CN105117810A - Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism - Google Patents
Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism Download PDFInfo
- Publication number
- CN105117810A CN105117810A CN201510616219.9A CN201510616219A CN105117810A CN 105117810 A CN105117810 A CN 105117810A CN 201510616219 A CN201510616219 A CN 201510616219A CN 105117810 A CN105117810 A CN 105117810A
- Authority
- CN
- China
- Prior art keywords
- electricity
- electricity consumption
- mrow
- user
- prediction
- 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
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
本发明涉及一种阶梯电价机制下居民用电中期负载预测方法，首先，采集居民用电数据，提取出阶梯电价机制下居民用电行为的属性特征，通过聚类分析识别出阶梯电价机制下居民不同的用电行为特征，将具有相同或相似的用电行为特征的用户归聚为同一用户类别；然后，针对每个用户类别分别建立起对应的负载预测模型，并进行预测；最后，汇总各个用户类别的预测结果。本发明创新性地将阶梯电价相关的指标引入聚类模型中；利用智能电表所提供的更精确，更全面的数据，更好地进行中期负载预测。同时，相比传统的每15分钟数据采集，在保证精度要求的前提下，本方法利用较少的数据采样点(每日数据)进行中期预测。The invention relates to a mid-term load forecasting method for residents' electricity consumption under the ladder electricity price mechanism. First, the residents' electricity consumption data is collected, the attribute characteristics of the residents' electricity consumption behavior under the ladder electricity price mechanism are extracted, and the residents under the ladder electricity price mechanism are identified through cluster analysis. According to different electricity consumption behavior characteristics, users with the same or similar electricity consumption behavior characteristics are grouped into the same user category; then, a corresponding load forecasting model is established for each user category, and the prediction is made; finally, all Prediction results for user categories. The present invention innovatively introduces indexes related to ladder electricity prices into the clustering model; utilizes more accurate and comprehensive data provided by smart meters to perform better mid-term load forecasting. At the same time, compared with the traditional data collection every 15 minutes, this method uses fewer data sampling points (daily data) for medium-term forecasting under the premise of ensuring the accuracy requirements.
Description
Technical Field
The invention relates to the field of load prediction and data mining, in particular to a method for predicting loads in the middle period of residential electricity consumption under a stepped electricity price mechanism.
Background
Since the implementation of the step electricity price of the residents, the step electricity price of the residents has achieved primary effect, the energy-saving consciousness of the residents is enhanced to a certain extent, and some bad electricity using habits are changed. Under the traditional single low-price electricity price system, the difference of electricity consumption behaviors (not electricity consumption) among user groups with different characteristics is not obvious.
However, under the new step electricity price system, the user groups with different characteristics (such as income, family structure, living habits and the like) will generate different responses to the step electricity price, and the difference of the electricity utilization behaviors caused by the different responses will gradually be highlighted. This also makes load prediction at stepped rates more complicated. The accurate load prediction can not only ensure the safe and stable operation of the power grid system, but also reduce the power operation cost and improve the economic benefit and the social benefit.
The load prediction can be divided into short-term, medium-term and long-term according to the time domain. Short term generally refers to predictions of hours, days, to days, the middle term refers to predictions of weeks, months, and the long term refers to predictions of years, or even longer, in the future. Compared with short-term prediction, the medium-term and long-term load time span is long, the required basic data amount is large, the interference of various factors is easy, and prediction errors are accumulated to become unreliable.
The scientific theory basis of the step electricity price is to distinguish different characteristic user groups by a market segmentation method and adopt different electricity price mechanisms aiming at different user groups so as to improve the resource allocation efficiency. China is still in the initial stage of comprehensively implementing the stepped electricity price, researches on the aspect of analyzing the electricity consumption behavior under the stepped electricity price in China are rare, and great research blanks exist in theory and practice.
Since the mid-eighties, a large number of electric load prediction researches mainly based on various electric load prediction models and methods have been carried out by domestic and foreign scholars. However, most of the models are single load prediction models, mainly including regression analysis, time series, neural network, support vector machine, and the like. The single load prediction model captures the correlation among variables by analyzing key factors influencing the electricity consumption of the area, and a model is constructed for prediction. However, almost all single load prediction models simply sum up the electricity consumption of the users, and an average behavior model of the area is established by taking the total electricity consumption of the area as a target, and the characteristics of the electricity consumption behaviors of different types of users are ignored.
In the existing research, a user subdivision method is generally based on single variables such as household income, electricity consumption and the like, and a user group is simply divided into high-income, medium-income and low-income user groups. Meanwhile, the number of user groups needs to be determined in advance. For example, the families of residents are divided into four categories according to the annual income of the families; the monthly average power consumption of the users is used as a unique index, and the residential users are preliminarily classified into three types of users, namely low-income users, medium-income users and high-income users according to the number density of the users in the neighborhood of the monthly average power consumption. However, the monthly power consumption of the user is not constant, but often varies with the change of the air temperature and the season. Recent researches indicate that factors influencing power consumption are complex, and more abundant influencing variables need to be considered in user subdivision, such as subdivision variables of annual total power consumption, average power price, power consumption increase rate, variation coefficient, load rate, payment rate and the like, so that customers in the power industry are subdivided. However, the step price related variables have not been added to the user segmentation in the prior art. On the other hand, many studies have been made at home and abroad on the prediction of the power load, and the previous studies have always been made to predict the total amount of power consumption of users in a certain area as a target. The traditional gross prediction establishes an average behavior model of users, and the model cannot reveal different behaviors of each customer group and neglects the difference of the power utilization behaviors of different types of users. Therefore, efficient and accurate user classification and capturing of the law of electricity utilization behaviors of various users are two important aspects of urgent need for improvement of smart grid planning under a stepped electricity price mechanism.
