather forecast system, wherein +.>And->Is the outdoor dry bulb temperature and wet bulb temperature, < + >>And->For the outdoor air relative humidity and air humidity ratio, +.>And->Wind speed and direction; />And->Respectively->And->The kth element of (a); />Is a price change threshold vector in a pricing scheme.
Further, after the pricing method is obtained, price risk hedging constraint is further performed, including:
wherein:a risk of revenue for the retailer on day t; />Is a cost risk to the user on day t.
A second aspect of the present application provides an intelligent pricing system for power retail, comprising: the data acquisition module is used for acquiring historical energy utilization behavior data of the target user; the behavior prediction module is used for inputting the historical energy consumption behavior data into a pre-trained energy consumption behavior learning model of the user and predicting the energy consumption behavior of the user; the demand response calculation module is used for calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption demand of the user according to the energy consumption behavior to obtain the demand response potential of the energy consumption behavior of the user; and the pricing method generation module is used for inputting the demand response potential into a pre-constructed pricing model to obtain a pricing method.
A third aspect of the present application provides an electronic device, comprising: the intelligent pricing system for the electric power retail comprises a memory and a processor, wherein the memory stores a computer program capable of running on the processor, and the intelligent pricing system for the electric power retail is characterized in that when the processor executes the computer program.
A fourth aspect of the present application provides a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the smart pricing method of electricity retail of any of the above.
The application provides an intelligent pricing method, system, electronic device and storage medium for electric power retail, which have the beneficial effects that: the method is suitable for personalized pricing of power retailers facing business users and resident users in practice, and the time-varying demand response potential of the users is developed to the greatest extent on the premise of controlling the power price fluctuation risk.
Drawings
In order to more clearly illustrate the embodiments of the application or the technical solutions in the prior art, the drawings that are necessary for the description of the embodiments or the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the application and that other drawings may be obtained from them without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of an intelligent pricing method for power retail according to an embodiment of the application;
FIG. 2 is a specific illustration of two phases of an intelligent pricing method for power retail according to an embodiment of the application;
FIG. 3 is a deep learning overall framework diagram of an intelligent pricing method for power retail in accordance with an embodiment of the present application;
FIG. 4 is a comparison of the front and rear of an intelligent pricing method for power retail using an embodiment of the application;
FIG. 5 is a distinction between time-of-use wholesale electricity prices and personalized retail electricity prices verified by the intelligent pricing method for power retail according to an embodiment of the present application;
FIG. 6 is a block diagram of an intelligent pricing system for power retail according to embodiments of the application;
fig. 7 is a schematic block diagram of an electronic device according to an embodiment of the present application.
Detailed Description
In order to make the objects, features and advantages of the present application more comprehensible, the technical solutions in the embodiments of the present application will be clearly described in conjunction with the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are only some embodiments of the present application, but not all embodiments of the present application. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Referring to fig. 1, an intelligent pricing method for power retail comprises:
s101, acquiring historical energy consumption behavior data of a target user;
s102, inputting historical energy consumption behavior data into a pre-trained energy consumption behavior learning model of a user, and predicting the energy consumption behavior of the user;
s103, calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption requirement of the user according to the electricity consumption behavior, and obtaining the requirement response potential of the user electricity consumption behavior;
s104, inputting the demand response potential into a pre-built pricing model to obtain a pricing method.
Specifically, the intelligent pricing method for power retail provided by the application is divided into two stages of building a behavior learning model and developing demand response potential at a user side. The method steps can be seen in fig. 2.
As can be seen from fig. 2, the two stages are respectively:
firstly, the power retailers firstly select relevant user energy consumption behavior characteristics, collect corresponding historical behavior data, secondly conduct characteristic processing on the data, input the extracted data into a deep learning model for training, and obtain an energy consumption behavior learning model of a specific user.
And step two, according to the result of the daily power market clearing and the DR period of release, the retailers predict the energy consumption behavior of the users under the price of the power wholesale market by adopting the trained energy consumption behavior learning model, evaluate the time-varying DR potential of the users, formulate personalized retail electricity prices for the users and stimulate the users to adjust the energy consumption behavior through the price. And finally, the user receives a new electricity price package through the intelligent electric meter, responds to the demand under the excitation of the electricity price, changes the original energy utilization behavior and reduces the electricity consumption at peak time.
