Power prediction-scheduling integrated method based on unbalanced data clustering

Abstract

The application provides a power prediction-scheduling integrated method based on unbalanced data clustering, relates to the technical field of demand response scheduling and comprises the following steps: obtaining historical energy consumption data of users through a load aggregator, predicting according to the historical energy consumption data and obtaining corresponding power demand prediction values; obtaining dynamic electricity price data, sending the dynamic electricity price data and the power demand prediction values to each household energy management system through the load aggregator, and constructing a power distribution optimization model by the household energy management system; solving the power distribution optimization model to obtain an optimal power distribution scheme; and performing response scheduling of power utilization equipment according to the optimal power distribution scheme by the household energy management system. The application compares the predicted power demand with the optimal power distribution value, realizes the response scheduling of electrical appliances and achieves the purpose of reducing the peak-to-average ratio of community energy consumption.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of demand response scheduling, and more particularly to an electricity prediction-scheduling integrated method based on unbalanced data clustering. BACKGROUND

[0002] At present, with the development of smart grid, demand response (DR) as a key technology to adjust the user side power demand to match the supply capacity is increasingly concerned. The traditional DR strategy mainly relies on price incentives, direct load control and other ways to guide users to reduce electricity consumption during peak load period, or increase electricity consumption during low price period, so as to achieve the goal of peak load shifting and improving the efficiency of power grid operation. Since residential electricity consumption accounts for a high proportion in energy consumption structure, it has important theoretical exploration value and practical application prospect to carry out targeted and high-precision demand response strategy research around residential communities.

[0003] However, the uncertainty of electricity demand is one of the key problems faced by community side demand response. Through the analysis of user energy consumption data, electricity demand prediction can estimate the user demand budget in advance, providing key support for demand response strategy. However, in actual life, user energy consumption data often shows significant unbalanced characteristics, which will affect the accuracy of clustering analysis and prediction modeling. And the existing methods still have the problems of disconnection between prediction and scheduling, and strong dependence on centralized data.

[0004] In addition, according to the different communication and control modes between different subjects, the smart community demand response system mainly adopts two typical coordination architectures: centralized and distributed architecture. In the centralized method, the power operator or load aggregator needs to collect a large amount of user side electricity consumption data, appliance operating status, life behavior mode and other information, and centrally model and uniformly schedule. However, this mode not only brings huge data communication and processing pressure, but also has serious privacy leakage risk. On the other hand, although the distributed demand response architecture has developed in recent years, most methods still rely on strong central control signals or global information, and the user autonomy is insufficient, especially lacking specific applications combining the independent decision-making ability of home energy management system (HEMS)

[0005] Therefore, how to provide an electricity prediction-scheduling integrated method based on unbalanced data clustering to solve the above problems is a problem that those skilled in the art need to solve. SUMMARY

[0006] Therefore, the application provides an electricity prediction-scheduling integrated method based on unbalanced data clustering, a resident user only interacts with a load aggregator, a prediction model in the load aggregator performs data enhancement on a minority class in an electricity demand state obtained through DBSCAN (Density-Based Spatial Clustering of Applications with Noise) clustering based on a synthetic minority over-sampling technique (SMOTE), so that an electricity demand prediction value is obtained under the condition of data imbalance. A household energy management system on the user side obtains a user demand budget based on the predicted electricity demand and real-time electricity price information, and optimally allocates electricity of community residents by using a Lagrange-KKT condition with the goal of maximizing individual interests of residents. Community DR compares the predicted electricity demand with the optimal electricity allocation value, realizes response scheduling of electrical appliances, reduces user energy consumption cost overhead, and reduces a community energy consumption peak-to-average ratio.

[0007] In order to achieve the above object, the application adopts the following technical scheme:

[0008] The electricity prediction-scheduling integrated method based on unbalanced data clustering comprises the following steps:

[0009] S1: obtaining historical energy consumption data of a user through a load aggregator, and obtaining a corresponding electricity demand prediction value through prediction according to the historical energy consumption data;

[0010] S2: obtaining dynamic electricity price data, sending the dynamic electricity price data and the electricity demand prediction value to each household energy management system through the load aggregator, and constructing an electricity allocation optimization model by the household energy management system;

[0011] S3: solving the electricity allocation optimization model to obtain an optimal electricity allocation scheme;

[0012] S4: performing response scheduling of electrical equipment according to the optimal electricity allocation scheme by the household energy management system.