In recent years, along with the popularization of intelligent electric meters, the existing resident real-time utilization and acquisition system can conveniently and quickly capture richer and more detailed resident real-time electricity utilization data under the stepped electricity price, and powerful data support is provided for identifying electricity utilization characteristics of different types of users.
At present, a common combined load prediction mode is to establish a plurality of different prediction models for the same input/output data set, then combine the prediction results, or perform weighted average according to appropriate weights, or adopt a more complex nonlinear combination model, and finally select a combination model with the best fitting degree or the minimum standard deviation.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a method for predicting the load in the middle period of residential electricity consumption under a stepped electricity price mechanism, which can accurately predict the load in the middle period and the long period.
The technical scheme of the invention is as follows:
a method for predicting the load of residents in the middle period of electricity utilization under a stepped electricity price mechanism comprises the steps of firstly, collecting resident electricity utilization data, extracting attribute characteristics of resident electricity utilization behaviors under the stepped electricity price mechanism, identifying different electricity utilization behavior characteristics of residents under the stepped electricity price mechanism through clustering analysis, and grouping users with the same or similar electricity utilization behavior characteristics into the same user category;
then, respectively establishing a corresponding load prediction model for each user category, and predicting;
and finally, summarizing the prediction results of all user categories.
Preferably, the resident electricity consumption data are divided into a plurality of classes according to different electricity consumption behavior characteristics to obtain a group of a plurality of data sets with different input and output, and then a corresponding load prediction model is established for each data set.
Preferably, after collecting the electricity consumption data of the residents, the following operations are performed:
1) data preprocessing: acquiring the daily electricity consumption of each user;
2) missing value processing: if the electricity consumption of a certain day is lost, calculating the difference of the accumulated electricity of the day before and after the electricity consumption loss time period, and averaging according to the lost days to be used as the lost electricity consumption of the certain day;
3) abnormal value processing: and filtering the daily electricity consumption exceeding the set index threshold range.
Preferably, during prediction, the user category to which the user to be predicted belongs is identified, then the corresponding load prediction model is selected for load prediction, and finally, the single prediction result is summarized to obtain the final overall prediction result.
Preferably, the attribute feature extraction of the residential electricity consumption behavior comprises cluster attribute selection and prediction input variable extraction; the clustering attributes comprise daily average power consumption of each user, a second step proportion, a third step proportion and high-temperature sensitivity; the predicted input variables include power usage over the past seven days, temperature of the day.
Preferably, the daily average power consumption is the total power consumption/total number of sampling days;
the second step ratio is the second step month/total month reached;
the third step proportion is the third step month/total month;
high temperature sensitivity-average daily power consumption/average daily power consumption.
Preferably, the clustering analysis is implemented by a fuzzy C-means clustering algorithm, each attribute feature of a user is affiliated to one or more user classes, and the degree of affiliation represents the degree of affiliation of the user to different user classes.
Preferably, the load prediction model is based on a self-organizing fuzzy neural network model and comprises an input layer, an ellipsoid basis function layer, a normalization layer, a weighted average layer and an output layer.
Preferably, the learning process of the self-organizing fuzzy neural network comprises parameter learning and structure learning;
the parameter learning enables the network to be converged quickly through an online recursion least square algorithm;
structure learning finds and selects neural network scales by automatically adding, modifying, or deleting self-organization of neurons in the ellipsoidal basis function layers.
Preferably, the structure learning includes the following operations:
1) (ii) an increase in neurons;
2) trimming neurons;
3) and merging the membership function and the fuzzy rule in the ellipsoid basis function layer.
The invention has the following beneficial effects:
the method of the invention is different from the traditional total load prediction model in the aspect of model mechanism; innovatively introducing indexes related to the stepped electricity prices into the clustering model; the mid-term load prediction is better performed by utilizing more accurate and comprehensive data provided by the intelligent electric meter. Meanwhile, compared with the traditional data acquisition of every 15 minutes, the method utilizes fewer data sampling points (daily data) to perform medium-term prediction on the premise of ensuring the precision requirement. However, the accurate medium-long term load is helpful to provide a series of decision support for developing the planning and construction of the intelligent power distribution network scientifically and reasonably, and is an important module for realizing the intelligent power grid.
The method provided by the invention takes a cluster analysis and load prediction algorithm as a core, provides a classified load prediction model, combines methods such as fuzzy C-means clustering (FCM) and self-organizing fuzzy neural network (SOFNN), and the like, can capture different characteristics of user electricity utilization behaviors under a stepped electricity price mechanism, and improves the overall middle-term load prediction precision. Accurate mid-term load prediction.
The classified load prediction model provided by the invention is different from the traditional combined prediction, and is characterized in that (1) in the classified load prediction model, each sub-model predicts a class of load, but the predicted input and output are different; (2) the prediction results of each submodel are only summarized, and the modeling process of linear weighting or nonlinear combination in the combined prediction model is not carried out; (3) the classified load prediction model can conveniently obtain two types of outputs, besides the traditional total amount prediction result, the prediction result of each electricity utilization type can be obtained, but the combined prediction model can only obtain the total amount prediction result.
Drawings
FIG. 1 is a schematic flow chart of the present invention (a framework including a load prediction model);
FIG. 2 is a basic framework of a self-organizing fuzzy neural network;
FIG. 3 is a graph comparing results of different categories of SSE and MIA index tests;
FIG. 4 is a clustered graph of residential electricity consumption;
FIG. 5 is a comparison of test set prediction results.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
The invention relates to a method for predicting the load of residents in the middle period of electricity consumption under a stepped electricity price mechanism, which comprises the following steps of firstly, acquiring resident electricity consumption data, extracting attribute characteristics of resident electricity consumption behaviors under the stepped electricity price mechanism, identifying different electricity consumption behavior characteristics of residents under the stepped electricity price mechanism through cluster analysis, and grouping users with the same or similar electricity consumption behavior characteristics into the same user category;
then, respectively establishing a corresponding load prediction model for each user category, and predicting;
and finally, summarizing the prediction results of all user categories.