Specifically, the first stage is used for establishing a behavior learning model for a user side, and the general deep learning framework of the behavior learning model comprises four parts of behavior characteristics, characteristic processing, deep learning and behavior prediction: the behavior characteristic part mainly inputs related information about the behavior of the energy consumption, including historical behavior, electricity price, temperature, humidity, wind power and time; the feature processing processes various feature information, such as normalization, so as to reduce the influence of non-uniform feature dimensions on learning ability; then, the relation between the energy consumption behavior and each characteristic, such as a multi-layer perceptron, a recurrent neural network and the like, is constructed through a deep learning technology; and finally predicting the energy utilization behavior under the characteristic information.
When the performance characteristics are selected, the following points can be obtained:
1) Historical behavior: the energy utilization behavior of each user has the characteristics. Energy usage levels and energy usage patterns associated with their own preferences are generated on various typical days, such as weekdays, weekends, holidays, etc. The historical behavior information can help learn the power consumption behavior patterns of the user on different typical days, and build a datum line of the energy consumption behavior learning model. The datum line outputs the energy utilization behavior result in combination with other characteristics, and the prediction performance is improved.
2) Electricity price: the user forms specific electricity price elasticity according to objective factors such as income, consumption level and the like, and the elasticity can generate different energy utilization behaviors along with the change of the electricity price. For example, washing machines and dryers are among the transferable load resources. When the electricity rate is increased, the user may shift the washing machine, the dryer to use in a low electricity rate period to reduce the electricity rate expenditure, thereby causing a change in the energy consumption behavior.
3) Temperature: the temperature mainly has a significant influence on the use of electric appliances having cold and hot load characteristics, such as air conditioners, water heaters and the like. Specifically, the user frequently uses the cooling load such as air conditioner to realize the cooling in the hot period in summer. The heat load is needed in winter to increase the temperature so as to improve the comfort level of life. In addition, high temperatures can have an impact on computer hardware operation, for example, data centers can utilize air conditioning to reduce the heat generated by server operation.
4) Humidity: similar to temperature, humidity mainly affects the load of a humidifier/dehumidifier, a fresh air system, etc. which can regulate indoor humidity. And when the humidity is too high/low, the user adjusts the indoor humidity through the electric equipment according to the self adaptation degree to the environment. Meanwhile, too high humidity can reduce the insulation strength of the electrical equipment, corrode metal elements, reduce the performance and service life of the equipment, and even cause electrical faults. Thus, some manufacturing power customers may develop indoor humidity standards to reduce the risk of electrical failure.
5) Wind power: due to house construction, outdoor wind speed and wind direction affect indoor aerodynamic force through ventilation openings such as doors and windows. The operation mode and the state of the air system such as the indoor air conditioner, the fresh air system and the like are directly acted, and the energy utilization behavior is changed. In addition, for users who like natural wind, the use of an air conditioner, a fan and a fresh air system can be reduced under the condition that the outdoor wind is sufficient, so that the energy consumption behavior of the users is changed.
6) Time: because of regular operation such as individual work and rest, users have time preference for the use of specific electric appliances, such as making breakfast in the morning with a bread machine, charging electric vehicles after going to work at night, and the like. Time of day preferences will constitute a typical usage behavior pattern. By learning the time characteristics, the construction of an accurate energy utilization behavior learning model is facilitated.
When the characteristic processing is carried out, the following processing modes are adopted corresponding to different energy utilization behavior characteristics:
1) Historical behavior: deep learning algorithm (LSTM-TPA layer) extraction of input features using LSTM-based time-mode attention mechanism
First, a one-dimensional convolutional neural network is used for improving the characteristic learning ability of the hidden state of the LSTM network. Assume thatIs a hidden state matrix of the LSTM layer, wherein the dimension m isFeature quantity, dimension u is time step, H LSTM Middle->Is the i-th hidden state of the LSTM layer. Hidden state matrix for previous u-1 stepPerforming one-dimensional convolution processing to extract a time mode matrix H in a convolution kernel range C ：
Wherein:is H C The ith row and jth column elements in H represent the ith row vector and jth convolution kernel +.>A processed result value; />Is a matrix after convolution operation; k is the number of convolution kernels; h i，q For the i-th row and q-th column element in H, wherein q=u-T-1+l is calculated in the convolution field of view; t is the convolution field of view; c j，l C is j The first element of (a);
second, a scoring mechanism s (·) is introduced to evaluate the hidden state h of the u-th step u With a convolution time model matrix H C Is a correlation of the row vectors of (a):
wherein:is H C The dimension K is the feature number of all convolution kernels after processing;mapping a matrix for attention in a scoring mechanism;
then, normalizing a scoring mechanism by using a Sigmoid (·) activation function to obtain a representative h u Andattention coefficient alpha of related information i ：
Finally, according to the obtained attention coefficient, carrying out attention weighting and addition operation, and outputting the hidden state under the time mode attention mechanism:
wherein:and->Are all a learnable parameter matrix of the TPA layer; />The (u) th step hidden state after the LSTM layer and the TPA layer are treated contains m features.