[0013] Preferably, the S1 specifically comprises:

[0014] S11: performing DBSCAN clustering processing on the historical energy consumption data, and abstracting the historical energy consumption data into different electricity demand states;

[0015] S12: performing data enhancement on a minority class in the electricity demand state by using a SMOTE oversampling method to obtain a corresponding balanced data set;

[0016] S13: using the DBSCAN method to identify the balance data set, and obtaining a corresponding energy demand state space;

[0017] S14: combining the energy demand state, the energy demand value and the exogenous time variable, training a model based on an expectation maximization algorithm, and obtaining a probability that the user n moves from an energy demand state i to a state j under a time state, and a probability that the user n observes an energy demand value o under an energy demand state i under a time state g;

[0018] S15: combining the result obtained in S14 to predict the energy demand state and the energy demand value.

[0019] Preferably, S2 specifically comprises:

[0020] S21: determining a user demand parameter according to the dynamic electricity price data and the energy demand prediction value;

[0021] S22: constructing an energy distribution optimization model with the user demand parameter as a target, and a specific expression is as follows:

[0022]

[0023] wherein, is a minimum total energy demand of the user, n is a maximum total energy demand of the user, represents distributed energy, n t represents a real-time electricity price, x represents a user demand price. RTP

[0024] Preferably, S3 specifically comprises:

[0025] Solving the energy distribution optimization model by using a Lagrange multiplier method and a KKT optimality condition, and obtaining an optimal energy distribution scheme in each time period in the family.

[0026] Preferably, S4 specifically comprises:

[0027] According to the solving result of S3 and the change of the dynamic electricity price data, the household energy management system controls the electrical equipment with a schedulable characteristic in the family, and realizes residential community demand response scheduling.

[0028] ​​Compared with the prior art, the application provides an electricity prediction-scheduling integrated method based on unbalanced data clustering, combines electricity demand prediction to perform early perception on future electricity demand budget of a family, and makes the scheduling strategy more forward-looking. In order to improve the accuracy of user behavior modeling, the application introduces an SMOTE-DBSCAN algorithm, performs unbalanced sample clustering analysis on original electricity data. The SMOTE algorithm is used to oversample rare behavior samples, the DBSCAN density clustering algorithm is used to classify users with high robustness, potential features and electricity modes of a small number of users are effectively mined, the demand prediction result is more personalized and comprehensive, and therefore more accurate input support is provided for subsequent budget estimation and optimized scheduling.

[0029] In the calculation process of the optimal electricity distribution, the model only depends on the real-time electricity price and user demand budget information provided by the load aggregator, and the optimal electricity distribution scheme is calculated by means of the Lagrange multiplier method and the KKT optimality condition, and the household energy management system responds to the optimal electricity distribution by scheduling the operation of household-level electrical appliances, so that the electricity bill expenditure of the user can be reduced.

[0030] The application uses a distributed complete dependence type architecture, and all optimization calculation and scheduling decisions are completed in the household energy management system at the user's local place. In this way, the household electricity data, electrical appliance state and user preference information do not need to be reported, and the risk of privacy leakage is eliminated from the root. In addition, the application is based on local data for fast solution, and has low calculation complexity, and can be deployed on a conventional household energy management system hardware platform. BRIEF DESCRIPTION OF DRAWINGS

[0031] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of the provided drawings.

[0032] Figure 1 A whole flowchart of an electricity prediction-scheduling integrated method based on unbalanced data clustering provided by the application is provided.

[0033] Figure 2 An electricity prediction-scheduling model architecture diagram provided by the embodiment of the application is provided.

[0034] Figure 3 A demand response function architecture diagram provided by the embodiment of the application is provided.