The core of the method is a classified load prediction model, the resident electricity consumption data are firstly divided into a plurality of classes according to different electricity consumption behavior characteristics to obtain a group of a plurality of data sets with different input and output, and then a corresponding load prediction model is established for each data set. The classification load prediction model is used for classifying all information contained in a single load prediction model, and different prediction models are adopted for different power utilization behavior characteristics, so that a more accurate prediction result is provided. For example, there are two different user groups, which are a user group with high price sensitivity and a user group with low price sensitivity, and the classification prediction can let us know the accurate behavior of each user group, whereas the traditional average model of the total amount can cause the prediction of the user group with high price sensitivity to be too high, and the prediction of the user group with low price sensitivity to be too low.
After the resident electricity consumption data are collected, the following operations are carried out, and the data preparation stage is completed:
1) data preprocessing: acquiring the daily electricity consumption of each user;
2) missing value processing: if the electricity consumption of a certain day is lost, calculating the difference of the accumulated electricity of the day before and after the electricity consumption loss time period, and averaging according to the lost days to be used as the lost electricity consumption of the certain day;
3) abnormal value processing: and filtering the daily electricity consumption exceeding the set index threshold range.
Specifically, after the data preparation stage is completed, the electricity consumption behavior characteristics of different users in the analysis period are captured and identified by using a clustering algorithm, and users with the same or similar electricity consumption behavior characteristics are clustered into a group. And then classifying the residential electricity consumption data, namely, the input and the output of each data class are different. Then, the most suitable load prediction model is established for different types of users (i.e. different data sets) respectively.
When prediction is carried out, the user category to which the user to be predicted belongs is firstly identified, then a corresponding load prediction model is selected for load prediction, and finally, single prediction results are summarized to obtain a final overall prediction result.
The attribute feature of the residential electricity consumption behavior extraction mainly comprises clustering attribute selection and prediction input variable extraction.
Through data analysis and pre-experiments, four groups of cluster attributes of daily average power consumption, second step proportion, third step proportion and high-temperature sensitivity of each user are extracted to reflect the load change rule of residents in a certain time period.
The daily average power consumption mainly depends on the number of various electrical appliances owned by the household, whereby the income level of the residents can be estimated. And the electricity utilization behaviors of residents with similar income levels are similar.
The second step proportion and the third step proportion can reflect the fluctuation of the electricity utilization of the users in the past months and the reflection of the step electricity price mechanism, and capture the long-term electricity utilization rule of each user under the step electricity price mechanism. For example, a user who is sensitive to price may consciously reduce the power consumption when fast forwarding to the next step to avoid entering the next step, thereby reducing the total electricity price.
In addition, meteorological factors, especially temperature, often influence the change of resident's power consumption. Especially, the household appliances such as an air conditioner and the like are continuously used at high temperature in summer, the use frequency is high, and the daily electric quantity is usually greatly improved. The high-temperature sensitivity index reflects the fluctuation condition of the power consumption of the user in high-temperature weather.
The calculation method of the four groups of cluster attributes comprises the following steps:
the daily average power consumption is the total power consumption/total days of sampling;
the second step ratio is the second step month/total month reached;
the third step proportion is the third step month/total month;
high temperature sensitivity is the average daily power consumption/average daily power consumption of high temperature; the high temperature day defined in this embodiment means a day having an average temperature of 25 ℃ or higher.
Meanwhile, the input variables of all the load prediction models mainly extract eight input attributes of the power consumption of the past seven days and the temperature of the current day. Wherein applying historical load data is beneficial for prediction, as a rolling prediction approach may be used. And if the temperature is unknown and needs to be predicted, it can be estimated using weather forecast data or the average temperature of the same day over the past few years for the region.
The classification load prediction model provided by the invention has the characteristics of good universality and compatibility, is suitable for free combination of different clustering methods and load prediction models under the framework, and comprises a common clustering method (K mean value, a self-organizing feature mapping neural network) and a load prediction model (regression, time sequence and support vector machine). In addition, different user classes can adopt completely different load prediction models for load prediction, which is beneficial to adopting the most suitable load prediction model for each user class, and the flexibility and the prediction accuracy of the classified load prediction model are greatly improved.
In this embodiment, a fuzzy C-means clustering algorithm (FCM) and a self-organizing fuzzy neural network (SOFNN) model are combined to perform classified load prediction.
Compared with the traditional K-means clustering algorithm, the FCM adds a fuzzy concept, so that each input vector (attribute feature) is not only affiliated to a specific cluster any more, but also expresses the degree of belonging to different clusters according to the affiliation degree of the input vector. I.e. each attribute characteristic of a user is belonging to one or more user classes, with the degree of membership indicating the degree to which it belongs to different user classes.
In addition, the SOFNN has the advantages that: firstly, the method is simple and easy to use, and even if a user does not deeply know a fuzzy system and a neural network, the SOFNN can automatically determine the structure of a model and identify the parameters of the model; second, the prediction accuracy is higher.
The basic idea of FCM is to obtain the membership degree of each sample point to all the clustering centers by continuously optimizing an objective function, further determine the category of the sample points, and finally achieve the purpose of automatically clustering the sample data.
Assume that the sample set is Z ═ Z1，z2，…，zNAnd N is the total number of samples. Dividing the fuzzy clustering groups into C fuzzy clustering groups, and solving a clustering center set as V ═ V { (V)1，v2，…，vCDividing data by adopting the following optimized objective function according to the principle of least square method:
where m is the fuzzy regulation parameter, ucnIs the degree of membership of class c of the nth sample, and0≤ucn≤1，U＝[Ucn]is a matrix of dimensions C × N.
