KR102133034B1 - Smart power demand management system using real-time collected customer load information - Google Patents

Abstract

The present invention discloses a smart power demand management system. The smart power demand management system of the present invention includes a communication unit that receives real-time power data and climate data, a storage unit that stores the received power data and climate data, and an algebraic prediction algorithm using the real-time power data and pre-stored power data. A first prediction unit for calculating a first set of predicted values, a climate change detection unit that analyzes the real-time climate data and determines that there is an abnormality in the climate change pattern when the climate change value is greater than or equal to a reference value, and the climate change detection unit When it is determined that there is an abnormal sign of the climate change pattern, the second prediction unit and the first prediction value set or the second prediction unit for calculating a second prediction value set for updating the first prediction value set using the power data and the climate data. And a notification unit for transmitting power demand related information to the participating customer terminal based on the second set of predicted values.

Description

Smart power demand management system using real-time collected consumer load information {SMART POWER DEMAND MANAGEMENT SYSTEM USING REAL-TIME COLLECTED CUSTOMER LOAD INFORMATION}

The present invention relates to a smart power demand management system, and more particularly, to a smart power demand management system using real-time collected consumer load information capable of quickly providing consumer demand information by collecting consumer load information more quickly. .

Recently, the global trend in the energy sector has focused on regulating carbon emissions, introducing renewable energy, and promoting energy saving. In order to promote this, the government's policy is to participate in the demand management market, operate the Renewable Energy Certificate (REC) trading market, and support the installation of an ESS (Energy Storage System). Accordingly, the power trading market is expected to expand rapidly in the future.

As the demand management market is in full swing, a power trading platform for power trading has been developed and used. The platform for demand resource providers is the main function, based on OpenADR, automatic KPX event message reception and response function, RRMSE automatic calculation function to determine whether registration is possible when registering demand management resources, customer registration and modification by resource managers Deletion function, CBL calculation function, history management function, data excel conversion and download function, gateway remote management function, siren alarm function when feeding orders, dedicated dashboard support function by participating customers, reliability and economics By DR's earnings and penalty It provides various functions such as automatic calculation and correction function, reliability DR and economical DR, a reduction status graph implementation function for each reduction, and a dashboard for smartphones.

However, in the conventional power trading platform, only basic functions of the demand management market are implemented, and thus, it is not possible to provide customers with information on market prediction. Therefore, in the case of reliability DR (demand response resource transaction according to real-time power supply instruction), the customer must reduce the contracted power usage at the designated time delivered one hour before the implementation of the demand management reduction instruction delivered from the power exchange. do. In addition, in the case of economical DR (demand response resource trading according to a pricing development plan), a problem has been pointed out that it is difficult to know advance information about the bid capacity and price for the next day's power trading bid market.

Accordingly, there is a need for a new concept of smart power demand management system development that can collect load information from a customer more quickly and provide information essential for a customer to establish a power usage plan more quickly.

Prior Literature 1: Published Patent No. 10-2013-0045006 (May 03, 2013) Prior Literature 2: Published Patent No. 10-2014-0075617 (June 19, 2014) Prior Literature 3: Published Patent No. 10-2017-0007071 (January 18, 2017) Prior Literature 4: Registered Patent No. 10-1724686 (April 03, 2017)

The present invention, devised by the above-mentioned necessity, collects consumer load information faster than the collection cycle of consumer load information required by the power exchange, and provides economic DR bid participation information, power usage scheduling information to customers participating in the power transaction bidding market, It is an object of the present invention to provide a smart power demand management system that collects consumer load information more quickly so that real-time information on power reduction can be provided at a faster time period.

In order to achieve the above object, a smart power demand management system using real-time collected consumer load information according to an embodiment of the present invention is installed at the consumer side, and the consumer load collected from a plurality of sensors collecting consumer load information in real time A first communication unit that receives information at a predetermined time period and receives climate data from an external server, and calculates a first set of predicted values by an algebraic prediction algorithm using the real-time collected consumer load information and pre-stored consumer load information. The prediction unit analyzes the real-time climate data and determines that there is an abnormality in the climate change pattern when the climate change value is higher than the reference value. The climate change detection unit determines that there is an abnormality in the climate change pattern. If it is determined, the second predictor calculates a second predicted value set for updating the first predicted value set using the consumer load information and climate data, and the second calculated by the first predictor or the second predictor It includes an analysis unit that analyzes a power usage pattern of a consumer using a set of 1 predicted values or the second set of predicted values, and a notification unit that delivers power demand related information to a participating customer terminal based on the analysis result of the analysis unit.