[0035] Figure 4 A real-time electricity price change diagram provided by the embodiment of the application is provided.

[0036] Figure 5 A before-and-after energy consumption comparison chart within a week is provided for the embodiment of the present application;

[0037] Figure 6 A before-and-after energy consumption cost and total amount comparison chart is provided for the embodiment of the present application. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0039] Embodiment 1

[0040] Referring to Figure 1 Embodiment 1 of the present application discloses an electricity prediction-scheduling integrated method based on unbalanced data clustering, comprising the following steps:

[0041] S1: obtaining historical energy consumption data of users through a load aggregator, and predicting to obtain corresponding electricity demand prediction values according to the historical energy consumption data;

[0042] S2: obtaining dynamic electricity price data, sending the dynamic electricity price data and the electricity demand prediction values to each household energy management system through the load aggregator, and constructing an electricity distribution optimization model by the household energy management system;

[0043] S3: solving the electricity distribution optimization model to obtain an optimal electricity distribution scheme;

[0044] S4: responding to scheduling of the household energy management system according to the optimal electricity distribution scheme.

[0045] In a specific embodiment, S1 specifically comprises:

[0046] S11: performing DBSCAN clustering processing on the historical energy consumption data, and abstracting the historical energy consumption data into different electricity demand states, wherein the historical energy consumption data can be historical energy consumption data of user n before t+1 time obtained by the load aggregator from a user smart meter;

[0047] S12: performing data enhancement on the minority class in the electricity demand state by using an SMOTE oversampling method to obtain a corresponding balanced data set;

[0048] S13: identifying the balanced data set by using a DBSCAN method to obtain a corresponding electricity demand state space , the demand state space defines the state set of the residents' electricity demand;

[0049] S14: Based on the electricity demand state, the electricity demand value and the exogenous time variable, the model is trained based on the expectation maximization algorithm to obtain the probability of user n transferring from energy demand state i to state j at time state (i.e., state transition probability ), and the probability of observing the energy demand value o for user n at time state g when the energy demand state is i (i.e., emission probability b n (o∣i,g) ), which also depends on the value of the exogenous variable z t . The state transition probability provides the maximum probability prediction of the next moment of electricity demand state, while the emission probability provides the maximum probability prediction of the next moment of electricity demand value;

[0050] S15: Combine the results obtained in S14 to predict the electricity demand state and the electricity demand value.

[0051] Specifically, the specific process of S11 includes:

[0052] S111: Collect historical energy consumption data of user n before t+1 from the smart meter ;

[0053] S112: Use the DBSCAN clustering method to cluster the historical energy consumption data , and abstract the observed energy consumption value into different electricity demand states.

[0054] The specific implementation process of the DBSCAN clustering method is as follows:

[0055] S1121: Randomly select a core point from the historical energy consumption data , and connect the points in its neighborhood to form a cluster .

[0056] S1122: For each newly added point in the cluster , the algorithm detects whether it is a core point. If the point is a core point, further core point determination is performed on the points in its neighborhood, and the points that meet the conditions are added to the cluster until it cannot be expanded any more. Then, the algorithm merges all samples in the cluster into the same energy demand state.

[0057] S1123: Remove all samples in the cluster from the historical energy consumption data . Repeat S1122 and S1123 until no more samples can be added to the existing cluster. Finally, after processing the historical energy consumption data After all samples in the cluster are classified, if there are samples that cannot be classified into any cluster, they are marked as noise points.

[0058] After clustering the historical energy consumption data using DBSCAN The observed energy consumption values are abstracted into different energy demand states.

[0059] The specific process of S12 includes:

[0060] To overcome the problem of DBSCAN clustering quality decline caused by data sparsity and uneven density distribution, the embodiment of the application uses SMOTE for data enhancement to balance data distribution and enhance density consistency. SMOTE can increase the data density of low-density areas (unconventional energy consumption data of users) by generating new synthetic samples in the data space, thereby improving the clustering effect and the accuracy of state space estimation. The specific implementation process of SMOTE data enhancement is as follows:

[0061] S121: Extract the frequency of occurrence of different energy demand states from the DBSCAN clustering results, and classify other states except the state with the highest frequency of occurrence as minority classes.