In addition, before fuzzy clustering, the extracted feature attributes need to be normalized, that is, the attribute values of the feature attributes are mapped to 0.1, so as to remove the influence of different magnitudes on the power consumption features of the users. Usually, a maximum and minimum value method is adopted to perform normalization processing on a data set, and the processing method is as follows:
in the formula zn' is the nth sample data normalized by the maximum minimum value method,andrespectively, a maximum value and a minimum value of the data sequence.
The load prediction model is based on a self-organizing fuzzy neural network model, and as shown in fig. 2, the SOFNN model comprises an input layer, an Ellipsoid Basis Function (EBF) layer, a normalization layer, a weighted average layer and an output layer. Wherein,
(1) a neuron i in the input layer 1, 2i；
(2) The neurons j 1, 2.. u in the EBF layer represent the precondition of a fuzzy rule, where each neuron multiplies the values of all membership functions as the output result ΦjThe specific algorithm is as follows:
in the formula, cijThe center of the membership function is represented,ijrepresenting the width of the membership function;
(3) the number of neurons in the normalization layer is generally the same as that of the EBF layer, and the corresponding output result ΨjComprises the following steps:
(4) the output of each neuron in the weighted average layer is obtained by multiplying the normalized layer output result by the weighted deviation w2The output result of the layer of neurons is fj＝w2Ψj。
(5) Each neuron in the output layer represents a variable obtained by summing up the output results in the weighted average layer, and therefore, the output result y of this layer is:
the learning process of the self-organizing fuzzy neural network (SOFNN) mainly comprises parameter learning and structure learning.
Parameter learning enables the network to converge quickly through an online recursive least squares algorithm.
Structure learning finds the most appropriate neural network size by automatically adding, modifying, or deleting the self-organization of neurons in the ellipsoidal basis function layer.
Structure learning mainly comprises three key steps:
1) (ii) an increase in neurons;
2) trimming neurons;
3) and merging the membership function and the fuzzy rule in the ellipsoid basis function layer.
Therefore, through parameter learning and structure learning algorithm in the SOFNN, an optimal network structure can be found for each cluster to predict.
Examples
In this embodiment, a 533-family resident in a certain area is used as a target to perform example analysis, since the meter reading date of the area is 11 days, the power consumption data from 11 days in 4 months to 10 days in 2015 as a training set, and the power consumption data from 11 days in 1 month to 10 days in 2015 as a test set are used for performing experiments.
The data preparation phase mainly comprises the following steps:
(1) data preprocessing: because the collected resident electricity consumption data recorded by the intelligent electric meter are all integrated values, the electricity consumption of each user on the day needs to be calculated by subtracting the electricity integrated value of the previous day from the electricity integrated value of the current day;
(2) missing value processing: after preprocessing, whether the data is missing or not needs to be detected. The method comprises the steps of calculating the difference of accumulated electric quantity of a day before and after a power consumption deficiency time period, averaging according to the number of days of deficiency, and filling up the deficient data;
(3) abnormal value processing: and filtering the data samples exceeding the set index threshold range, for example: street lamps, industrial users, and long-term, unsupervised residents, etc.
FCM clustering analysis
The traditional FCM algorithm needs a user to determine the number of clusters in advance. In the cluster analysis, the determination of the number of clusters has a great influence on the clustering result. In order to objectively determine the appropriate cluster number, the embodiment finds out the optimal cluster number parameter through comparative analysis mainly by calculating the Sum of Squared Error (SSE) and meanlndexadequacy (mia) value of each heuristic class number. The SSE and MIA calculation is as follows:
wherein n iscDenotes the number of sample data in class c, zckRepresenting the kth sample in class c. And (3) selecting the optimal clustering number according to the index change trend, and obtaining a clustering result, wherein when the number of the clusters exceeds 6, the curve is flatter and flatter along with the increase of the number of the clusters, and the reduction trend of the SSE index value and the MIA index value is obviously weakened as shown in figure 3.
Meanwhile, in order to ensure that each cluster center has a certain number of samples, the number of clusters is set to 6, and the clustering result is shown in table 1.
TABLE 1 results of six Cluster centers
| Clustering | Number of | Daily average power consumption | Second step ratio | Third step ratio | Sensitivity to high temperature |
| 1 | 75 | 0.0572 | 0.0277 | 0.0046 | 0.3535 |
| 2 | 113 | 0.1616 | 0.3794 | 0.0931 | 0.6148 |
| 3 | 81 | 0.1965 | 0.8253 | 0.0712 | 0.4212 |
| 4 | 118 | 0.2944 | 0.5457 | 0.4166 | 0.5946 |
| 5 | 89 | 0.3384 | 0.3976 | 0.5485 | 0.5988 |
| 6 | 57 | 0.4695 | 0.0645 | 0.9272 | 0.5224 |
By observing the numerical value of the K-means clustering center and combining the sample characteristics of each type, the following characteristics can be summarized:
(1) the average electricity consumption of the first type of users is the least, the electricity consumption of each type is increased in sequence, and the average electricity consumption of the sixth type of users is the most.
(2) The electricity consumption of the first class of users basically stays within the electricity consumption specified by the first step electricity price, and the electricity consumption reaching the second step and the third step is little; the electricity consumption of the second type of users is evenly distributed in the electricity consumption specified by the first and second elevator prices; the electricity consumption of the third type of users basically stays in the electricity consumption specified by the second step electricity price, and the electricity consumption of the first step and the third step is very small; the users in the fourth and fifth classes stay in the second and third steps. Wherein, the electricity consumption of the fourth type of users falls on the second ladder more often, and the electricity consumption of the fifth type of users falls on the third ladder more often; the electricity consumption of the sixth type of users basically stays within the electricity consumption specified by the third step of electricity price, and the electricity consumption of the first step and the second step is very small.