The historical behavior time sequence is an energy consumption behavior curve of the user on the past day. Given a givenFor the energy consumption behavior curve at the t-th day N time periods (corresponding to the electricity price clearing time granularity), then +.> Is historical behavior information of the past S days at the t-th day. Extraction of user presence using proposed temporal pattern attention mechanism (TPA-LSTM layer)Electric power consumption modes of different typical days, and predicting the energy consumption behavior standard value of a user on the t th day, and outputting the standard value as +.>
2) Electricity price: after data is removed, performing feature processing through an LSTM layer, a Sigmoid layer and dot multiplication operation
The electricity price time sequence is an electricity price package formulated by retailers for power users, and 24/48/96 time period electricity prices can be formed every day according to the market clearing time granularity of 1-h/30-min/15-min. The neural network layer for processing electricity prices is constructed by considering typical daily effects. Zero-index of the corresponding electricity price data is obtained through time sequence learning (LSTM layer) and conventional attention layer (Sigmoid layer) of the historical behavior data. And performing dot multiplication operation on the index and the electricity price data, and eliminating the low-importance electricity price data. Given a givenFor a price package containing N time periods of electricity price on day t, for L t Go about p t Is processed to obtain the hidden state of electricity price related to the behavior of the user (expressed by the function LSTM>
Attention index related to electricity price obtained through Sigmoid layer
Then through dot product operationGet the weighted electricity price data characteristic hidden state +.>
3) Temperature data:
the temperature time series includes outdoor dry bulb temperature and outdoor wet bulb temperature data, givenAndthe time series data of the outdoor dry bulb temperature and the outdoor wet bulb temperature on the t th day are the dimension corresponding to the time of electricity price. Similar to electricity price time sequence data processing, the data characteristic hidden states of the outdoor dry bulb temperature and the outdoor wet bulb temperature after attention weighting are obtained through an LSTM layer, a Sigmoid layer and dot multiplication operation>And->
4) Humidity data:
the humidity time series includes an outdoor air relative humidity and an outdoor air humidity ratio. Given a givenAndfor the data of the relative humidity of the outdoor air and the humidity ratio of the outdoor air on the t th day, the data characteristic hidden state of the weighted relative humidity of the outdoor air and the humidity ratio of the outdoor air is obtained through the LSTM layer, the Sigmoid layer and the dot product operationAnd->
5) Wind power data:
the wind time series includes wind speed and wind direction. Given a givenAnd->For the data of the wind speed and the wind direction on the t-th day, obtaining the data characteristic hidden state of the wind speed and the wind direction after attention weighting through an LSTM layer, a Sigmoid layer and dot multiplication operation>And->
In one embodiment, a training method for a user with a behavior learning model includes: pre-constructing a user energy consumption behavior learning model; selecting historical user behavior characteristics, inputting a user behavior learning model, wherein the behavior characteristics comprise: historical behavior, electricity price, temperature, humidity, wind power and time; performing feature processing on the user behavior feature by adopting a multi-class attention mechanism in the behavior learning model; and deep learning is carried out on the user behavior characteristics processed by the characteristics in the behavior learning model by adopting a stacked neural network, so that a trained user behavior learning model is obtained.
In particular, the present application mines associations between input features and functional behavior based on deep learning models of multiple classes of attention mechanisms. The corresponding neural network layer is set as follows:
splicing the processed electricity price, temperature, humidity and wind power data through fusion layers, deep data mining through stacked full-connection layers, learning the influence of different electricity price, temperature, humidity and wind power on energy utilization behaviors, and outputtingCo-dimensional mixed feature hidden state->Finally, will->Superimposed on->And performing depth data mining again by stacking the full connection layers, and outputting user energy behaviors. The overall framework for deep learning is shown in fig. 3.
The deep learning model shown in fig. 3 may be generalized to business users and residential users: for business users, the energy consumption behavior is single, a more stable energy consumption behavior mode is provided, and the model can be used for focusing on the dependence of behavior prediction on historical behaviors; for resident users, the energy uncertainty is strong, the history behavior is unstable, the model gives less weight, and the influence caused by other characteristic factors is considered more.