In this case, it further includes a real-time power information collection module that temporarily stores and manages user load information collected from the plurality of sensors.

In this case, the real-time power information collection module transmits and receives the consumer load information through the plurality of sensors and real-time packet communication based on the Internet protocol.

In this case, the real-time power information collection module classifies the corresponding sensor using the Internet protocol address assigned to each of the plurality of sensors, but supplements the unique sensor uniquely assigned to the corresponding sensor. Discern.

On the other hand, the analysis unit predicts the next day's expected power consumption and estimated power reduction by analyzing the trend of power use increase or decrease based on the first set of predicted values or the second set of predicted values.

In this case, the notification unit transmits a first notification message on whether to participate in economical DR bidding using the estimated power consumption of the next day.

In this case, the notification unit transmits a second notification message on whether a specific consumer can achieve reliability DR using the estimated power reduction amount.

In this case, the notification unit transmits to the second notification message, including a graph capable of confirming in real time the current reduction amount compared to the reduction target amount.

According to various embodiments of the present invention, by collecting consumer load information faster than the collection cycle of consumer load information required by the power exchange, economical DR bid participation information, power usage scheduling information, power to customers participating in the power transaction bidding market By providing real-time information on the amount of reduction at a faster time period, the consumer can monitor the amount of power reduction more quickly.

In addition, it is possible to promote convenience of participating customers in the demand management business by providing visualization information on power scheduling at a rapid cycle to the consumer.

1 is a view for illustratively illustrating a power demand management platform to which a smart power demand management system according to an embodiment of the present invention is applied,
2 is a timing diagram for explaining a process of collecting consumer load information in real time according to an embodiment of the present invention;
3 is a block diagram illustrating a smart power demand management system according to an embodiment of the present invention,
4 is a view for explaining a real-time power information collection system according to an embodiment of the present invention,
5 is a view for explaining real-time power information collection according to an embodiment of the present invention,
6 is a view for explaining the difference between the predicted value and the actual value of the maximum peak power using algebraic prediction,
7 is an exemplary diagram for specifically describing a prediction algorithm of the smart power demand management system shown in FIG. 3;
8 is a flowchart for explaining a training process of support vector regression for implementing the second prediction unit illustrated in FIG. 3,
9 is a timing diagram for explaining a process for predicting reliability DR according to another embodiment of the present invention, and
10 is a timing diagram for explaining a process for predicting economic DR according to another embodiment of the present invention,

Hereinafter, preferred embodiments for carrying out the present invention will be described with reference to the drawings. These embodiments are merely illustrative of an example for the purpose of the present invention, and freely modified embodiments may be derived within the scope of the technical spirit of the present invention. First, a structure of a smart power demand management system according to an embodiment of the present invention will be described below.

1. Smart power demand management system

1 is a view for explaining a power demand management platform according to an embodiment of the present invention. Referring to FIG. 1, the power demand management platform includes a smart power demand management system 100, a consumer terminal 200, a power exchange server 300, and a real-time power consumption collection module 400. In addition, it may include a cloud server (S) capable of storing or reading data, and a network (N) connecting the smart power demand management system 100 and the receiver terminal 200.

The smart power demand management system 100 according to an embodiment of the present invention includes power demand forecasting, pricing power generation plan information, power market price forecasting information, power purchase cost forecasting information, reliability demand management, economical demand management, evaluation analysis, real time It can perform various functions related to power demand management, such as information services.

The smart power demand management system 100 serves as a relay that interconnects the consumer terminal 200 and the power exchange 300, and may communicate with the consumer terminal 200 through a public network.

The smart power demand management system 100 may transmit demand management enforcement information to the consumer terminal 200 in the form of an SMS guide message through the network N. Alternatively, the smart power demand management system 100 may transmit result analysis information such as reliability DR and related incentive payment to the consumer terminal 200 through the network N.

The consumer terminal 200 may provide a real-time demand management information service according to the management of the smart power demand management system 100. In addition, the consumer terminal 200 may provide demand reduction enforcement information or power usage real-time information to the demand reduction execution information or the power usage real-time collection module 400.

The power exchange 300 may provide overall information regarding demand management implementation in connection with the smart power demand management system 100. For example, when a situation in which a power supply instruction is required, the power exchange 300 may issue a reliability DR and transmit it to the smart power demand management system 100. Alternatively, the power exchange 300 may evaluate the consumer who performed the demand reduction in response to the reliability DR and deliver the incentive to the smart power demand management system 100.