(3) It can be seen that the electricity consumption of the first, third and sixth users cannot be reduced ideally because the electricity consumption of the users is stable and almost constant within a certain electricity consumption step, and the fine adjustment of the electricity price of the users cannot cause the significant change of the electricity consumption. For the second, fourth and fifth types of users, the influence of the step electricity rates on the users may be obvious, because the electricity consumption of the three types of users is distributed in different step electricity rates, and step spans generally occur. When the electricity consumption of one step jumps to the electricity consumption of another step, the users with high price sensitivity can correspondingly reduce the electricity consumption according to the assumption of the rational economics scholars, and more electricity consumption is prevented from being used under higher electricity price.
Meanwhile, drawing out the electricity consumption curve clusters of the residents in each cluster according to the original data set, as shown in fig. 4, it can be found that the electricity consumption of the users in the category 1 is very small and very average, is basically concentrated below 10kWh, and is insensitive to high-temperature weather; users in category 3 are also not sensitive to high temperature weather; and the power consumption of other categories is obviously increased in summer, the corresponding high-temperature sensitivity indexes are high, and users in the categories are sensitive to high-temperature weather.
SOFNN load prediction
And summing the residential electricity consumption data according to the clustering results, and building corresponding SOFNN models respectively. Through multiple simulation experiments and tracking error correction, the estimated better SOFNN pre-experiment parameter is 0.01, sigma00.1, krmse 0.01 and kd (i) 0.01(i 1, 2, 8).
The accuracy of the prediction is measured by the commonly used Mean Absolute Percent Error (MAPE) measurement and the maximum absolute percent error (ME), and is calculated as follows:
wherein, ynAndthe actual value and the predicted value of the total power consumption of the area are shown,indicating the number of days predicted.
The rolling error of the model corresponding to each cluster is shown in table 2.
TABLE 2 Rolling prediction error of model corresponding to clustering
| Error index | Model 1 | Model 2 | Model 3 | Model 4 | Model 5 | Model 6 |
| MAPE | 22.87％ | 7.27％ | 5.14％ | 3.86％ | 2.79％ | 3.29％ |
| ME | 75.34％ | 17.43％ | 21.34％ | 9.80％ | 14.25％ | 8.90％ |
In general, the rolling prediction error of each model is within an acceptable range except for model 1. The reason is that some users exist in the cluster 1 corresponding to the model 1, the randomness of the electricity utilization behaviors of the users is particularly strong, and the electricity utilization rules of the users are difficult to distinguish. In the classification prediction model, the electricity utilization behaviors of the users are greatly different from those of other regular users, and the users are classified into one class. However, due to the characteristics of strong randomness and the like, the electricity utilization behavior of the users cannot be predicted accurately. Fortunately, the proportion of the power consumption predicted by the model 1 relative to the total consumption is small, and the prediction accuracy of the classification prediction model is not greatly influenced.
And summarizing the prediction results of the models, and obtaining the final classified load prediction model result as shown in fig. 5. MAPE values of the single load prediction model and the classification load prediction model are respectively 3.34% and 2.78%, and are both within 4%, and the precision result is satisfactory. Compared with a single load prediction model, the overall prediction precision of the classified load prediction model is improved by 0.56%. In addition, at the time of day 14 (i.e., day 1/25 of 2015), the prediction accuracy of the two models simultaneously drops to 90% or less, and the relevant historical data findings are reviewed, because the actual power consumption suddenly drops to a large extent due to regional blackouts. For such an emergency, it is difficult for the prediction model to respond to and correct it in time, which affects the overall prediction accuracy to some extent.
The above examples are provided only for illustrating the present invention and are not intended to limit the present invention. Changes, modifications, etc. to the above-described embodiments are intended to fall within the scope of the claims of the present invention as long as they are in accordance with the technical spirit of the present invention.
Claims (10)
1. A method for predicting the load of the middle period of residential electricity consumption under the mechanism of stepped electricity prices is characterized in that,
firstly, collecting resident electricity consumption data, extracting attribute characteristics of resident electricity consumption behaviors under a step electricity price mechanism, identifying different electricity consumption behavior characteristics of residents under the step electricity price mechanism through clustering analysis, and grouping users with the same or similar electricity consumption behavior characteristics into the same user category;
then, respectively establishing a corresponding load prediction model for each user category, and predicting;
and finally, summarizing the prediction results of all user categories.
2. The method according to claim 1, wherein the residential electricity load prediction method in the middle stage under the stepped electricity price mechanism is characterized in that the residential electricity data is divided into a plurality of classes according to different electricity consumption behavior characteristics to obtain a group of a plurality of data sets with different inputs and outputs, and then a corresponding load prediction model is established for each data set.
3. The method for predicting the load of the middle stage of residential electricity consumption under the stepped electricity price mechanism according to claim 1, wherein the following operations are performed after the data of the residential electricity consumption is collected:
1) data preprocessing: acquiring the daily electricity consumption of each user;
2) missing value processing: if the electricity consumption of a certain day is lost, calculating the difference of the accumulated electricity of the day before and after the electricity consumption loss time period, and averaging according to the lost days to be used as the lost electricity consumption of the certain day;
3) abnormal value processing: and filtering the daily electricity consumption exceeding the set index threshold range.
4. The method according to claim 1, wherein in the prediction, a user category to which a user to be predicted belongs is identified, then a corresponding load prediction model is selected for load prediction, and finally, a single prediction result is summarized to obtain a final overall prediction result.