In this embodiment, the collected real data set is divided into a training set and a validation set according to a certain proportion, the training set is used for training the deep learning model, and the validation set is used for determining the network structure and controlling the model parameters. And finally, inputting the historical characteristic data of the target user into a trained model to predict.
In one embodiment, the patent proposes a quantitative indicator of time-varying DR potential according to user energy consumption behavior, and proposes a pricing model of personalized electricity price packages from the point of view of the power retailers, so as to maximize development of user time-varying DR potential.
Firstly, a time-varying demand response potential quantitative evaluation needs to be carried out, and different DR potentials (time-varying DR potentials) of the same user in different time periods are defined as the ratio of the power consumption change which can be cut down/transferred to the original power consumption demand of the user: (the larger its value, the greater the time-varying DR potential representing the user over the period, and vice versa),
wherein: beta t，k Time-varying DR potential for the user at the kth period on the t-th day; d, d t，k And Δd t，k The electricity demand of the user at the kth period of the t-th day and the electricity demand that can be cut down/diverted, respectively.
On the other hand, there is a need to build a retailer personalized pricing model based on a usage behavior learning model:
after the day-ahead power market has cleared wholesale prices, the retailer will build the following optimization model by formulating an optimal personalized retail power price package to maximize the user's time-varying DR potential during the kth period of the t-th day. The objective function is equation 1 to maximize the user's time-varying DR potential at the kth period on day t by optimizing the personalized retail price package. Equations 2 and 3 estimate the user's performance at wholesale price and retail price packages by using the trained performance learning model. Equation 4 employs a personalized retail pricing scheme for the DR potential of the user during the kth period on day t. Equation 5 sets a price change threshold for price risk hedging constraint, and defines the price change amplitude of each period.
Wherein:personalized retail electricity price packages representing the t-th day for decision variables; />For the user +.>Estimating the following energy utilization behavior; />The price of wholesale electricity generated for market clearing before the t day;for the user +.>Estimating the following energy utilization behavior; θ (·) is the energy consumption behavior learning model after training; />For the weather forecast collection on day t, obtained from the weather forecast system, wherein +.>And->Is the outdoor dry bulb temperature and wet bulb temperature, < + >>And->For the outdoor air relative humidity and air humidity ratio, +.>And->Wind speed and direction; />And->Respectively->And->The kth element of (a); />Is a price change threshold vector in a pricing scheme.
For retailers, their price risk is reflected as a revenue risk; for the user, the price risk is embodied as cost risk, and can be described by the following mathematical formula:
wherein:a risk of revenue for the retailer on day t; />Is a cost risk to the user on day t.
In one embodiment, verification is also performed, as follows:
the embodiment of the application adopts a real electricity price, temperature, humidity, wind power data and weather data combined data simulation method in the British regions of 2019 and 2020 to construct a user historical behavior database according to 8:1: the proportion of 1 is divided into a training set, a verification set and a test set, the simulation is carried out by inputting the energy utilization behavior model based on deep learning provided by the patent, and the simulation is compared with several advanced deep learning models, and the model description is shown in table 1.
Table 1 comparative scenario with behavior learning model
Through prediction, all prediction results show similar results, and compared with LSTM-1, LSTM-2 and HM models (energy-consumption behavior learning models without attention mechanisms), the MCAHM model provided by the patent shows predictions closer to a true value at other positions. It can be seen that the comprehensive predictive performance of the proposed behavior learning model is optimal after using multiple types of attention mechanisms.
Secondly, the trained model MCAHM is adopted to simulate on 2020 data in the whole year, price signals positively correlated with load levels are adopted to guide the implementation time of DR, the result is shown in fig. 4 and 5, wherein fig. 4 describes the electricity utilization behaviors of users before and after the personalized retail pricing method is adopted, and different color blocks represent different types of electricity utilization loads of the users. It can be seen that part of electricity load is transferred from electricity peak period to low valley period under the personalized pricing method, so that the whole electricity utilization form is smoother, and a good demand response effect is achieved. Fig. 5 compares the difference between time-of-use wholesale electricity prices and personalized retail electricity prices. It can be seen that the personalized retail electricity price curve approximately accords with the daily load curve change trend, and shows a double peak characteristic at noon and evening, and reflects the adjustment mechanism of price signals to loads.