[0062] S122: In the minority state corresponding category, randomly select a sample , calculate its distance from other samples using Euclidean distance, and thus obtain its K nearest neighbors.

[0063] S123: Randomly select a sample from the K nearest neighbors of the sample , and generate a new sample using linear interpolation.

[0064] The generation formula of the new sample is as follows:

[0065] (1)

[0066] In the formula, δ is a linear interpolation random number, used to control the interpolation ratio, ranging from 0 to 1; is the synthetic data sample of user n in energy demand state s;

[0067] S124: Repeat steps S122-S123 for samples in each minority energy state until a sufficient number of synthetic samples are generated. Assume that the demand state with the highest frequency corresponds to class To ensure data balance, set the sampling rate for each minority class to , where S(·) represents the number of samples in a certain energy demand state.

[0068] S14 specifically includes: estimating parameters by EM algorithm , is the initial state distribution, E-step calculates the posterior probability of each pair of state transitions given the observation sequence, and M-step maximizes the expected log-likelihood to update the model parameters, where the model can be any neural network model.

[0069] S15 specifically comprises:

[0070] S151. Electric energy demand state prediction:

[0071] predicted electric energy demand state at time t+1 can be estimated by the state at time t and the time information z at time t+1 t+1 = g, the formula is expressed as:

[0072] (2)

[0073] In the formula, denotes the state transition probability;

[0074] S152. Electric energy demand value prediction:

[0075] predicted electric energy demand value at time t+1 can be estimated by the predicted electric energy demand state and the time information z at time t+1 t+1 . The emission probability is fitted using a Gaussian mixture distribution in the EM algorithm, i.e. , the electric energy demand prediction value can be generated based on the conditional expectation, and the specific expression is:

[0076] (3)

[0077] In the formula, is the weight of the kth Gaussian component, are the mean and variance, respectively, and the specific principle of the prediction model can be referred to as shown in Figure 2 , the prediction model uses historical energy consumption information and time information, and performs data enhancement on the DBSCAN algorithm based on the SMOTE oversampling algorithm, thereby obtaining the predicted electric energy demand, and the predicted electric energy demand data is specifically defined as the electric energy demand prediction value of a user in a certain period. In the prediction process, the resident user only needs to provide the overall energy consumption information of the family, without providing the electric appliance use information, thereby protecting the electricity privacy of the user. The DR scheduling model receives the electric energy demand prediction value from the prediction model, combines the real-time electricity price information, and performs electric energy distribution.

[0078] In a specific embodiment, S2 specifically comprises:

[0079] S21: Determine user demand parameters based on dynamic electricity price data and electricity demand forecasts. User demand parameters may include user demand budget, user satisfaction, and electricity cost.

[0080] S22: Construct an energy allocation optimization model with user demand parameters as the objective, the specific expression of which is:

[0081] (4)

[0082] in, For users n The minimum total power demand, For users n Maximum total electricity demand x Indicates the allocated electrical energy. RTP t Indicates real-time electricity price. Indicates the price requested by the user;

[0083] In designing the optimal power allocation objective, this invention aims to improve user satisfaction while reducing electricity costs. Therefore, maximizing individual user benefits can be expressed as the difference between the user's utility function and the electricity cost.

[0084] Specifically, S21 includes:

[0085] S211: Obtain dynamic electricity price data and predicted electricity demand data to calculate the user demand budget. The dynamic electricity price is specifically defined as the average of the Real-Time Pricing (RTP) price. The user demand budget (UDB) is defined as:

[0086] (5)

[0087] In the formula, For user n, the electricity cost budget for time period t. RTP avg This is the average of the real-time electricity price. The predicted electricity demand of user n during time period t;

[0088] According to UDB's definition, a user's budget is positively correlated with the projected electricity demand: for each residential user, the higher the projected electricity demand, the higher the electricity budget the user is willing to pay to meet future electricity needs.