The demand reduction execution information and the power consumption real-time collection module 400 may collect the demand reduction execution information and the power usage real-time information in real time from the consumer terminal 200. The collection module 400 may automatically update the collected information in real time or at this time on the web cloud server S. Referring to FIG. 2, a process of collecting consumer load information in real time according to an embodiment of the present invention will be described.

The power exchange 300 requests the Nth power meter reading request from the demand management provider 100 to check the consumer load information (S201). The demand management provider 100 requests the M-th customer load information before and after the meter request of the power exchange 300 (S202). The N-th request of the power exchange 300 and the M-th request of the demand management provider 100 may be made simultaneously or bi-directionally. Participating customer 200 responds to the sensing value in response to the M-th request by the sensor installed on the consumer side as M-th customer load information (S203). The demand management provider 100 analyzes the M-th customer load pattern based on the M-th response (S204). After analyzing the Mth load pattern, power demand information is provided to the participating customers (S205). After the M-th request (S202), the process from the M-th request information provision (S204) is repeated 2 to 5 times (S205 to S209). If a certain time passes after this repetition process, power meter reading information for the N-th request is provided (S210). After providing the Nth power meter reading information (S210), the power exchange 300 issues an N+1th power meter reading request (S211).

As described above, at least two or more head load checks are performed between the Nth metering request and the N+1th metering request of the power exchange 300, and preferably 5 customer load checks are performed. As such, when customer load information is collected in real time to analyze the customer's power usage pattern, it is possible to accurately collect or predict customer load information even in an emergency.

The web cloud server (S) provides a service capable of real-time checking and data writing/reading through the data connection with the smart power demand management system 100, real-time information on demand reduction execution information and power consumption that are automatically updated.

3 is a block diagram illustrating a smart power demand management system according to an embodiment of the present invention. Referring to FIG. 3, the smart power demand management system 100 includes a communication unit 110, an analysis unit 120, a first prediction unit 130, a climate change detection unit 140, a second prediction unit 150, and It includes a notification unit 160.

The communication unit 110 may receive real-time information from the web cloud server S. Alternatively, the communication unit 110 may receive consumer power information from the consumer terminal 200 through the network N. In addition, the communication unit 110 may receive power market information in real time from the power exchange 300. In addition, the communication unit 110 may receive real-time weather information from a weather station server, an internet portal service provider, or other weather-related web server.

The analysis unit 120 may store the received real-time power market information and weather information in a database. The analysis unit 120 may store pre-stored power market information and weather information for a period of time. The analysis unit 120 may store power market information and weather information in a form for data analysis. The analysis unit 120 may analyze power information and climate information to analyze power usage patterns. The analysis unit 120 may analyze the power usage pattern using a set of prediction values predicted by the first prediction unit 130 and the second prediction unit 150 described below.

The first prediction unit 130 may predict the first set of predicted values according to an algebraic prediction algorithm based on real-time power market information (eg, consumer load information) and pre-stored power market information. For example, the first prediction unit 130 may predict power peak load data at a specific future time point using past power peak load data and real-time power peak load data.

The first prediction unit 130 may predict information of a specific future time point for time series data using an algebraic prediction algorithm (AP Algorithm). Detailed description of this will be described below with reference to a separate drawing.

The climate change detection unit 140 analyzes climate information received in real time to detect whether there are abnormal signs in the pattern of climate change. When the climate change detection unit 140 determines that the climate change pattern is an abnormal sign, the climate change value is extracted and stored in the storage unit.

The climate change detection unit 140 includes a set of weather information of the day consisting of the maximum temperature, minimum temperature, average temperature, maximum humidity, minimum humidity, and average humidity of the day, and the maximum temperature, minimum temperature, average temperature, maximum humidity, and minimum humidity of the previous day. By comparing each other's weather data set consisting of average humidity, the climate change pattern can be detected as an anomaly when the change width is greater than the reference value. For example, if the maximum temperature of the day (T _max1 ) is more than the maximum temperature (T _max2 ) of the previous day, the climate change detection unit 140 generates an abnormal sign in the climate change pattern. Can be judged. In addition, if a sudden change is detected in the minimum temperature, average temperature, maximum humidity, minimum humidity, and average humidity, the abnormality is determined by comparing with each reference value.

The second prediction unit 150 may update the first predicted value set predicted by the first predictor 130 to the second predicted value set when the abnormality indication of climate change is detected by the climate change detector 140. have. The second predictor 150 may update the first set of predicted values to the second set of predicted values according to a support vector regression algorithm. The support vector regression (SVR) algorithm will be described below with separate drawings.