Referring to fig. 6, the embodiment of the present application further provides an intelligent pricing system for power retail, including: the system comprises a data acquisition module 1, a behavior prediction module 2, a demand response calculation module 3 and a pricing method generation module 4; the data acquisition module is used for acquiring historical energy consumption behavior data of the target user; the behavior prediction module is used for inputting historical energy utilization behavior data into a pre-trained energy utilization behavior learning model of the user and predicting the energy utilization behavior of the user; the demand response calculation module is used for calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption demand of the user according to the electricity consumption behavior, so as to obtain the demand response potential of the user electricity consumption behavior; the pricing method generation module is used for inputting the demand response potential into a pre-built pricing model to obtain a pricing method.
Wherein the behavior prediction module 2 comprises: the device comprises a model construction unit, a feature selection unit, a feature processing unit and a deep learning unit; the model construction unit is used for pre-constructing an energy consumption behavior learning model for the user; the feature selection unit is used for selecting historical user behavior features and inputting a user behavior learning model, and the behavior features comprise: historical behavior, electricity price, temperature, humidity, wind power and time; the feature processing unit is used for carrying out feature processing on the user behavior feature by adopting a multi-class attention mechanism in the behavior learning model; the deep learning unit is used for deep learning the user behavior feature processed by the feature processing in the behavior learning model by adopting the stacked neural network, so as to obtain a trained user behavior learning model.
In one embodiment, the feature processing unit is configured to: is provided withIs a hidden state matrix of the LSTM layer, wherein the dimension m is the feature quantity, the dimension u is the time step, and H LSTM Middle->The i-th hidden state of the LSTM layer;
hidden state matrix for previous u-1 stepsPerforming one-dimensional convolution processing to extract a time mode matrix H in a convolution kernel range C ：
Wherein:is H C The ith row and jth column elements in H represent the ith row vector and jth convolution kernel +.>A processed result value; />Is a matrix after convolution operation; k is the number of convolution kernels; h i，q For the i-th row and q-th column element in H, wherein q=u-T-1+l is calculated in the convolution field of view; t is the convolution field of view; c j，l C is j The first element of (a);
a scoring mechanism s (·) is introduced to evaluate the hidden state h of the u-th step u With a convolution time model matrix H C Is a correlation of the row vectors of (a):
wherein:is H C The dimension K is the feature number of all convolution kernels after processing;mapping a matrix for attention in a scoring mechanism;
normalizing a scoring mechanism by using a Sigmoid (·) activation function to obtain a representative h u Andattention coefficient alpha of related information i ：
And according to the obtained attention coefficient, carrying out attention weighting and addition operation, and outputting the hidden state under the time mode attention mechanism:
wherein:and->Are all a learnable parameter matrix of the TPA layer; />The (u) th step hidden state after the LSTM layer and the TPA layer are treated contains m features.
Specifically, the electricity price characteristics are processed: obtaining zero-one index of corresponding electricity price data through a time sequence learning layer and a conventional attention layer of historical behavior data; performing dot multiplication operation on the zero-one index and the electricity price data, and removing low-importance electricity price data, wherein the method specifically comprises the following steps of:
given a givenFor a price package containing N time periods of electricity price on day t, for L t Go about p t Processing to obtain the hidden state of electricity price related to the behavior of energy consumption>The timing learning layer is expressed as a function LSTM (.):
attention index related to electricity price is obtained through conventional attention layer
Then through dot product operationGet the weighted electricity price data characteristic hidden state +.>
The processing of the temperature, humidity and wind power characteristics comprises the following steps:
given a givenAnd->And->Andthe time series data of the outdoor dry bulb temperature, the outdoor wet bulb temperature, the relative humidity of the outdoor air, the outdoor air humidity ratio, the wind speed and the wind direction are the t day, and the dimension corresponds to the time of electricity price; obtaining the data characteristic hidden state of the outdoor dry bulb temperature and the outdoor wet bulb temperature after attention weighting through a time sequence learning layer, a conventional attention layer and dot multiplication operationAnd->And->And->
The deep learning unit is used for splicing the processed electricity price, temperature, humidity and wind power data through the fusion layer, carrying out deep data mining through the stacked full-connection layer, learning the influence of different electricity price, temperature, humidity and wind power on the energy utilization behavior, and outputting the processed electricity price, temperature, humidity and wind power dataCo-dimensional mixed feature hidden state->Will->Superimposed on->And performing depth data mining again by stacking the full connection layers, and outputting user energy behaviors.
The ratio calculating method comprises the following steps:
wherein beta is t，k Time-varying DR potential for the user at the kth period on the t-th day; d, d t，k And Δd t，k The electricity demand of the user at the kth period of the t-th day and the electricity demand that can be cut down/diverted, respectively.