[0089] S212. Calculate user satisfaction :

[0090] The electricity consumption satisfaction can be measured by the utility function. The utility function is used in economics to quantify the satisfaction a consumer obtains from consuming a particular good. A typical logarithmic utility function is used to measure the user's satisfaction with the allocated electricity, and its mathematical expression is:

[0091] (6)

[0092] In the formula, the utility function of user n at time period t is The electricity consumption cost is composed of two key parameters, the user demand price and the allocated electricity x, and is a strictly monotonic increasing function of and x. From the function characteristics, in the process of electricity allocation, the higher the user demand price, the greater the corresponding utility function value; similarly, the more electricity allocated, the utility function value also increases accordingly. In addition, this function relationship reflects the law of diminishing marginal utility of user demand for electricity, which is specifically manifested as the satisfaction of users brought by each unit of electricity increases gradually decreases with the increase of the amount of electricity allocated.

[0093] S213. Calculate the electricity consumption cost

[0094] The electricity consumption cost reflects the economic interests of the user. The cost of user n at time period t depends on the allocated electricity n and the real-time electricity price t , and its mathematical expression is: RTP

[0095] (7) In a specific embodiment, S3 specifically comprises:

[0096] The Lagrange multiplier method and the KKT optimality condition are used to solve the electricity allocation optimization model to obtain the optimal electricity allocation scheme of each time period in the family.

[0097] Specifically, S3 specifically comprises:

[0098] Let λ1 and λ2 be the Lagrange factors, and construct the Lagrange function for equation (4), and the specific expression is:

[0099]

[0100] (8) According to the KKT condition, the optimal solution must satisfy the following conditions:

[0101]

[0102]

[0103] ​With the above conditions, when x is within the constraint range, the constraint does not work, at this time λ1, λ2 is 0, and the optimal solution is When x is at the constraint boundary, the situation is different: if at this time λ2 is 0, and the optimal solution is ; if at this time λ1 is 0, and the optimal solution is .

[0104] Substituting equation (12) into the expression of the optimal solution, the optimal power distribution of user n at time period t can be summarized as:

[0105] (13)

[0106] The above formula shows that the DR model can reduce the distribution of power at high price periods and increase the distribution of power at low price periods by RTP avg / RTP t adjusting the optimal distribution of power dynamically.

[0107] In a specific embodiment, S4 specifically comprises:

[0108] According to the solution of S3 and the change of dynamic electricity price data, the household energy management system controls the electric equipment with schedulable characteristics in the household to realize residential community demand response scheduling, and the specific process can include the following steps:

[0109] In the optimal power distribution response mechanism, the model responds to the optimal power distribution through the intelligent electric appliance scheduling strategy. Specifically:

[0110] If the real-time electricity price of time period t is higher than the average electricity price, i.e. RTP avg / RTP t <1, it is considered that the electricity price of this period is too high for user n, and the demand flexibility is reduced. The distributed power is mainly used to respond to the demand of non-movable electric appliances, and the demand of flexible load (such as movable electric appliances) is transferred to the low electricity price period.

[0111] If the real-time electricity price of time period t is lower than the average electricity price, i.e. RTP avg / RTP t ≥1, it is considered that the electricity price of this period is relatively low for user n, and the demand flexibility is increased. The user can arrange some flexible load to run at this period, and the running power of controllable electric appliances can be flexibly conditioned to meet the power comfort of the user as much as possible. ​​

[0112] Figure 3 The specific architecture of the demand response function is shown, and the demand response scheduling algorithm runs on the HEMS. The HEMS on the market provides users with various functions, such as power demand feedback, remote switch control of electrical appliances, and flexible adjustment of electrical appliance energy consumption based on consumer preferences. Electrical appliances are divided into three categories, the first category is non-shiftable appliances (NSAs), which usually do not participate in demand response, and therefore are not controlled by the HEMS; the second category is shiftable appliances (SAs), which are directly switched by the HEMS, and the third category is controllable appliances (CAs), which are usually flexibly adjusted by the HEMS. The electrical equipment with schedulable characteristics refers to shiftable appliances and controllable appliances.

[0113] Shiftable appliances (SAs), also known as schedulable loads or flexible loads, refer to devices with flexible running time and delayed electrical energy demand. Such devices are intelligently controlled by the home energy management system, and when the system is in a power consumption peak period, their operation can be temporarily interrupted and restarted after the real-time electricity price falls, thereby realizing the flexibility of electrical energy demand adjustment. Taking typical household appliances as an example, for example, the working cycle of a washing machine includes multiple discrete stages, and each stage can be interrupted after the end of the running, and the running can be resumed after a few minutes to a few hours; similarly, as a cascading device, the dryer can be flexibly delayed for several hours according to the household power consumption state.

[0114] Controllable appliances (CAs), also known as thermal inertia loads, refer to electrical appliances with periodic working characteristics that are affected by the environment. The running state of such devices can be dynamically adjusted according to the preferences of end users, and their power demand is constrained between the specified minimum and maximum values during the electrical energy distribution process. CAs mainly include electric water heaters, air conditioning systems, and refrigeration devices, etc. with thermal storage characteristics. A common feature of these devices is that they have a certain thermal inertia, i.e. they can maintain their temperature state for a certain period of time, which allows their running state to be flexibly adjusted under the premise of ensuring user comfort.

[0115] The application also provides a system applying the power prediction-scheduling integrated method based on unbalanced data clustering according to any one of the above embodiments, comprising:

[0116] A prediction module is configured to obtain historical energy consumption data of a user through a load aggregator, and to obtain a corresponding power demand prediction value by predicting the historical energy consumption data.

[0117] a model construction module, configured to acquire dynamic electricity price data, and send the dynamic electricity price data and electricity demand prediction value to each household energy management system through a load aggregator, and the household energy management system constructs an electricity distribution optimization model;

[0118] a solution module, configured to solve the electricity distribution optimization model to obtain an optimal electricity distribution scheme;

[0119] a scheduling module, configured to cause the household energy management system to perform response scheduling of the electrical equipment according to the optimal electricity distribution scheme.

[0120] Embodiment 2

[0121] The detailed process of the method provided in Embodiment 1 is as follows:

[0122] I. Residential information in the simulation scenario and information of electrical appliances of residents Based on the database disclosed by Pecan Street. In order to verify the preliminary feasibility of the proposed DR scheduling algorithm, a simulation scenario is designed for analysis of a single typical residential user whose energy consumption is positively correlated with real-time electricity price. The specific information of the user is shown in Table 1, which includes 6 NSA, 2 CA and 4 SA. The maximum electricity demand is 16 Kwh, and the minimum electricity demand that must be met is 4.44 Kwh. The average value of the real-time electricity price is 0.065 .

[0123] Table 1 User information

[0124]

[0125] II. Introducing two indicators of electricity consumption cost and electricity consumption to evaluate the optimization performance of the DR algorithm

[0126] (1) Electricity consumption cost. The main reason why residential users agree to change the use time of household appliances to change the electricity consumption habit and thus transfer the electricity demand in the high electricity price period to the low electricity price period is that they expect to reduce the energy consumption cost by participating in the DR plan. The energy consumption cost of the user is:

[0127]

[0128] wherein, is the actual energy consumption value of the user n after the scheduling in period t, and T is the length of daily scheduling, which is 48 because the scheduling is performed every half hour.

[0129] (2) Energy Consumption. This comparison focuses on the total actual energy consumption of residents within one week before and after the DR model's scheduling. The actual energy consumption after scheduling is generated by the HEMS decision-making process. Changes in energy consumption are a key indicator of the feasibility of the DR strategy: a significant increase in energy consumption after scheduling indicates that the DR strategy has failed to achieve reasonable energy allocation and cannot effectively transfer energy demand over time; a significant decrease in energy consumption after scheduling means that the DR strategy is too aggressive and fails to meet users' electricity needs. Total Energy Consumption:

[0130]

[0131] III. This case study uses real-time electricity price data provided by PJM. For example... Figure 4 As shown, the horizontal axis of the graph is divided into half-hour increments, spanning from Monday 00:00-00:30 (marked as 1) to Sunday 23:30-24:00 (marked as 336), fully covering the week. The diagonal lines, cross lines, and horizontal lines in the graph represent three different price ranges: the low point (0.04...). -0.08 ), high point (0.08) -0.10 ) and peak (0.10) -0.14 These electricity price data that change over time can help visualize the effectiveness of the proposed DR algorithm.