5. The method for predicting the load in the middle period of residential electricity consumption under the stepped electricity price mechanism according to claim 1, wherein the extracting of the attribute features of the behavior of residential electricity consumption comprises clustering attribute selection and prediction input variable extraction; the clustering attributes comprise daily average power consumption of each user, a second step proportion, a third step proportion and high-temperature sensitivity; the predicted input variables include power usage over the past seven days, temperature of the day.
6. The method for predicting the load in the middle stage of residential electricity consumption under the stepped electricity rate scheme according to claim 5,
the daily average power consumption is the total power consumption/total days of sampling;
the second step ratio is the second step month/total month reached;
the third step proportion is the third step month/total month;
high temperature sensitivity-average daily power consumption/average daily power consumption.
7. The method according to claim 1, wherein the clustering analysis is performed by a fuzzy C-means clustering algorithm, each attribute feature of a user belongs to one or more user categories, and the degree of membership indicates the degree of the user belonging to different user categories.
8. The method for predicting the load in the middle period of residential electricity consumption under the stepped electricity price mechanism according to claim 1, wherein the load prediction model is based on a self-organizing fuzzy neural network model and comprises an input layer, an ellipsoid basis function layer, a normalization layer, a weighted average layer and an output layer.
9. The method for predicting the load in the middle period of residential electricity consumption under the stepped electricity price mechanism according to claim 8, wherein the learning process of the self-organizing fuzzy neural network comprises parameter learning and structure learning;
the parameter learning enables the network to be converged quickly through an online recursion least square algorithm;
structure learning finds and selects neural network scales by automatically adding, modifying, or deleting self-organization of neurons in the ellipsoidal basis function layers.
10. The residential electricity intermediate load prediction method under the stepped electricity rate scheme according to claim 9, wherein the structure learning includes the operations of:
1) (ii) an increase in neurons;
2) trimming neurons;
3) and merging the membership function and the fuzzy rule in the ellipsoid basis function layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510616219.9A CN105117810A (en) | 2015-09-24 | 2015-09-24 | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510616219.9A CN105117810A (en) | 2015-09-24 | 2015-09-24 | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105117810A true CN105117810A (en) | 2015-12-02 |
Family
ID=54665792
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510616219.9A Pending CN105117810A (en) | 2015-09-24 | 2015-09-24 | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105117810A (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106160232A (en) * | 2016-08-20 | 2016-11-23 | 国网山东滨州市沾化区供电公司 | A kind of load monitoring equipment for power system |
| CN106296465A (en) * | 2016-08-23 | 2017-01-04 | 四川大学 | A kind of intelligent grid exception electricity consumption behavioral value method |
| CN106295877A (en) * | 2016-08-09 | 2017-01-04 | 四川大学 | A kind of intelligent grid electric energy usage amount Forecasting Methodology |
| CN106570581A (en) * | 2016-10-26 | 2017-04-19 | 东北电力大学 | Attribute association based load prediction system and method in energy Internet environment |
| CN107316093A (en) * | 2016-04-26 | 2017-11-03 | 华为技术有限公司 | The method and device of a kind of rolling forecast |
| CN107515908A (en) * | 2017-08-11 | 2017-12-26 | 新智数通（北京）技术服务有限公司 | A kind of data processing method and device |
| CN108363343A (en) * | 2018-05-17 | 2018-08-03 | 广东美的制冷设备有限公司 | Household electric appliance control method, system and computer storage media |
| CN109063922A (en) * | 2018-08-21 | 2018-12-21 | 国网湖北省电力有限公司鄂州供电公司 | A kind of distribution transformer heavy-overload prediction technique based on cell occupancy rate |
| CN109284851A (en) * | 2018-06-11 | 2019-01-29 | 西安交通大学 | A classification method of consumer electricity behavior suitable for demand-side response |
| CN109461019A (en) * | 2018-09-21 | 2019-03-12 | 华南理工大学 | A kind of dynamic need response pricing method based on Fuzzy Reinforcement Learning |
| CN109543908A (en) * | 2018-11-26 | 2019-03-29 | 国家电网有限公司 | A kind of prediction technique of the electricity consumption of resident based on resident's household electrical appliance accounting energy |
| TWI663570B (en) * | 2017-10-20 | 2019-06-21 | 財團法人資訊工業策進會 | Power consumption analyzing server and power consumption analyzing method thereof |
| CN110533256A (en) * | 2019-09-05 | 2019-12-03 | 天津华春智慧能源科技发展有限公司 | Method for predicting heat supply load on demand side |
| CN110782074A (en) * | 2019-10-09 | 2020-02-11 | 深圳供电局有限公司 | Method for predicting user power monthly load based on deep learning |
| CN111141950A (en) * | 2019-12-03 | 2020-05-12 | 深圳供电局有限公司 | A method for judging and repairing abnormality of time-sharing meter code of metering automation system |
| CN111177651A (en) * | 2019-12-03 | 2020-05-19 | 深圳供电局有限公司 | Time-sharing missing code fitting method for electric meter of metering automation system |
| CN111210260A (en) * | 2019-12-25 | 2020-05-29 | 深圳供电局有限公司 | Electricity price data processing method, device, computer equipment and storage medium |