The pricing model is:
wherein,personalized retail electricity price packages representing the t-th day for decision variables; />For the user +.>Estimating the following energy utilization behavior; />The price of wholesale electricity generated for market clearing before the t day; />For the user +.>Estimating the following energy utilization behavior; θ (·) is the energy consumption behavior learning model after training; />For the weather forecast collection on day t, obtained from the weather forecast system, wherein +.>And->Is the outdoor dry bulb temperature and wet bulb temperature, < + >>And->For the outdoor air relative humidity and air humidity ratio, +.>And->Wind speed and direction; />And->Respectively->And->The kth element of (a); />Is a price change threshold vector in a pricing scheme.
After the pricing method is obtained, price risk hedging constraint is also carried out, and the price risk hedging constraint comprises the following steps:
wherein:a risk of revenue for the retailer on day t; />Is a cost risk to the user on day t.
Referring to fig. 7, an embodiment of the present application provides an electronic device, which includes: the system comprises a memory 601, a processor 602 and a computer program stored on the memory 601 and executable on the processor 602, wherein the processor 602 implements the smart pricing method for power retail described in the foregoing.
Further, the electronic device further includes: at least one input device 603 and at least one output device 604.
The memory 601, the processor 602, the input device 603, and the output device 604 are connected via a bus 605.
The input device 603 may be a camera, a touch panel, a physical key, a mouse, or the like. The output device 604 may be, in particular, a display screen.
The memory 601 may be a high-speed random access memory (RAM, random Access Memory) memory or a non-volatile memory (non-volatile memory), such as a disk memory. The memory 601 is used for storing a set of executable program codes and the processor 602 is coupled to the memory 601.
Further, the embodiment of the present application also provides a computer readable storage medium, which may be provided in the electronic device in each of the above embodiments, and the computer readable storage medium may be the memory 601 in the above embodiments. The computer readable storage medium has stored thereon a computer program which when executed by the processor 602 implements the smart pricing method for power retail as described in the previous embodiments.
Further, the computer-readable medium may be any medium capable of storing a program code, such as a usb (universal serial bus), a removable hard disk, a Read-Only Memory 601 (ROM), a RAM, a magnetic disk, or an optical disk.
In the several embodiments provided by the present application, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, and for example, the division of the modules is merely a logical function division, and there may be additional divisions when actually implemented, for example, multiple modules or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or modules, which may be in electrical, mechanical, or other forms.
The modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed over a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module. The integrated modules may be implemented in hardware or in software functional modules.
The integrated modules, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present application.
It should be noted that, for the sake of simplicity of description, the foregoing method embodiments are all expressed as a series of combinations of actions, but it should be understood by those skilled in the art that the present application is not limited by the order of actions described, as some steps may be performed in other order or simultaneously in accordance with the present application. Further, those skilled in the art will appreciate that the embodiments described in the specification are all preferred embodiments, and that the acts and modules referred to are not necessarily all required for the present application.
In the foregoing embodiments, the descriptions of the embodiments are emphasized, and for parts of one embodiment that are not described in detail, reference may be made to the related descriptions of other embodiments.
The foregoing describes an intelligent pricing method, system, electronic device and storage medium for power retail, and the present disclosure should not be construed as limiting the disclosure in view of the foregoing, as those skilled in the art will appreciate that modifications may be made in terms of specific embodiments and application scope in accordance with the teachings of the present disclosure.

Claims (10)

1. An intelligent pricing method for power retail, comprising:
acquiring historical energy consumption behavior data of a target user;
inputting the historical energy utilization behavior data into a pre-trained energy utilization behavior learning model of the user, and predicting the energy utilization behavior of the user;
calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption requirement of the user according to the energy consumption behavior, and obtaining the requirement response potential of the energy consumption behavior of the user;
and inputting the demand response potential into a pre-built pricing model to obtain a pricing method.
2. The intelligent pricing method for power retail sales of claim 1,
the training method of the behavior learning model for the user comprises the following steps:
pre-constructing a user energy consumption behavior learning model;
selecting historical user behavior characteristics and inputting the user behavior learning model, wherein the behavior characteristics comprise: historical behavior, electricity price, temperature, humidity, wind power and time;
performing feature processing on the user behavior feature by adopting a multi-class attention mechanism in the behavior learning model;
and deep learning the feature-processed user behavior features in the behavior learning model by adopting a stacked neural network to obtain a trained user behavior learning model.