[0132] Four, Figure 5 The changes in user energy consumption over time before and after scheduling were compared. As can be seen from the figure, the scheduling method provided by this invention achieves peak shaving and valley filling effects. In the original energy consumption curve, energy consumption is lower during off-peak hours (corresponding to periods of low electricity prices). After DR scheduling, some peak-hour electricity demand is shifted to off-peak hours, allowing more flexible electricity demand to be met during low-price periods.

[0133] V. In Figure 5 On this basis, Figure 6 The performance of the scheduling method provided by this invention was further quantified using data, comparing the user's electricity consumption cost and electricity consumption before and after scheduling. As shown in the figure, after scheduling, the electricity consumption cost decreased from 129.8 to 118.2, a reduction of 8.9%; simultaneously, electricity consumption also decreased, from 1814.3 to 1780.8, a reduction of 1.8%. This result indicates that the scheduling method provided by this invention can reduce electricity costs while basically meeting the user's electricity needs.

[0134] The various embodiments described in this specification are implemented in a progressive manner, each embodiment focusing on the differences from other embodiments, and the same or similar parts between embodiments can be mutually referred to. For the apparatus disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0135] The above description of disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A power prediction-scheduling integrated method based on unbalanced data clustering, characterized in that, The method comprises the following steps: S1: obtaining historical energy consumption data of users through a load aggregator, and performing prediction according to the historical energy consumption data to obtain corresponding predicted values of electric energy demand; S2: obtaining dynamic electricity price data, sending the dynamic electricity price data and the predicted values of electric energy demand to each household energy management system through the load aggregator, and constructing an electric energy distribution optimization model by the household energy management system, specifically comprising: S21: determining user demand parameters according to the dynamic electricity price data and the predicted values of electric energy demand; S22: constructing an electric energy distribution optimization model with the user demand parameters as the target, and the specific expression is: wherein, is the minimum total electrical energy demand of the user, n is the maximum total electrical energy demand of the user, is the minimum total electrical energy demand of the user, n is the maximum total electrical energy demand of the user, x denotes the allocated electrical energy, RTP t denotes the real-time electricity price, denotes the user demand price; S3: solving the electric energy distribution optimization model to obtain an optimal electric energy distribution scheme; S4: performing response scheduling of electric equipment according to the optimal electric energy distribution scheme by the household energy management system. 2.The method of claim 1, wherein, The S1 specifically comprises: S11: performing DBSCAN clustering processing on the historical energy consumption data, and abstracting the historical energy consumption data into different electric energy demand states; S12: performing data enhancement on the minority classes in the electric energy demand states by using an SMOTE oversampling method to obtain a corresponding balanced data set; S13: identifying the balanced data set by using a DBSCAN method to obtain a corresponding electric energy demand state space; S14: combining the electric energy demand state, the electric energy demand value and the exogenous time variable, training a model based on an expectation maximization algorithm to obtain a probability that user n transfers from an energy demand state i to a state j in a time state, and a probability that user n observes an energy demand value o in an energy demand state i in a time state g; S15: combining the results obtained in S14 to perform prediction of the electric energy demand state and the electric energy demand value. 3.The method of claim 1, wherein, The S3 specifically comprises: Solving the electric energy distribution optimization model by using a Lagrange multiplier method and a KKT optimality condition to obtain an optimal electric energy distribution scheme in each time period in the household. 4.The method of claim 1, wherein, The S4 specifically comprises: According to the solving result of S3 and the change of the dynamic electricity price data, the household energy management system controls the electric equipment with a schedulable characteristic in the household to realize residential community demand response scheduling.

Images