| CN111553586A (en) * | 2020-04-26 | 2020-08-18 | 南方电网科学研究院有限责任公司 | Energy-saving strategy pushing system, method and equipment based on step electricity price |
| CN111786824A (en) * | 2020-06-23 | 2020-10-16 | 中国电力科学研究院有限公司 | Data center energy efficiency ratio optimization method, system, device and readable storage medium |
| CN111815060A (en) * | 2020-07-16 | 2020-10-23 | 广东电网有限责任公司 | A method and device for short-term load forecasting in electricity consumption areas |
| CN111915116A (en) * | 2019-05-10 | 2020-11-10 | 国网能源研究院有限公司 | Electric power resident user classification method based on K-means clustering |
| CN112036777A (en) * | 2020-11-05 | 2020-12-04 | 江苏智臻能源科技有限公司 | Determination method for measuring and calculating user types based on power consumption of air conditioner and water heater |
| CN112561193A (en) * | 2020-12-23 | 2021-03-26 | 上海亿边科技有限公司 | AI intelligent power load prediction system based on FPGA inference acceleration |
| WO2021077977A1 (en) * | 2019-10-23 | 2021-04-29 | 华能大理风力发电有限公司 | Method for analyzing transaction data in wind power bidding market, device, apparatus, and medium |
| CN112836886A (en) * | 2021-02-09 | 2021-05-25 | 国网山西省电力公司吕梁供电公司 | An energy service method, device, computer equipment and storage medium |
| CN113868481A (en) * | 2021-10-25 | 2021-12-31 | 联想(北京)有限公司 | Component acquisition method, device, electronic device and storage medium |
| CN114091480A (en) * | 2021-11-18 | 2022-02-25 | 北京理工大学 | An ensemble method for machine translation quality estimation using multiple pre-trained models |
| CN114511360A (en) * | 2022-04-20 | 2022-05-17 | 华能江西能源销售有限责任公司 | Electricity purchase management method, system, readable storage medium and computer equipment |
| CN114742153A (en) * | 2022-04-08 | 2022-07-12 | 国家电网有限公司大数据中心 | Power utilization behavior analysis method based on one graph of power distribution network |
| CN115809942A (en) * | 2023-02-06 | 2023-03-17 | 四川智源能诚售电有限公司 | Power consumption data prediction method, device and storage medium |
| CN117932311A (en) * | 2024-03-21 | 2024-04-26 | 杭州可当科技有限公司 | Intelligent user identification method of intelligent internet terminal based on 5G network |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
-
2015
- 2015-09-24 CN CN201510616219.9A patent/CN105117810A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008430A (en) * | 2014-05-29 | 2014-08-27 | 华北电力大学 | Method for establishing virtual reality excavation dynamic smart load prediction models |
Non-Patent Citations (6)
| Title |
|---|
| LENG G 等: ""An on-line algorithm for creating self-organizing fuzzy neural networks"", 《NEURAL NETWORKS》 * |
| LIN B Q 等: ""Designation and influence of household increasing block electricity tariffs in China"", 《ENERGY POLICY》 * |
| 刘继东: ""电力需求侧响应的效益评估与特性分析"", 《中国博士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 徐娟 等: ""基于负荷分类的短期负荷预测方法分析及应用"", 《河北电力技术》 * |
| 徐娟: ""基于负荷分类的短期负荷预测方法研究"", 《万方数据知识服务平台 学位论文》 * |
| 李培强 等: ""基于模糊聚类的电力负荷特性的分类与综合"", 《中国电机工程学报》 * |
Cited By (38)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107316093B (en) * | 2016-04-26 | 2021-01-05 | 华为技术有限公司 | Rolling prediction method and device |
| CN107316093A (en) * | 2016-04-26 | 2017-11-03 | 华为技术有限公司 | The method and device of a kind of rolling forecast |
| CN106295877A (en) * | 2016-08-09 | 2017-01-04 | 四川大学 | A kind of intelligent grid electric energy usage amount Forecasting Methodology |
| CN106160232A (en) * | 2016-08-20 | 2016-11-23 | 国网山东滨州市沾化区供电公司 | A kind of load monitoring equipment for power system |
| CN106296465B (en) * | 2016-08-23 | 2020-05-08 | 四川大学 | A method for detecting abnormal power consumption behavior of smart grid |
| CN106296465A (en) * | 2016-08-23 | 2017-01-04 | 四川大学 | A kind of intelligent grid exception electricity consumption behavioral value method |
| CN106570581B (en) * | 2016-10-26 | 2019-06-28 | 东北电力大学 | Load prediction system and method under energy internet environment based on Attribute Association |
| CN106570581A (en) * | 2016-10-26 | 2017-04-19 | 东北电力大学 | Attribute association based load prediction system and method in energy Internet environment |
| CN107515908A (en) * | 2017-08-11 | 2017-12-26 | 新智数通（北京）技术服务有限公司 | A kind of data processing method and device |
| TWI663570B (en) * | 2017-10-20 | 2019-06-21 | 財團法人資訊工業策進會 | Power consumption analyzing server and power consumption analyzing method thereof |
| CN108363343A (en) * | 2018-05-17 | 2018-08-03 | 广东美的制冷设备有限公司 | Household electric appliance control method, system and computer storage media |
| CN109284851A (en) * | 2018-06-11 | 2019-01-29 | 西安交通大学 | A classification method of consumer electricity behavior suitable for demand-side response |
| CN109063922A (en) * | 2018-08-21 | 2018-12-21 | 国网湖北省电力有限公司鄂州供电公司 | A kind of distribution transformer heavy-overload prediction technique based on cell occupancy rate |
| CN109063922B (en) * | 2018-08-21 | 2021-04-09 | 国网湖北省电力有限公司鄂州供电公司 | Power distribution transformer overload prediction method based on cell survival rate |
| CN109461019A (en) * | 2018-09-21 | 2019-03-12 | 华南理工大学 | A kind of dynamic need response pricing method based on Fuzzy Reinforcement Learning |
| CN109543908A (en) * | 2018-11-26 | 2019-03-29 | 国家电网有限公司 | A kind of prediction technique of the electricity consumption of resident based on resident's household electrical appliance accounting energy |
| CN111915116A (en) * | 2019-05-10 | 2020-11-10 | 国网能源研究院有限公司 | Electric power resident user classification method based on K-means clustering |
| CN110533256A (en) * | 2019-09-05 | 2019-12-03 | 天津华春智慧能源科技发展有限公司 | Method for predicting heat supply load on demand side |
| CN110782074A (en) * | 2019-10-09 | 2020-02-11 | 深圳供电局有限公司 | Method for predicting user power monthly load based on deep learning |
| WO2021077977A1 (en) * | 2019-10-23 | 2021-04-29 | 华能大理风力发电有限公司 | Method for analyzing transaction data in wind power bidding market, device, apparatus, and medium |
| CN111177651A (en) * | 2019-12-03 | 2020-05-19 | 深圳供电局有限公司 | Time-sharing missing code fitting method for electric meter of metering automation system |
| CN111141950B (en) * | 2019-12-03 | 2022-06-14 | 深圳供电局有限公司 | A method for judging and repairing abnormality of time-sharing meter code of metering automation system |
| CN111141950A (en) * | 2019-12-03 | 2020-05-12 | 深圳供电局有限公司 | A method for judging and repairing abnormality of time-sharing meter code of metering automation system |
| CN111210260A (en) * | 2019-12-25 | 2020-05-29 | 深圳供电局有限公司 | Electricity price data processing method, device, computer equipment and storage medium |
| CN111553586A (en) * | 2020-04-26 | 2020-08-18 | 南方电网科学研究院有限责任公司 | Energy-saving strategy pushing system, method and equipment based on step electricity price |
| CN111786824A (en) * | 2020-06-23 | 2020-10-16 | 中国电力科学研究院有限公司 | Data center energy efficiency ratio optimization method, system, device and readable storage medium |
| CN111815060A (en) * | 2020-07-16 | 2020-10-23 | 广东电网有限责任公司 | A method and device for short-term load forecasting in electricity consumption areas |
| CN112036777A (en) * | 2020-11-05 | 2020-12-04 | 江苏智臻能源科技有限公司 | Determination method for measuring and calculating user types based on power consumption of air conditioner and water heater |
| CN112561193A (en) * | 2020-12-23 | 2021-03-26 | 上海亿边科技有限公司 | AI intelligent power load prediction system based on FPGA inference acceleration |
| CN112836886A (en) * | 2021-02-09 | 2021-05-25 | 国网山西省电力公司吕梁供电公司 | An energy service method, device, computer equipment and storage medium |
| CN113868481A (en) * | 2021-10-25 | 2021-12-31 | 联想(北京)有限公司 | Component acquisition method, device, electronic device and storage medium |
| CN114091480A (en) * | 2021-11-18 | 2022-02-25 | 北京理工大学 | An ensemble method for machine translation quality estimation using multiple pre-trained models |
| CN114742153A (en) * | 2022-04-08 | 2022-07-12 | 国家电网有限公司大数据中心 | Power utilization behavior analysis method based on one graph of power distribution network |
| CN114511360A (en) * | 2022-04-20 | 2022-05-17 | 华能江西能源销售有限责任公司 | Electricity purchase management method, system, readable storage medium and computer equipment |
| CN114511360B (en) * | 2022-04-20 | 2022-07-08 | 华能江西能源销售有限责任公司 | Electricity purchase management method, system, readable storage medium and computer equipment |
| CN115809942A (en) * | 2023-02-06 | 2023-03-17 | 四川智源能诚售电有限公司 | Power consumption data prediction method, device and storage medium |
| CN117932311A (en) * | 2024-03-21 | 2024-04-26 | 杭州可当科技有限公司 | Intelligent user identification method of intelligent internet terminal based on 5G network |
| CN117932311B (en) * | 2024-03-21 | 2024-05-31 | 杭州可当科技有限公司 | Intelligent user identification method of intelligent internet terminal based on 5G network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105117810A (en) | Residential electricity consumption mid-term load prediction method under multistep electricity price mechanism | |
| CN110097297B (en) | Multi-dimensional electricity stealing situation intelligent sensing method, system, equipment and medium | |
| WO2022135265A1 (en) | Failure warning and analysis method for reservoir dispatching rules under effects of climate change | |
| CN116227637B (en) | Active power distribution network oriented refined load prediction method and system | |
| CN113554466A (en) | Construction method, forecasting method and device for short-term electricity consumption forecasting model | |
| CN105389636A (en) | Low-voltage area KFCM-SVR reasonable line loss prediction method | |
| CN111932402A (en) | Short-term power load bidirectional combined prediction method based on similar day and LSTM | |
| CN110689195A (en) | A method of daily power load forecasting | |
| CN104680261A (en) | Power load operation control method based on load curve clustering of major clients | |
| CN115660225A (en) | Electricity load prediction management method and system based on ammeter communication module | |
| CN115907822A (en) | A Relevance Mining Method of Load Characteristic Index Considering Regional and Economic Impact | |
| CN111428745A (en) | Clustering analysis-based low-voltage user electricity utilization feature extraction method | |
| CN110675020A (en) | High-price low-access user identification method based on big data | |
| CN111798333A (en) | A method and system for energy use assessment and electricity safety analysis | |
| CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
| CN117291299B (en) | Moon electricity quantity prediction method considering various influence factors | |
| CN112307675B (en) | Neural network-based temperature-sensitive load separation identification method and system | |
| CN104143117B (en) | Method for extracting correlation coefficient between special load and daily load of power grid | |
| CN111160993A (en) | User practical and stage release coefficient analysis method based on user load application | |
| CN112508254A (en) | Method for determining investment prediction data of transformer substation engineering project | |
| CN110175705B (en) | Load prediction method and memory and system comprising same | |
| CN119209484A (en) | A short-term power load forecasting method based on smart meter | |
| CN113837486B (en) | RNN-RBM-based distribution network feeder long-term load prediction method | |
| CN115829418A (en) | Power consumer load characteristic portrait construction method and system suitable for load management | |
| CN117236532B (en) | A method and system for predicting peak power load based on load data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20151202 |
|
| RJ01 | Rejection of invention patent application after publication |