3. The intelligent pricing method for power retail sales of claim 2,
the multiple types of attention mechanisms are time-mode attention mechanisms (temporal pattern attention, TPA) based on long short-term memory (LSTM), and the method extracts user historical behavior characteristics, and comprises the following steps:
is provided withIs a hidden state matrix of the LSTM layer, wherein the dimension m is the feature quantity, the dimension u is the time step, and H LSTM Middle->The i-th hidden state of the LSTM layer;
hidden state matrix for previous u-1 stepsPerforming one-dimensional convolution processing to extract a time mode matrix H in a convolution kernel range C ：
Wherein:is H C The ith row and jth column elements in H represent the ith row vector and jth convolution kernel +.>A processed result value; />Is a matrix after convolution operation; k is the number of convolution kernels; h i,q For the i-th row and q-th column element in H, wherein q=u-T-1+l is calculated in the convolution field of view; t is the convolution field of view; c j,l C is j The first element of (a);
a scoring mechanism s (·) is introduced to evaluate the hidden state h of the u-th step u With a convolution time model matrix H C Is a correlation of the row vectors of (a):
wherein:is H C The dimension K is the feature number of all convolution kernels after processing; />Mapping a matrix for attention in a scoring mechanism;
normalizing a scoring mechanism by using a Sigmoid (·) activation function to obtain a representative h u And h i C Attention coefficient alpha of related information i ：
And according to the obtained attention coefficient, carrying out attention weighting and addition operation, and outputting the hidden state under the time mode attention mechanism:
wherein:and->Are all a learnable parameter matrix of the TPA layer; />The (u) th step hidden state after the LSTM layer and the TPA layer are treated contains m features.
4. The intelligent pricing method for power retail sales of claim 2,
the multi-class attention mechanism is used for processing electricity price, temperature, humidity, wind power and time characteristics, and comprises the following steps:
processing electricity price characteristics: obtaining zero-one index of corresponding electricity price data through a time sequence learning layer and a conventional attention layer of historical behavior data; performing dot multiplication operation on the zero-one index and the electricity price data, and removing low-importance electricity price data, wherein the method specifically comprises the following steps of:
given a givenFor a price package containing N time periods of electricity price on day t, for L t Go about p t Processing to obtain the hidden state of electricity price related to the behavior of energy consumption>The timing learning layer is expressed as a function LSTM (.):
attention index related to electricity price is obtained through conventional attention layer
Then through dot product operationGet the weighted electricity price data characteristic hidden state +.>
The processing of the temperature, humidity and wind power characteristics comprises the following steps:
given a givenAnd->And->And->The time series data of the outdoor dry bulb temperature, the outdoor wet bulb temperature, the relative humidity of the outdoor air, the outdoor air humidity ratio, the wind speed and the wind direction are the t day, and the dimension corresponds to the time of electricity price; obtaining the hidden state of the data characteristics of the outdoor dry bulb temperature and the outdoor wet bulb temperature after attention weighting through a time sequence learning layer, a conventional attention layer and dot multiplication operation>Andand->And->
5. The intelligent pricing method for power retail sales of claim 4,
the deep learning includes:
the processed electricity price, temperature, humidity and wind power data are spliced through fusion layers, deep data mining is carried out through stacked full-connection layers, influences of different electricity prices, temperatures, humidity and wind power on energy utilization behaviors are learned, and output are carried outCo-dimensional mixed feature hidden state->Will->Superimposed on->And performing depth data mining again by stacking the full connection layers, and outputting user energy behaviors.
6. The intelligent pricing method for power retail sales of claim 1,
the calculating method of the ratio comprises the following steps:
wherein beta is t,k Time-varying DR potential for the user at the kth period on the t-th day; d, d t,k And Δd t,k The user's electricity demand during the kth period on the t-th day and the electricity demand that can be cut down/diverted,
the pricing model is as follows:
wherein,personalized retail electricity price packages representing the t-th day for decision variables; />For the user +.>Estimating the following energy utilization behavior; />The price of wholesale electricity generated for market clearing before the t day;for the user +.>Estimating the following energy utilization behavior; θ (·) is the energy consumption behavior learning model after training; />For the weather forecast collection on day t, obtained from the weather forecast system, wherein +.>And->Is the outdoor dry bulb temperature and wet bulb temperature, < + >>And->For the outdoor air relative humidity and air humidity ratio, +.>And->Wind speed and direction; />And->Respectively->And->The kth element of (a); />Is a price change threshold vector in a pricing scheme.
7. The intelligent pricing method for power retail sales of claim 1,
after the pricing method is obtained, price risk hedging constraint is also carried out, and the price risk hedging constraint comprises the following steps:
wherein:a risk of revenue for the retailer on day t; />Is a cost risk to the user on day t.