The notification unit 160 generates information on power demand based on the first set of predicted values or the second set of predicted values predicted by the first predictor 130 and the second predictor 150, and the generated power demand Information on the information is transmitted to the consumer terminal 200 or the power exchange 300. For example, when it is predicted that the reliability DR will be issued based on the first set of predicted values or the second set of predicted values, the notification unit 160 generates and transmits the reliability DR issued prediction information to the user terminal 200. In addition, when the bidding power amount and price for economic DR are predicted based on the first predicted value set or the second predicted value set, the notification unit 160 generates economical DR information and transmits it to the user terminal 200.

In the case of the present invention, the real-time power information collection module 400 collects the power information of the consumer in real time, performs load pattern analysis using the collected power information, and then schedules the exhibition use to each consumer, and the current reduction amount. Provides actual information, warning message information, etc. against the target. The notification unit 160 delivers the power demand information to the user terminal 200 together with economic DR and reliability DR information or separately. Hereinafter, a specific process in which power information is collected in real time by the real-time power information collection module 400 and provided to the smart power demand management system 100 will be described.

2. Real-time power information collection module

4 is a view for explaining a real-time power information collection module according to an embodiment of the present invention. Referring to FIG. 4, the real-time power information collection module 400 is directly or indirectly connected to the smart power demand management system 100 to provide power information collected in real time. The real-time power information collection module 400 collects user load information in a time period within 5 minutes, preferably 1 minute to 2 minutes, and provides the collected user load information to the smart power demand management system 100 Can.

According to the present invention, the smart power demand management system 100 may use real-time collected consumer load information. At this time, the consumer load information collected in real time means load information collected from the consumer at time intervals of 1 minute to 2 minutes. In general, load information is collected from the consumer at a time interval of 5 minutes or more, but when the time interval is 5 minutes or more, a sudden load usage change occurring at the maximum time interval may not be predicted.

Therefore, the smart power demand management system 100 according to the present invention collects consumer load information in units of 1 minute from sensors installed on the consumer's side to collect load usage of the consumer. This will be described with reference to a separate drawing.

5 is a view for explaining real-time power information collection according to an embodiment of the present invention. Referring to FIG. 5, the real-time power information collection module 400 is connected to a plurality of sensors (node_1, node_1 to node_N) installed on the user side through the Internet.

The real-time power information collection module 400 receives sensing values collected from a plurality of sensors (node_1, node_1 to node_N) using packet communication based on the Internet protocol.

The real-time power information collection module 400 collects consumer load information from a plurality of sensors (node_1, node_1 to node_N) at a predetermined time period (for example, 1 minute or 2 minutes). If the predetermined time is collected in less than 1 minute, the consumer load information can be obtained more accurately. However, since the width of the load change is not large within 1 minute, it is preferable to request the consumer load information in a period of 1 to 2 minutes since an overload can be expected in the network when requesting the consumer load within 1 minute.

The plurality of sensors (node_1, node_1 to node_N) installed on the consumer's side is embedded in the consumer's power meter or separately installed to monitor the load power of the consumer. The plurality of sensors (node_1, node_1 to node_N) have unique Internet protocol addresses, and have unique values assigned uniquely to identify devices.

Since the real-time power information collecting module 400 can communicate with a plurality of sensors (node_1, node_1 to node_N) through the Internet, the real-time power information collecting module 400 has a plurality of sensors (node_1, node_1 to node_N) may be received in real time. In this way, the plurality of sensors node_1, node_1 to node_N) is a configuration that can be constructed by the Internet of Things (Internet of Thing).

3. Algebraic Prediction Algorithm

6 is a diagram for explaining a difference between a predicted value and an actual value of peak power using algebraic prediction. Referring to FIG. 6, the predicted value and the actual value of the peak power value based on the algebraic prediction algorithm described below may coincide and may differ.

Using the algebraic prediction algorithm according to an embodiment of the present invention, the past data for peak power is modeled as a Hankel matrix, and then the verified data is added to n rows of the Hankel matrix to correct the Hankel matrix. By doing so, a practical algebraic prediction model can be constructed.

This is a technique for short-term time series prediction based on the identification of skeleton algebraic series. In other words, it aims to build a mathematical problem in the engineering model of the process and project it in the future. The skeleton of the algebraic sequence is identified based on the Rank of the Hankel Matrix.

The Hanken matrix is named after Herman Hankel, and is widely used for system identification when a sequence of output data is given when realization of the basic state space model is required.

The concept of Hankel ranking of sequences is proposed for numerical sequence identification. However, the Hankel sequence of sequences is a concept independent of the realization of the state space of the system. The Hankel class does not approximate the analytical model of the underlying dynamic system, but describes the algebraic relationship between elements in a sequence. In addition, this algebraic relationship can be described by the Hankel matrix composition and formula as shown in Equation 1 below.

As shown in Equation 1 above, for a given number of observations with 2n elements, a model of the process of estimating past behavior into the future can be expressed.

In Equation 1, P _k-1 is an observation value at the present time. Assuming that the sequence is logarithmic and its Hankel rank is equal to n, the next element of the sequence P _k can be determined in the form of a Hankel matrix as shown in Equation 2 below.

In Equation 2, the only unknown is P _k or a determinant that regards the determinant of a Hankel matrix as 0 can be expressed as a set of linear algebraic equations for solving P _k as in Equation 3 below.

Equation 3 is a general formula for algebraic progression such as arithmetic progression and geometric progression. However, Equation 3 is not easy to be applied to a real-time time series due to existing interference and noise in a real system.

To solve this, we need to create a more realistic prediction model using the concept of the Hankel class. Equation 3 above is improved to identify a better representation of the skeleton algebraic sequence in order to predict a more accurate future value of the time series. Equation 3 may be modified as in Equation 5 below.

Equation 4 is still a valid equation for solving P _k . To increase the accuracy of the prediction, additional rows are added on the matrix represented by Equation (3) to obtain more information. Equation 5 may be simply expressed as Equation 6 below.

It is important that the number of measurements (m) exceeds the number of system parameters (n). That is, m>n. Therefore, it is possible to filter out measurement errors in the estimation process and obtain a good quality estimate. In the method of linear least error squared estimation, the goal is to minimize the sum of squares of errors or residuals. Using the conventional process of obtaining a pseudoinverse matrix, Equation 7 below can be obtained.

Where [P ^T P] ^-1 P ^T is the left pseudo-inverse of P, and ^a is the optimal or smallest least squares estimate of a. The P matrix implementation is very similar to creating a Hankel matrix using a series of hourly data from the previous day.

However, the most important part of maximizing the proposed methodology is figuring out the optimal values of the row vectors that have proven to be groundbreaking for the implementation of the proposed prediction strategy. The linear least error squared estimation method finds the average of a series of measurements. When the set of measurements has a Gaussian error distribution, the Tennessee value is generally accepted as the best estimate. It can also be adversely affected by the presence of bad data. Therefore, to set the best estimate based on the data provided, you should obtain as much historical data as possible.

Therefore, irrelevant data can be ignored during data preparation and preprocessing. According to an embodiment of the present invention, it is possible to determine an appropriate amount of data to be used, preferably by testing using a historical data set of 10 to 40 days.

At the point of occurrence of the climate change event shown in FIG. 6, it is necessary to update the first set of predicted values predicted by the algebraic prediction algorithm. The update of the first set of predicted values predicted by the first predicting unit 130 is performed by a training model that trains a support vector regression model using climate information and power market information using a support vector regression algorithm. Can be updated to a second set of predicted values.

4. Hybrid prediction algorithm

7 is an exemplary diagram specifically illustrating a prediction algorithm of the smart power demand management system illustrated in FIG. 3. Referring to FIG. 7, the first prediction unit 130 generates a first set of predicted values including power peak values at a specific time point based on past power peak information and real-time power peak information by an algebraic prediction algorithm. The climate change detection unit 140 transmits the first set of predicted values to the notification unit 160 when there is no significant change between the same-day climate information and the previous-day climate information using real-time collected climate information. If a large change between the same-day climate information and the previous-day climate information is detected, the first predicted value set and the detected climate change value are transmitted to the second prediction unit 150. The second predictor 150 updates the first predicted value set to the second predicted value set using the first predicted value set and the climate change value by the trained SVR module described below. The second prediction unit 150 transmits the second set of prediction values to the notification unit 160.

5. Support Vector Regression (SVR) model training

8 is a flowchart for explaining a training process of support vector regression for implementing the second prediction unit illustrated in FIG. 3. Referring to FIG. 8, in order to set a trained SVR module, data is generated using a process of generating a first set of predicted values (S501 to S503), a process of detecting climate change (S504 to S506), and a set of first predicted values and a climate change value. After filtering, set the trained SVR module through the data training with the selected data.

At this time, support vector regression (SVR) is an extension of the support vector machine (SVM), an algorithm for numerical prediction. The SVM algorithm is a state-of-the-art machine learning algorithm that can be applied to classification tasks. The SVM method finds the maximum margin hyperplane between two classes by applying an optimization method using training data. Decision boundaries are defined as a subset of training data called support vectors. Kernel functions can be used to form nonlinear decision boundaries while keeping computational complexity low.

SVR also creates decision boundaries that can be represented by several support vectors and can be used with kernel functions to create complex nonlinear decision boundaries. Like Linear Regression, SVR tries to find the function that best fits the training data. Unlike linear regression (LR), SVR ignores the error and defines the tube around the regression line using user-specified parameters. The error is ignored and an attempt is made to maximize the flatness of the line as well as minimize the error.

The SVR for time series prediction can be expressed as Equation 8 below. The given training data is (x1, y1), (x2, y2),. . . , (xn, yn). Where x is the input vector and y is the related output of x.

Where xi is mapped to the high dimensional space and ζ _i is the upper training error (ξ * i is lower) dependent on the ε-sensitive tube. |yi-(wTφ (xi) + b) | ≤ ε. The parameters controlling the regression quality are the cost of the error c, the area of the tube ε and the mapping function φ.

The constraint of Equation 8 above is that most data xi is tube |yi-(wTφ (xi) + b) | Implies that we want to include in ≤ ε. If xi is not within this tube, there is an error ζ _i or ξ*i, which is minimized in the objective function. For traditional least squares regression, ε is always 0 and the data is not mapped to high dimensional space. However, SVR is a more general and flexible method of handling regression problems.

The main parameters of SVR are ε and C. As described above, ε defines an error-free tube around the regression function, so it controls how well the function fits the training data. Parameter C controls the balance between training error and model complexity. The smaller C increases the number of training errors. A larger C increases the penalty for training errors and results in behavior similar to a strong margin SVM.

The principle of the SVR formula is constructed from a similar point of view as the SVM, and once trained, the SVR generates predictions using Equation 9 below.

In order to implement the SVR in the present invention, input learning data xi and yi represent climate-related parameters and load deviation, respectively. When training of the input data pair is ready, the trained SVR module can be used to approximate the yn value given the input xn.

According to an embodiment of the present invention, if the climate is not input to the time series model, the prediction capability may be limited in general. In other words, the impact of climate variables such as temperature and humidity is important. In a neural network approach, minimum distance measurements can be used to identify appropriate historical patterns of load and temperature readings to estimate network weights. This is to overcome the problem of rapid changes in weather patterns. In a similar way, historical patterns of load, temperature and humidity are identified as the basis for SVR module training. When a significant change between the previous day and the current daytime weather conditions is found, an increased error of the first set of prediction values using the algebraic prediction algorithm is observed.

In one embodiment of the present invention, the algebraic prediction algorithm uses the concept of the Hankel matrix, and the input data may use power data of the previous year of the year to be predicted. AR, ARMA and SVR methods can also be implemented to indicate the performance of algebraic prediction algorithms compared to other prediction models. Therefore, the performance of the algebraic prediction algorithm can be confirmed by increasing the input of historical data and changing the value of the row vector of the modified Hankel matrix.

In addition, iterative one-step forward prediction can be implemented to evaluate the performance of each predictive model on the day when significant changes in weather conditions are found. It is understood that even if the first predicted value set is predicted by the algebraic prediction algorithm using the past power data, an error may occur in the first predicted value set without considering climate change.

Therefore, by training the SVR module, it is possible to reduce the error of the first predicted value set by updating the first predicted value set with the second predicted value set. Hereinafter, a process of training the SVR module will be described.

Referring to FIG. 8, the process of building a trained SVR module begins with preparing a significant amount of historical data to build a more stable reward module. Once the trained SVR module is built, the preliminary peak load output of the algebraic prediction method can be easily adjusted on the day of significant changes in weather conditions.

The SVR training procedure can be summarized as follows starting from the SVR module configuration. As a first step, hourly historical data of load, temperature and humidity for the past 2-3 years are collected. As a second step, AP simulation is performed in summer or winter on the collected past data. As a 33rd step, according to the simulation result, a day in which a discrepancy between the actual peak load and the predicted peak load is important is identified. As a fourth step, the collected data is checked by filtering out those with corresponding significant deviations in weather parameters (ie, ΔT _MAX , ΔT _AVE and ΔH _AVE ) between the current and previous dates. In the fifth step, the filtered data is trained using the normalized value of the peak load mismatch (first predicted value set) as ΔT _MAX , ΔT _AVE and ΔH _AVE as the SVR and the input training set X and the output training set Y. Identifies suitable kernel functions as well as optimal parameter settings for ε and C. As a sixth step, the trained SVR module is stored.

When the trained SVR module is prepared to compensate for the first set of predicted values (peak load), the following procedure is performed for real-time manipulation.

In the first step of adjustment, 14-hour prediction is performed by an algebraic prediction algorithm to determine peak load conditions.

In the second step of adjustment, check if the most recent weather forecast is significantly different from the previous day's weather conditions.

If a significant difference is found in the third step of adjustment, proceed to the fourth step. If not found, the initial algebraic prediction algorithm result is used as the final peak load prediction.

In the fourth step of adjustment, the current ΔT _MAX , ΔT _AVE and ΔH _AVE and trained SVR modules as input variables are used to predict the corresponding increase or decrease in peak load.

In the fifth step of adjustment, for the peak load correction given by the set P ₁ , P ₂ , .... P _i , ....., P _N representing the predicted load for the next N time, by the following equation (10) Update AP results.

Here, P ₀ is the same as the initial prediction P ₁ for 1 hour, P _max is the peak load based on the result of the first prediction value, ΔP _SVR is the peak load adjustment value of the trained second predictor, P _i,new Is a predicted value updated at i hour by the second predicting unit.

In the sixth step of adjustment, a new time zone prediction set is stored with the updated peak load condition. For actual implementation, the only data available is the demand recorded in the previous year. So a trained SVR module that can be built for the summer season is based on three months of summer data: July, August and September data. Optimal settings for SVR training as kernel functions are parameter values of ε = 0.001, C = 10 and radial basis function (RBF).

For the input training set X and Y used for the SVR building, X is a vector with three components ΔT _MAX , ΔT _AVE and ΔH _AVE and Y is the normalized value in the set of all Ys (peak load mismatch) to be. Since the temperature rise in summer means a positive increase in peak load demand, when ΔT _MAX and ΔT _AVE have a positive value along with a positive Y value, the pair of X and Y values can be considered as viable data. have.

Although it is more reasonable to consider the hourly temperature and humidity data for the correction of the results of the algebraic prediction algorithm, it is much easier to perform and set the actual implementation by entering only the maximum and average values in the SVR compensation module.

6. Reliability DR issuance process

9 is a timing diagram illustrating a process for predicting reliability DR according to an embodiment of the present invention. Referring to FIG. 9, the demand management provider receives power market information from the electronic exchange 300 using the smart power demand management system 100 (S901). The smart power demand management system 100 receives consumer information from the participating customers 200 (S902). The smart power demand management system 100 receives weather information from the outside (S903). The smart power demand management system 100 determines whether the reliability DR is predicted in a mixed manner by the algebraic prediction algorithm and the trained SVR module (S904). When the reliability DR is predicted to be issued within a predetermined time, the smart power demand management system 100 transmits the reliability DR announcement to the participating customer terminal 200 (S905). Participating customer 200 prepares for the power reduction instruction (S906). The power exchange 300 issues a demand reduction event to the demand management provider 100 (S907). Demand management The demand management provider 100 notifies the participating customers 200 of a demand reduction event (S908). Participating customer 200 performs power reduction at a predetermined time (up to 1 hour after receiving the instruction) (S909). Participating customer 200 transmits meter reading data to demand management provider 100 (S910). The demand management provider 100 notifies the power exchange 300 of the meter reading data (S911). The power exchange 300 calculates the reliability DR result based on the meter reading data (S912). The power exchange 300 pays the settlement amount to the demand management provider 100 (S913). The demand management provider 100 distributes the settlement money to the participating customers 200 (S914).

As described above, according to an embodiment of the present invention, before the reliability DR is issued by the power exchange 300, the smart power demand management system 100 predicts whether the reliability DR is issued, and participates in forecasting by the customer 200 By delivering to, the participating customer 200 can make preparations for power reduction more smoothly.

7. Economical DR issuance process

10 is a timing diagram illustrating a process for predicting economical DR according to an embodiment of the present invention. Referring to FIG. 10, the demand management provider 100 receives power market information from the exchange 300 using the smart power demand management system 100 (S1001). The smart power demand management system 100 receives consumer information from the participating customers 200 (S1002). The smart power demand management system 100 receives weather information from the outside (S1003). The smart power demand management system 100 predicts the bidding capacity and the estimated price using the above-described algebraic prediction algorithm and the first set of predicted values or the second set of predicted values predicted by the trained SVR module (S1004). The demand management provider 100 provides the bidding capacity and the predicted price to the participating customers 200 (S1005). Participating customer 200 determines whether to bid (S1006). The participating customer 200 proposes a power supply contract bid to the demand management provider 100 when the bidding capacity and the expected price condition are acceptable (S1007). The demand management provider 100 provides power supply contract bidding information (S1008). The power exchange 300 selects the participating customers 200 that meet the price conditions and notifies the demand management provider 100 of the power supply contract decision (S1009). The demand management provider 100 notifies the power supply contract decision (S1010). Participating customer 200 supplies power according to the conditions of the power supply contract (S1011).

As described above, according to an embodiment of the present invention, the bidding capacity and the expected price for economic DR are predicted and provided by the demand management provider 100 to the participating customer 200, so that the participating customer 200 has a power supply contract Since the bid capacity and the estimated price, which are important factors for participating in bidding, can be confirmed in advance, the effect of being able to participate in bidding is exerted.

On the other hand, that all components constituting the embodiments of the present invention are described as being combined or operated as one, the present invention is not necessarily limited to these embodiments. That is, within the scope of the present invention, all of the constituent elements may be selectively combined and operated. In addition, although all of the components may be implemented by one independent hardware, a part or all of the components are selectively combined to perform a part or all of functions combined in one or a plurality of hardware. It may be implemented as a computer program having a. The codes and code segments constituting the computer program can be easily deduced by those skilled in the art of the present invention. Such a computer program is stored in a computer-readable non-transitory computer readable media, and read and executed by a computer, thereby implementing an embodiment of the present invention.

Here, the non-transitory readable recording medium means a medium that stores data semi-permanently and that can be read by a device, rather than a medium that stores data for a short time, such as registers, caches, and memory. . Specifically, the above-described programs may be stored and provided on a non-transitory readable recording medium such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM, and the like.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific embodiments described above, and it is usually in the technical field to which the present invention pertains without departing from the gist of the present invention as claimed in the claims. Of course, various modifications can be implemented by a person having knowledge of these, and these modifications should not be individually understood from the technical idea or prospect of the present invention.

100: smart power demand management system 110: Communication Department
120: analysis unit 130: first prediction unit
140: climate change detection unit 150: second prediction unit
160: notification unit 200: prisoner terminal
300: power exchange server 400: real-time power information collection module

Claims (8)

In the smart power demand management system using real-time collected consumer load information,
A communication unit that is installed on the consumer side and receives the consumer load information collected from a plurality of sensors collecting real-time consumer load information in a predetermined number of times at a time period of 1 minute to 2 minutes, and receives climate data from an external server;
A real-time power information collection module that receives and stores the consumer load information collected from the plurality of sensors through the communication unit and temporarily stores and manages them;
When the consumer load information is received by the predetermined number of times, power peak load data at a specific future time point is generated by an algebraic prediction algorithm by using power peak load data of the real-time collected consumer load information and pre-stored past power peak load data. A first predicting unit calculating a set of predicted values;
A climate change detection unit that analyzes the climate data and determines that there are abnormal signs in the climate change pattern when the climate change value is greater than or equal to a reference value;
If it is determined by the climate change detection unit that there is an abnormal sign of a climate change pattern, the first predicted value set is updated by a support vector algorithm using the consumer load information and climate data to calculate a second predicted value set. 2 prediction unit;
An analysis unit for analyzing a power usage pattern of a consumer by using the first prediction value set or the second prediction value set calculated by the first prediction portion or the second prediction portion; And
On the basis of the analysis results of the analysis unit includes a notification unit for transmitting information related to power demand to the participating customer terminal,
The real-time power information collection module transmits and receives the consumer load information through the plurality of sensors and real-time packet communication based on the Internet protocol, and distinguishes the corresponding sensors by using the Internet protocol addresses assigned to each of the plurality of sensors, A smart power demand management system that uses real-time collected consumer load information to identify the sensor by using the unique value uniquely assigned to the sensor. delete delete delete According to claim 1,
The analysis unit, by using the first set of predicted values or the second set of predicted values by analyzing the tendency of the increase or decrease in power usage, predicts the next day's expected power consumption and the expected power reduction amount Smart using real-time collected consumer load information Power demand management system. The method of claim 5,
The notification unit, a smart power demand management system using real-time collected consumer load information, characterized in that using the estimated power consumption of the next day to send a first notification message about whether to participate in economical DR bidding. The method of claim 5,
The notification unit, a smart power demand management system using real-time collected consumer load information, characterized in that for transmitting a second notification message about whether a specific consumer can achieve reliability DR using the estimated power reduction. The method of claim 7,
The notification unit, a smart power demand management system using real-time collected consumer load information, characterized in that the second notification message includes a graph to check in real time the current reduction amount compared to the reduction target amount.

Images