8. An intelligent pricing system for power retail, comprising:
the data acquisition module is used for acquiring historical energy utilization behavior data of the target user;
the behavior prediction module is used for inputting the historical energy consumption behavior data into a pre-trained energy consumption behavior learning model of the user and predicting the energy consumption behavior of the user;
the demand response calculation module is used for calculating the ratio of the electricity consumption change which can be cut down or transferred by the user to the original electricity consumption demand of the user according to the energy consumption behavior to obtain the demand response potential of the energy consumption behavior of the user;
and the pricing method generation module is used for inputting the demand response potential into a pre-constructed pricing model to obtain a pricing method.
9. An electronic device, comprising: a memory, a processor, on which a computer program is stored which is executable on the processor, characterized in that the processor, when executing the computer program, implements the method according to any one of claims 1 to 7.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any of claims 1 to 7.
CN202310838437.1A 2023-07-07 2023-07-07 Intelligent pricing method, system, electronic device and storage medium for electric power retail Pending CN117132334A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310838437.1A CN117132334A (en) 2023-07-07 2023-07-07 Intelligent pricing method, system, electronic device and storage medium for electric power retail

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310838437.1A CN117132334A (en) 2023-07-07 2023-07-07 Intelligent pricing method, system, electronic device and storage medium for electric power retail

Publications (1)

Publication Number Publication Date
CN117132334A true CN117132334A (en) 2023-11-28

Family

ID=88851684

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310838437.1A Pending CN117132334A (en) 2023-07-07 2023-07-07 Intelligent pricing method, system, electronic device and storage medium for electric power retail

Country Status (1)

Country Link
CN (1) CN117132334A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117410988A (en) * 2023-12-11 2024-01-16 广东领卓能源科技有限公司 Charging control method and device for new energy charging station

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117410988A (en) * 2023-12-11 2024-01-16 广东领卓能源科技有限公司 Charging control method and device for new energy charging station
CN117410988B (en) * 2023-12-11 2024-03-29 广东领卓能源科技有限公司 Charging control method and device for new energy charging station

Similar Documents

Publication Publication Date Title
Huang et al. A novel hybrid deep neural network model for short‐term electricity price forecasting
Ahmad et al. Utility companies strategy for short-term energy demand forecasting using machine learning based models
Homod et al. A novel hybrid modelling structure fabricated by using Takagi-Sugeno fuzzy to forecast HVAC systems energy demand in real-time for Basra city
Beccali et al. Short-term prediction of household electricity consumption: Assessing weather sensitivity in a Mediterranean area
Liu et al. Analysis and accurate prediction of user’s response behavior in incentive-based demand response
KR20200034023A (en) A method for creating demand response determination model for hvac system and a method for demand response
Li et al. Meta-learning strategy based on user preferences and a machine recommendation system for real-time cooling load and COP forecasting
CN117132334A (en) Intelligent pricing method, system, electronic device and storage medium for electric power retail
CN112418495A (en) Building energy consumption prediction method based on longicorn stigma optimization algorithm and neural network
CN112365056A (en) Electrical load joint prediction method and device, terminal and storage medium
Wang et al. Multi-criteria comprehensive study on predictive algorithm of heating energy consumption of district heating station based on timeseries processing
Sharif et al. Generation of whole building renovation scenarios using variational autoencoders
Sun et al. Data-driven occupant-behavior analytics for residential buildings
Wang et al. Coordinated residential energy resource scheduling with human thermal comfort modelling and renewable uncertainties
Wang et al. Modeling of regional electrical heating load characteristics considering user response behavior difference
Wang et al. An XGBoost-Based predictive control strategy for HVAC systems in providing day-ahead demand response
Sheikh et al. An integrated decision support system for multi-target forecasting: A case study of energy load prediction for a solar-powered residential house
Kumar et al. Forecasting indoor temperature for smart buildings with ARIMA, SARIMAX, and LSTM: A fusion approach
Fux et al. Short-term thermal and electric load forecasting in buildings
Kaur et al. Predicting power for home appliances based on climatic conditions
Nguyen et al. A machine learning-based approach for the prediction of electricity consumption
CN115759458A (en) Load prediction method based on comprehensive energy data processing and multi-task deep learning
CN115374908A (en) Short-term load prediction method and system suitable for flexible interconnection real-time control of distribution network
Ming et al. Typical commercial user adjustable potential assessment method
Wang Application of deep learning model in building energy consumption prediction

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination