KR20140075617A - Method for estimating smart energy consumption - Google Patents

Abstract

An intelligent energy consumption estimation method includes (a) a step in which a first least square support vector machine part included in a building energy consumption estimation apparatus generates the estimated amount of current power consumption of a building facility by using a least square support vector machine regression algorithm (LS-SVR) as to the past amount of power consumption of the building facility; and (b) a step in which a second least square support vector machine part of the building energy consumption estimation apparatus generates the estimated amount of power consumption of a specific date in the future after the amount of the current power consumption of the building facility by using LS-SVR as to the generated estimated amount of current power consumption and the generated amount of current power consumption of the building facility.

Description

[0001] The present invention relates to a method for estimating smart energy consumption,

The present invention relates to building energy management technology, and more particularly, to intelligent energy consumption prediction method.

(RTP) tariff system under the smart grid (Smart Grid) environment, a system that imposes a predetermined fee for each time period, or a system that imposes a peak charge for a fixed number of times and a fixed time period (CPP) are being reviewed.

This is primarily aimed at reducing the known energy consumption and, secondarily, managing the energy consumption within a controllable range (peak power management), thereby lowering the cost of production (such as nuclear or coal-fired power) and green energy Solar power, wind power or fuel cell) to control power surplus power at a controllable level.

KEPCO is implementing a contracted electricity tariff (equal-tariff electricity tariff) in the short term prior to real-time tariffs, which will charge electricity fees. In terms of the charging standard of the contracted power plan, the monthly electricity bill = basic charge + use charge.

The base charge is multiplied by the unit price of the contract based on the maximum annual power consumption per hour, and it is set as the annual reference charge. In other words, lowering the annual maximum power consumption (peak power consumption) is the key to reducing the base charge. The usage fee is charged by the KEPCO as a high rate in the high energy consumption period (time interval). The electricity charge (power consumption) of each section is multiplied by the usage charge (usage time). Therefore, it is important to lower both the base rate and the usage fee to lower the cost of electricity.

Future energy policies should be based on uniform load contract power (ie, exceeding a certain range or below) to contract energy costs at a constant load maintenance condition that requires stability of load fluctuation to be able to predict and control energy consumption while suppressing energy consumption. (Ie, stable energy consumption contracts) to the energy supply through low-cost energy generation sources, and to prevent generation of high-cost sources such as diesel or LNG due to unplanned peak power .

It will be differentiated from the existing real-time power plan (RTP) and peak control plan, lowering the cost of power generation, and the electric power tariff system for stable supply of electricity.

The technical object (object) to be solved by the present invention is to provide an intelligent energy consumption prediction method which can be used for reducing the energy consumption cost of a building under the contract power plan, Generating device, and a method of providing a power usage instruction using a power usage instruction generating device of a building facility.

According to an aspect of the present invention, there is provided an apparatus for generating a power usage guide for a building equipment, the apparatus comprising: And predicts the energy consumption according to the time during the day of the building equipment used for the future specific day based on the predicted temperature, the predicted humidity, the day of the week, and the season of a specific future in the future Demand forecasting section; And estimating a demand of the building equipment to be used for a specific day in the future by reflecting the real-time electricity price, the output capacity of the building facility, and the maximum efficiency of the building equipment, And instructions for generating instruction data.

According to another aspect of the present invention, there is provided a method of providing power usage guidelines using a power usage guidelines generating apparatus for a building facility, the method comprising: (a) On the basis of the power amount data according to the past humidity, the electricity amount data according to the past day, the electricity amount data according to the past season, and the predicted temperature, the humidity, the day of the week, and the season Estimating an amount of energy consumption according to time during a day of the building equipment used for a certain future work; And (b) a guidance unit of the power usage guide generation apparatus uses the predicted result of the demand forecast unit to reflect a real-time power charge, an outputable capacity of the building facility, and a maximum efficiency of the building facility, And generating energy usage guideline data of the building equipment to be used.

The method for generating a power usage guide for building equipment according to the present invention and the method for providing power usage instructions using a power usage guide generating device for a building facility are provided by providing instructions on power usage for building equipment, Building energy consumption and building energy costs (power charges) can be reduced, and building energy can be optimized.

Future energy policies should be based on uniform load contract power (ie, exceeding a certain range or below) to contract energy costs at a constant load maintenance condition that requires stability of load fluctuation to be able to predict and control energy consumption while suppressing energy consumption. (Ie, stable energy consumption contracts) to the energy supply through low-cost energy generation sources, and to prevent generation of high-cost sources such as diesel or LNG due to unplanned peak power .

It will be differentiated from the existing real-time power plan (RTP) and peak control plan, lowering the cost of power generation, and the electric power tariff system for stable supply of electricity.

The present invention also predicts power consumption using two least-squares supporting vector machine regression algorithms (LS-SVR), thereby effectively predicting power demand.

BRIEF DESCRIPTION OF THE DRAWINGS For a more complete understanding of the drawings used in the detailed description of the present invention, a brief description of each drawing is provided.
FIG. 1 is a block diagram illustrating an apparatus 100 for generating power usage guidelines for a building facility according to an embodiment of the present invention.
FIG. 2 is a flow chart illustrating a method 200 for providing a power usage instruction using a power usage instruction generating device of a building facility according to an embodiment of the present invention.
FIG. 3 is a graph showing output results provided by the power usage guidelines generating apparatus 100 of the building facility of FIG. 1 or the power usage guidelines providing method 200 of FIG.
FIG. 4 is a diagram showing a configuration of a matlab of the power usage guide generating apparatus 100 of the building facility shown in FIG.
FIG. 5 is a graph illustrating a time shift of a building facility using energy usage instruction data generated by the apparatus 100 of FIG.
FIG. 6 is a graph illustrating the Smart Regulation Control of building equipment using the energy use instruction data generated by the apparatus 100 of FIG.
FIG. 7 is a diagram showing an example of a simulation result of the distributed control and the uniform control described with reference to FIGS. 5 and 6. FIG.
FIG. 8 is a diagram showing another example of a simulation result of the distributed control and the uniform control described with reference to FIGS. 5 and 6. FIG.
FIG. 9 is a diagram showing a configuration of a building energy management system to which the present invention described with reference to FIG. 1 can be applied.
10 is a diagram showing an example of the simulation result of the uniform control of the building equipment using the energy use instruction data generated by the apparatus 100 of FIG.
11 is a diagram showing an example of a simulation result of dispersion control of building equipment using energy use instruction data generated by the apparatus 100 of FIG.
FIG. 12 is a block diagram illustrating an apparatus 300 for generating a power usage guide of a building equipment according to another embodiment of the present invention.
13 is a functional block diagram corresponding to Fig.
Fig. 14 is a view for explaining another embodiment of the guidance unit 340 in Fig.
FIG. 15 is a view for explaining an embodiment of the SR-Index control method using the output value of the guidance unit 340 of FIG.
16 is a view for explaining an intelligent energy consumption predicting method according to another embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS For a better understanding of the present invention, and the objects attained by the practice of the invention, reference should be made to the accompanying drawings, which illustrate embodiments of the invention, and to the description in the accompanying drawings.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be described in detail with reference to embodiments of the present invention with reference to the accompanying drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. Like reference numerals in the drawings denote like elements.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having", etc., are used to specify that a feature, a number, a step, an operation, an element, a component, or a combination thereof, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the contextual meaning of the related art and, unless expressly defined herein, are to be construed as either ideal or overly formal Do not.

The present invention is based on a cost-based real-time rate plan in a Smart Grid environment, and it is possible to utilize energy for a solution to reduce building energy cost rather than a conventional building energy saving solution by changing a paradigm of energy consumption that a consumer is exposed to You can provide usage guidelines (guidelines) for the technology and how to implement it. In summary, the present invention provides a Smart Regulation Management Theory (SRMT) that can reduce building energy consumption and building energy costs (power charges) under the Smart Grid energy contract price policy. Theory) algorithm and a control method using the same.

FIG. 1 is a block diagram illustrating an apparatus 100 for generating power usage guidelines for a building facility according to an embodiment of the present invention.

Referring to FIG. 1, a power usage guide generating apparatus 100 of a building facility includes a prediction unit (Predictor) 105 and an indication unit (Indicator) 110.

The demand predicting unit (or the electric power demand predicting unit) 105 estimates the electricity amount data (electric power consumption data) according to the temperature in the past (for example, the past one year), the humidity in the past (For example, one day in the previous year) (for example, a special day such as a weekend (holiday) or election day), the amount of electricity in the past (for example, one year in the previous year) Data, and buildings used for certain future days based on the predicted temperature of a particular day in the future (e.g., the outside temperature of the building), the predicted humidity (e.g., outside humidity of the building) ) It is possible to predict the energy consumption according to the time of day of the facility. In other words, the demand forecasting unit 105 estimates the demand amount of the building equipment based on the electricity amount data of the building equipment according to the temperature before the specific day in the future, the electricity amount data of the building equipment according to the humidity before the future, The power quantity data of the building equipment according to the day of the week, the power quantity data of the building equipment according to the season before the future specific day, the temperature and humidity (temperature and humidity information) predicted on the future specific day, It is possible to predict the energy consumption amount according to the time during the day of the building equipment used for the future specific day based on the corresponding day of the week and season (day of the week and season information). The predicted temperature and predicted humidity of a specific future of the future may be provided from the weather database server of the meteorological office to the power usage guidelines generating apparatus 100 (or demand forecasting section 105) via a communication network. The least square support vector machine (least squares method) is used to predict the energy consumption over time of the building equipment used in the future on a certain day by using the electricity data according to the temperature of one year in the previous year or the electricity data according to the humidity in the previous year. (Regression) and an optimization method such as differential evolution algorithm (DEA) can be used.

An example of an operation (function) or an operation method of the demand predicting unit 105 using the least squares method and the difference evolutionary algorithm will be described as follows.

In each of the electricity amount data for one day in the past (for example, one year before the past), the predicted temperature of the future specific day for estimating the energy demand of the building equipment and the electricity amount data (Multiplication) a function value including a temperature, and then an average value of the multiplied power amount data is obtained, so that building power amount data for a day of a future specific day can be predicted. The value obtained by dividing the building power data for one day of the future specific day by the predicted temperature of the future specific day may indicate the amount of power when the temperature of the building changes by 1 (° C). For example, it is assumed that three pieces of power amount data for the past one day exist as the first to third power amount data. A value obtained by multiplying the first power amount data by 25 (占 폚) / 30 (占 폚) which is an example of the function value and a value obtained by multiplying the second power amount data by 26 (占 폚) / 30 The power amount data may be averaged by multiplying 32 (占 폚) / 30 (占 폚), which is an example of the function value, to estimate the building power amount data for one day used for a future specific day. 25 (占 폚), 26 (占 폚) and 32 (占 폚) are the temperature of the past specific day and 30 (占 폚) is the predicted temperature of the future specific day.

On the other hand, the operation of the demand predicting unit 105, which calculates the electricity amount data for one day of the future specific day that takes into account (reflects) the humidity using the least squares method and the difference evolutionary algorithm, Is similar to the example of the operation of the demand predicting unit 105 that calculates the power amount data for one day of a future specific day, so that the explanation thereof is based on the demand forecast A description of an example of the action of the unit 105 can be referred to.

In another embodiment of the present invention, when estimating the energy consumption amount of the building equipment used in the future on a specific day by using the electricity quantity data according to the temperature for the previous year or the electricity quantity data according to the humidity for the previous year, Such as the Electricity Consumption Predictor Model Predictive Control Method, the Optimization Pointing Method Golden Section Search Method, or the Probability Modeling of the Non-Gaussian in the noise.

The guidance unit 110 may reflect energy consumption of the building equipment on a specific day in the future, reflecting the real-time electricity charge, the output capacity of the building facility, and the maximum efficiency of the building equipment, It is possible to generate optimal energy usage guideline data according to the time of day of a building facility that can be used for cost reduction. The energy use instructions may be provided to a control unit for controlling the building equipment and may be implemented in a control unit for controlling the building equipment and may be implemented using any one or more of the following methods of uniform control, time shift control, pattern control, SR-index, Energy Diamond, or HVAC (BEMS) control, such as the control of a building. For example, the maximum efficiency of a building installation is as follows. In the case of an ice storage facility of a building, the maximum output efficiency can be obtained at a temperature of 3 (캜), and in the case of a battery facility of a building, if the discharge time is extended to an optimal discharge time (for example, 3 hours) The energy can be discharged at the highest rate (for example, 60 (%)).

The power usage guide generating apparatus 100 may further include an estimation unit or a tracking unit (not shown). The follow-up control unit may turn on / off each of the building equipments based on the energy use instruction data generated by the guidance unit on a predetermined time basis (for example, every 15 minutes, ) Operation. That is, the present invention estimates the energy consumption amount of the building based on the amount of electricity in the past (for example, one year in the previous year) and environmental factors (temperature, humidity, day of the week, or season) , And the building facility can provide a method of consuming energy for a predetermined period of time (for example, 15 minutes) by tracking the provided usage instruction data.

As described above, the present invention can predict (estimate) the energy of a building used in a certain future work, and generate the optimal energy resource based on the predicted building energy And provides an indication of optimal energy use. The optimal control may include, for example, equal control, peak control, time shift, pattern control, or BAS (building automation system) facility optimal control, Building energy or building energy costs can be reduced. In another embodiment of the present invention, by using the ideal guidance, which is a guide for optimal energy use, based on the current usage energy, Can be controlled.

FIG. 2 is a flow chart illustrating a method 200 for providing a power usage instruction using a power usage instruction generating device of a building facility according to an embodiment of the present invention. The power usage instruction providing method 200 may be applied to the power usage instruction generating apparatus 100 of the building equipment of FIG.

2 and 1, in the demand forecasting step 205, the demand forecasting unit 105 included in the power-usage-guide generating apparatus 100 calculates the power consumption amount data according to the past temperature, the electricity amount data according to the past humidity, Based on past power day data, past power year data, and predicted temperature, predicted humidity, day of the week, and season of the future for a particular day, Can predict the energy consumption over time. In other words, the demand forecasting unit 105 estimates the demand amount of the building equipment based on the electricity amount data of the building equipment according to the temperature before the specific day in the future, the electricity amount data of the building equipment according to the humidity before the future, The power quantity data of the building equipment according to the day of the week, the power quantity data of the building equipment according to the season before the future specific day, the temperature and humidity (temperature and humidity information) predicted on the future specific day, It is possible to predict the energy consumption amount according to the time during the day of the building equipment used for the future specific day based on the corresponding day of the week and season (day of the week and season information).

According to the instruction generating step 210, the guideline 110 of the power usage guideline generating apparatus 100 generates real-time electricity charges, output capacity of the building facility, It is possible to generate energy usage guideline data according to the time of day of the building equipment which can be used to reduce the energy consumption cost of the building equipment on a certain day in the future in order to reflect the highest efficiency of the facility.

FIG. 3 is a graph showing output results provided by the power usage guidelines generating apparatus 100 of the building facility of FIG. 1 or the power usage guidelines providing method 200 of FIG.

In FIG. 3, Estimate indicates the predicted power amount data according to the time during one day of the building equipment used in the future specific day, the indicator indicates the energy use instruction data based on the predicted power amount data, And indicates a tracking control value according to the data. In FIG. 3, Tracking can be displayed in the form of a bar graph, and the horizontal axis of the graph shows an example of a real time electricity tariff (low price, high price) over time.

FIG. 4 is a diagram showing a configuration of a matlab of the power usage guide generating apparatus 100 of the building facility shown in FIG.

Referring to FIG. 4, the estimator corresponds to (corresponds to) the demand predicting unit 105 and the guidance unit 110 shown in FIG. The LS-SVR in the estimator can refer to a functional block of least squares, DEA # 1, Dynamic Model, and DEA # 2 can represent functional blocks of a differential evolution algorithm.

Weight W in the estimator can indicate the output capacity, or peak efficiency, of the building facility. The Bias B in the estimator may refer to the electricity amount data according to past days or the electricity amount data according to the past season. The SR-Index Reference, which is the output, can indicate the output of the power usage instruction generator 100 of the building facility.

FIG. 5 is a graph illustrating a time shift of a building facility using energy usage instruction data generated by the apparatus 100 of FIG.

As shown in FIG. 5, the distributed control can be a control method for reducing the cost of moving the energy use of the daily expensive fare band to the cheapest time according to the time difference (3.5 times) rate. In Fig. 5, the horizontal axis represents the time during one day, and the vertical axis represents the electric power (electric energy).

The control method of the distributed control stores the energy based on the daily power consumption and the power rate forecast (e.g., the energy usage instruction data generated by the apparatus 100 of FIG. 1), and when the most expensive, It can mean a method of using it with efficiency. By using the control method of the distributed control, it is possible to reduce the energy usage fee by preliminarily using the unpredictable latent load that can be generated at an expensive fare (time zone). The solution of the distributed control may be Load Sharing, Optimum nonlinear Discharging, or Programing Logic Control (PLC), and the target facility of the distributed control may be the ice storage heat or the battery installed in the building.

Distributed control will be described in detail below. In order to lower the basic charge and the usage fee of the contract power plan, first, the equal control described below calculates the amount of power that can be lowered to the maximum by analyzing the maximum power per year in order to lower the peak power as a reference of the base charge, Power can be suppressed. Distributed control can then induce power to be used at a lower time to lower power usage fees.

The distributed control can be interlocked with the SR-Index. An example of a smart regulation (SR) -Index method is disclosed in Korean Patent Application No. 10-2011-0095135.

In the case of exceeding the level 2 or more determined by the SR-index, or when the threshold of the level 3 is not reached, a "0" step (normal operation) as an equal control mode is implemented. At this time, the indicator of the Smart Regulation Management System (SRMS) described with reference to FIG. 1 can be determined in a price-dependent manner. However, the conversion from the difference between the predictor and the indicator into the amount of electric power can be performed in the same manner as described below. Converting the difference between the predictor and the indicator to the amount of power can be done by distributing the appropriate ratio to the energy charge potential (for example, ice storage heat or rectifier) and controlling the charge based on the distributed value. That is, the discharge current value can be adjusted by adjusting the rectifier output voltage, and the ice storage heat output value can be adjusted by adjusting the ice storage temperature or the cooler valve value.

FIG. 6 is a graph illustrating the Smart Regulation Control of building equipment using the energy use instruction data generated by the apparatus 100 of FIG.

As shown in FIG. 6, the uniformity control can be a control that reduces the annual maximum peak power to suppress the base rate (7460 won per KW).

The control method of the uniform control can mean control that lowers the basic charge rate by calculating the maximum power consumption that appears 1 to 3 times a year and finding the optimum point that is acceptable and controlling it. The solution of the uniform control may be, for example, a linear control, a load sharing, or a golden section search method, and the target facility of the equal control may be a battery installed in the building, a generator, ice storage heat,

Equal control will be described in detail below. In order to lower the base charge of the contract power plan, the equal control calculates the amount of power that can be lowered to the maximum by analyzing the maximum peak power in order to lower the peak power, which is the standard of the base charge, have.

Equal control can be interlocked with the SR-Index described above. After passing the threshold of Level 3 determined by the Golden Pointing Algorithm, the mode implements the "1" step, which is the uniform control mode. At this time, it is possible to convert the difference between the predictor and the indicator into the electric power, divide it into the energy charge potential (for example, the ice storage heat or the rectifier) at an appropriate ratio, and follow up the charge based on the distributed value. That is, the discharge current value can be adjusted by adjusting the rectifier output voltage, and the ice storage heat output value can be adjusted by adjusting the ice storage temperature or the cooler valve value.

A control method in the case of deviating from the range of the distributed control or the uniform control in conjunction with the SR-index through the SRMT will be described as follows.

The Zero Base Reference of the SR-index uses the difference between Indicate and Estimated. The level ± 1 of the SR-index is the Energy Recharge Potential and is determined as Wide. Level ± 2 of the SR-index is determined by the Energy Saving Capable Potency for the pleasantness, and Level ± 2 can be stingered if it passes the threshold of Level ± 1 of the SR-index. Level 4 (generator operation) of SR-index is determined by separate Golden Pointing Algorithm. Level 3 of the SR-index is determined as the Energy Saving Capable Potential for the inconvenient level. However, Level 3 is a preparatory stage for Level 4 Work-in Time and consumers should not be prolonged because they can appeal to inconvenience.

FIG. 7 is a diagram showing an example of a simulation result of the distributed control and the uniform control described with reference to FIGS. 5 and 6. FIG.

An example of the electricity rate calculation method of the contracted power plan applied to the distributed control or the uniform control is as follows. Monthly Electricity Charge = Base Rate + Usage Charge = [Type 7430 * (Maximum Peak in Year)] + Rate to be used by time zone Peak suppression control (Equivalent control) is applied when the maximum peak is multiplied by the base unit price For example, in the case of the Yongsan KT building, the monthly peak of 1437kw ⇒ 1238kw is 10676 thousand ⇒ 9198 thousand won, saving 1,478 thousand won per month, and annual savings of about 17,743 thousand won, which is 2.9% Through distributed control, KRW 16,575,000 can be saved by a total of 2.7%.

The total savings of 5.6% = 31,151,000 won (total electricity cost of 615,741,000 won) can be achieved through equal control (peak control) and distributed control.

FIG. 8 is a diagram showing another example of a simulation result of the distributed control and the uniform control described with reference to FIGS. 5 and 6. FIG.

The above-mentioned equal control (or peak control) may be a control method for lowering the basic charge for one year (KW * 7430 won) as much as the peak power is reduced during the year. Distributed control can be a control method that directs the consumption of expensive time periods to less time.

FIG. 9 is a diagram showing a configuration of a building energy management system to which the present invention described with reference to FIG. 1 can be applied.

Referring to FIG. 9, the KT-BEMS (Katy-building energy management system) is an intelligent enterprise energy management system that performs a reduction of electricity through reduction of electric power consumption and a reduction of a charge through consumption pattern control (peak control or time shift) have.

The building energy management system is a smart energy management system (SEMS) that creates an energy equalization control algorithm that allows for prediction, indication, and tracking of energy usage, It can be an integrated management system that intelligently reduces the energy costs of a building through indexes (SR-Index) and energy management (KPI). An example of an integrated management system that intelligently reduces the energy cost of a building through energy management (KPI) is disclosed in Korean Patent Application No. 10-2011-0112984.

The building energy management system can be a statistical analysis and wide area management platform that integrates management of KT enterprise energy, and it can provide information of electric energy management system EEMS (existing N / W energy management system ELITE) and KT building energy management system (BAS) It can be the top platform to determine the optimal control policy of energy saving through statistical analysis of KT transfer energy and to provide information.

The building energy management system may include the following six algorithm methods.

First, Smart Regulation Management Theory (SRMT) is an intelligent control algorithm that can be used to control uniformity control, TIME shift, pattern control, SR-index, energy management (KPI) and BEMS (HVAC optimal control) , Prediction (Prediction), which predicts the energy of a building used in a specific future, Indication (Indication) for energy use, which can be created when optimally using energy resources based on predication, And may include control to cause the energy resource to estimate the intelligent control indication.

Next, the uniformity control (peak suppression) is a control algorithm that suppresses the base rate by lowering the peak power per year (7460 won per KW), calculates the maximum power usage that occurs 1 to 3 times a year, , And may be a control for lowering the basic charge rate by controlling the same.

Next, Time-Shift (Distributed Control) is an algorithm to reduce electricity charges by shifting the energy use of expensive fare band to the most inexpensive time of day according to the time difference (3.5 times) rate. It can be a way to store energy on a basis and use it at optimal efficiency when it is most expensive. More specifically, decentralized control can reduce energy bills by preemptively using unanticipated potential loads that can be incurred in expensive fare bands.

Next, Pattern (Behavior Based Control) is a technology (algorithm) that analyzes residents' behavior pattern, day, season, or weather and finds unnecessary energy consumption without inconvenience to residents and controls comfortably. , Or a bidet or the like, it is possible to perform intelligent control so as to match the correlation pattern of the related devices.

Next, the SR-Index is a programming indicator (algorithm) for controlling the system through the SRMT. The SR-Index is a programming indicator (algorithm) for regulating the information such as the past power usage, weather and weather forecast, current temperature, humidity, tenant, The power demand forecast of the day can be obtained, and the power data of the past one year can be analyzed to obtain the nine steps of SR-Index (Level) ± 1 to 4.

Finally, Energy Diamond is a Main Display method (algorithm) that allows you to view daily and monthly energy consumption of energy consumption details of individual building or group of buildings by a single screen. Can be displayed.

10 is a diagram showing an example of the simulation result of the uniform control of the building equipment using the energy use instruction data generated by the apparatus 100 of FIG. FIG. 10 shows the wave form of the performance of the uniform control.

The equivalent control (or peak control) shown in FIG. 10 may be a control that suppresses the base rate by lowering the maximum peak power per year (7460 won per KW), and a solution of equal control is linear control, load sharing , Or Golden Section Search Method, and the equipments to be controlled equally can be Energy Storage, Generator, Battery, Generator, Ice Storage, or Heat Storage Tank.

11 is a diagram showing an example of a simulation result of dispersion control of building equipment using energy use instruction data generated by the apparatus 100 of FIG. FIG. 11 illustrates a wave form of performance of the distributed control.

The distributed control shown in Fig. 11 can be a control for shifting the use time of the expensive fare band to the lowest cost according to the time zone differential (163.2 won ⇒ 45.5 won: 3.5 times), and the solution of the distributed control is Load The target facility of the dispersion control may be a water tank, a septic tank, a lower / middle water, a heat pump, a rectifier, a battery, an ice storage heat, a cooler, a storage tank, or the like.

FIG. 12 is a block diagram illustrating an apparatus 300 for generating a power usage guide of a building equipment according to another embodiment of the present invention.

Referring to FIG. 12, the power usage guidelines generating apparatus 300 of the building facility may be a more specific embodiment of the power usage guidelines generating apparatus 100 of the building facility shown in FIG. The power usage guide generation device 300 of the building facility can be implemented as a server.

The power usage guide generating device 300 of the building facility includes a power input instruction generating device 300 for building equipment to which the contract power plan is applied and includes an input unit 305, a demand predicting unit 310, and a guiding unit 340 do. The contract electricity rate may be calculated as the sum of the base fee and the usage fee, and the basic fee is a value obtained by multiplying the peak power (KW), which is the maximum power consumption for 15 minutes during the year, The price can be obtained by adding the value (for example, the contracted power + the maximum peak per year * 7790 won), and the usage fee can be obtained by the power per hour (KW) (or electricity consumption per hour) * per unit time (KRW / KW).

The input unit 305 receives past information on the amount of electricity consumed and the real time electric power charge used in the demand predicting unit 310 and the instruction unit 340 through the user of the electric power use instruction generating apparatus 300 or another device Can be input.

The demand predicting unit 310 estimates the past power consumption amounts of the building equipment before and after the current temperature outside the building equipment using a least square support vector machine and a differential evolution algorithm evolution algorithm can be applied to output the final predicted power consumption over a day's time of a building facility used for a future specific day after (immediately after) the current power consumption.

The demand prediction unit 310 may include a support vector machine unit 315, a dynamic model unit 320, a summation unit 325, a support vector machine A support vector machine update unit 330, and a model update unit 335.

The support vector machine unit 315 can generate the predicted power consumption amount of the future specific day and the predicted power consumption amount of the past one week before the future specific day by applying the least squares method to the past power consumption amounts.

The model update unit 335 measures a difference value between an actual power consumption amount of one week before a future specific day and a predicted power consumption amount of one week before the future specific day, A dynamic model that minimizes the value multiplied by the difference between the outside temperature outside the building for a period of two weeks and the outside temperature of the building one week before the future for a certain day (for example, the absolute value of the difference) Dynamic model value) can be generated.

The dynamic modeling unit 320 applies a difference value between the temperature average of seven days before (immediately before) the future specific date and the predicted temperature of the future specific day to the dynamic model (for example, Difference value) to output the amount of change in power consumption according to the temperature of the future specific day.

The summation unit 325 can output the final predicted power consumption amount by summing up the predicted power consumption amount generated by the support vector machine unit 315 and the power consumption amount variation output from the dynamic model unit 320 .

The support vector machine update unit 330 calculates an error value between the final predicted power consumption amount output from the summation unit 325 and the current power consumption amount of the building equipment and minimizes the error value and supplies the error to the support vector machine unit 315 May update the tuning parameters used by the < / RTI >

The guiding unit 340 is used for the future specific day based on the final predicted power consumption output from the demand predicting unit 310 (or the summation unit 325), the real-time electricity charge, and the accumulated electric energy of the building equipment The energy usage instruction data according to the time of day of the building equipment to be generated. The charging unit time of the real-time electricity charge (real-time electric charge) may be 15 minutes.

The instruction unit 340 includes an average operation unit 345, a multiplication unit 350, and a summation unit 355. [ The averaging unit 345 can calculate the final predicted power (final predicted power over time) by dividing the final predicted power consumption output from the demand predicting unit 310 (or the summation unit 325) by 24 hours.

The multiplier 350 multiplies the final predicted power, the real-time power charge, and the weight for the real-time power charge, and outputs the result.

The summing unit 355 sums up the output value of the multiplier 350 and the bias value (deviation value) indicating the peak power amount that can be lowered by using the accumulated electric energy of the building equipment, and outputs the energy use instruction data have.

13 is a functional block diagram corresponding to Fig. That is, Fig. 13 may correspond to Fig.

13 and 12, the estimator may include the demand predicting unit 310 and the guidance unit 340 of FIG. The LS-SVR included in the estimator is a Least-square Support Vector Machine corresponding to (corresponding to) the support vector machine unit 315 in FIG. 12, and is a component for performing prediction through regression analysis. c and sigma are the tuning parameters of the LS-SVR regression technique. E_k (or E (t-w (k)) is historical data input to the LS-SVR and may be the power consumption (power consumption) of the previous k-th previous week of the current day.

The LS-SVR can predict the future power consumption (EP (t), EP (tw (1)) using the past power consumption (E_k) (Tw (k)) may be the power consumption predicted over the LS-SVR of the previous kth previous week of the same day from the present.

The LS-SVR can be regression-regulated by using weekly data for each week by constructing an individual model considering characteristics of each day by using past data (E_k). Basically, it can be configured to use the same weekly data for the last 4 weeks. If the simulation result shows that the training data is longer than the period, the performance will be worse. It is possible.

The LS-SVR uses the power consumption data of E (tw (5)), E (tw (4)), E (tw (3) ) Last week. EP (t-w (1)) is used in the Model Updater corresponding to the model update unit 335 shown in Fig. 12 in the next step. The LS-SVR also uses the power consumption data of E (tw (4)), E (tw (3)), E (tw (2) You can predict the future specific day (Target Day) power consumption. EP (t) can be used in combination with the dynamic model included in Estimator which is a model optimized for temperature change.

The LS-SVR (least square method) will be described as follows.

The support vector machine (SVM) is a better algorithm for learning neural networks. It is a hyperplane or maximum-margin hyperplane satisfying the following equation (1) for classifying two classes in the binary classification problem, Decision boundaries). The constraint of the hyperplane is expressed by Equation (2) below.

[Equation 1]

&Quot; (2) "

Y = <w, x> + b = f (x), which is a linear function (linear discriminant function)

Equation (1) can be expressed as an optimization problem of an objective function having an optimal separation plane (Hyperplane) as in the following Equation (3) or a penalty term as in Equation (4).

&Quot; (3) &quot;

&Quot; (4) &quot;

는 슬랙(slack) 변수이다.In Equation (4), C is a penalty parameter or a cost coefficient for adjusting the relative size between the training error and the penalty point,

Is a slack variable.

Equation (4) can be expressed by Equation (5) below in the case of Support Vector Machine Regression (SVR).

&Quot; (5) &quot;

는 슬랙(slack) 변수이고, [수학식 5]는 [수학식 3]의 최적화 문제를 풀기 위하여 오차(슬랙(slack) 변수)를 포함하는 식으로서 SVR에서의 최적화 문제에 대한 라그랑제(Lagrange) 함수로 언급될 수 있다.In Equation (5)

Is a slack variable and Equation (5) is an equation including an error (slack variable) to solve the optimization problem of Equation (3) as a Lagrange for optimization problem in SVR. Can be referred to as a function.

The nonlinear training data of the input space are mapped to the kernel space by the kernel function and the optimal separation surface of the linear space in the feature space (high dimensional space) Can be found. It is SVM for Regression (regression) which is applied to the function approximation and can be called SVR. Describing in detail, a kernel function is used to extend the linear SVR as shown in Equation (5) to a nonlinear function. Also, the optimization problem as shown in Equation (5) can be solved simply by considering the dual problem (dual objective function). If Equation (5) is set as a pair problem, Equation (6) is obtained.

&Quot; (6) &quot;

), Gauss Kernel Function(RBF Kernel Function)(

), 또는 Sigmoid kernel function(

)일 수 있다. [수학식 6]을 만족하는 초평면을 구하는 식인 f(x)가 LS-SVR의 출력값(예측값)일 수 있고, x가 과거 전력 소비량 데이터일 수 있고, 상기 f(x)의 상수항 b 값은 Karush-Kuhn-Tucker 경계 조건들을 이용하여 결정될 수 있다. 상기 초평면은 고차원의 공간에서 최대 마진(maximum margin)을 갖는 선형 분리 초평면(linear separating hyperplane)을 의미할 수 있다.Equation (6) can be referred to as a dual problem of a Lagrange function to which a kernel function is applied. In Equation (6), k (x _i , x _j ) indicates a kernel function. The kernel function may be, for example, the Polynomial Kernel function (

), Gauss Kernel Function (RBF Kernel Function)

), Or Sigmoid kernel function (

). (X) may be an output value (predicted value) of LS-SVR, x may be past power consumption amount data, and the constant b value of f (x) may be Karush -Kuhn-Tucker boundary conditions. The hyperplane may refer to a linear separating hyperplane having a maximum margin in a high dimensional space.

Power consumption is closely related to changes in temperature (temperature). However, since the change in power consumption due to a sudden temperature change can not be predicted only with the LS-SVR, it is necessary to find out the relationship with the power consumption according to the change in temperature and to combine it with the dynamics indicated by the dynamic model And is performed by the model unit 320 and the model updater 335 indicated by DEA # 2.

First, the relationship between temperature and power consumption can be implemented in the form of a first-order model having a dynamic model (Dynamics) of the dynamic model unit 320 of FIG. 12 using a mathematical technique. Here, the actual environmental factor (temperature) varies depending on the situation, so that the dynamic characteristic of the model continuously changes. In order to solve this problem, a variable model, rather than a fixed model, is implemented through a model updater, which is a model update unit 335 of FIG.

A model that minimizes the error between last week's power consumption (EP (tw (1)) predicted by LS-SVR and actual power consumption last week (E (tw (1) And apply it so that it can be reflected in this forecast amount. In order to minimize this error, the parameters of the model are optimized by using a differential evolution algorithm (DEA) method, which is an optimization algorithm. The DEA in the estimator is a component that solves the optimization problem with the Differential Evolution Algorithm. T_k is a temperature outside the building (outdoor temperature) input to a dynamic model or a model updater, and may be the temperature of the previous k-th previous week of the same day based on the current day. The T_k may include a temperature outside the building of the current or future specific day (temperature forecast information of the future specific day).

The optimization procedure of the model using DEA (differential evolution algorithm) will be described as follows.

First, in order to obtain a temperature change that occurs more rapidly than other days, obtain the temperature variation by calculating the difference between the temperature average of the previous 7 days and the temperature distribution of Target Day (target day or future specific day) .

Second, for the same reason, the temperature change is obtained by calculating the difference between the temperature average of the previous 7 days and the temperature distribution of the same day of the previous week (Target Day - 7 days) from the same day of the last week.

Third, the difference between the last week's power consumption predicted by the LS-SVR and the actual last week power consumption is measured as an error, and the result of multiplying the measured error by the temperature change obtained from the second is minimized In the first model, a and b parameters are found.

Fourth, the optimized first model is applied to the temperature change of the target day last week using the temperature change obtained in the first procedure.

Fifth, the relationship between the predicted power consumption (EP (t)) estimated through LS-SVR and the relationship between the primary model and the temperature change is summarized and the result of the sum of the Prediction SVR and the model The final predicted power consumption (O_E_P) is obtained.

The optimization procedure of the dynamic model will be described in detail below.

Using the difference between the actual power amount data of one week before the target day and the predicted power amount predicted using the LS-SVR (based on the data from -5 weeks to -2 weeks), the dynamic model (Dynamic Model) can be obtained. The dynamic model can be obtained in the form of b / (as + 1) where each of a and b is tau (time constant) and Gain (gain). That is, each of b and a may be a parameter that determines the gain (Gain) of the transfer function between the external temperature and the power consumption, and Dynamics (dynamic model, dynamic model).

The dynamic model shows how the power consumption changes when the temperature changes by 1 占 폚, and it is obtained as shown in the following Equation (7). The dynamic model may include a temperature change that is a factor for the degree to which the temperature change of the reference date (target date) changes abruptly compared to other days.

&Quot; (7) &quot;

Error = (power consumption - predicted power consumption) × temperature change in 7 days before the future specific date = temperature change in the future specific day × dynamic model

The error between the predicted value and the actual power consumption is assumed to be caused by a temperature change (temperature change between 0 and 24 hours), and a dynamic model that minimizes the error in the above equation .

In other words, solving the optimization problem that minimizes the error, we solve the solution by using the optimized DEA (Difference Evolution Algorithm) and obtain the parameters a and b of the dynamic model do.

The dynamic model thus obtained is used for the target day and the amount of change in the power consumption amount according to the temperature change is obtained using the time unit temperature forecast of the target day and the dynamic model , A graph of the sum of the power consumption realized by the LS-SVR and the model according to the temperature change can be obtained in addition to the power consumption amount EP (t) obtained using the SVR. In addition, the dynamic model is a model predictive control that predicts a model using the relationship between temperature change and power consumption. Then, the dynamic model can be optimized by applying an error minimizing parameter using an updater.

O_E_P, which is the output of the estimator, is the predicted final predicted power consumption sum of the current day's LS-SVR and the dynamic model according to the temperature change, As shown in FIG. An example of a graph of O_E_P is shown in the upper right portion of FIG.

The above-described DEA (difference evolution algorithm) will be described in detail below.

DEA is an optimization technique that is an optimal control algorithm that optimizes key parameters to meet desired objectives. DEA differs from GA (genetic algorithm), which is one of the most widely used optimization techniques, because it does not encode entities into binary numbers and expresses the individuals constituting the population as vectors and generates new entities through their arithmetic operations. In other words, DEA is easier to implement than GA and can control the optimization process with fewer control factors.

The LS-SVR updater, which is the support vector machine update unit 330 of FIG. 12, optimizes C and Sigma (?), Which are the most important parameters in predicting using LS-SVR, and updates the prediction value to be a prediction value. Where C and sigma are SVM parameters, C is a cost coefficient, and sigma is a bandwidth function of the RBF function in a Gaussian radiator low function (RBF kernel function), which is a kernel function required to learn SVM Or kernel parameter).

The operation of the LS-SVR Updater is performed after the current power consumption is measured every hour, and the algorithm used for optimization is DEA (Differential Evolution Algorithm). E_T input to the LS-SVR Updater is the power consumption of the current day of the week.

When the current power consumption is measured every hour, the error is calculated by comparing with the predicted power consumption, and parameter tuning of C, Sigma is performed to minimize the RMS (root mean square) error value derived from the error. The RMS error value (RMSE) is calculated by the following equation.

When the parameter tuning is performed, a prediction value that is the most similar to the measured power consumption until the current time is newly calculated. This is because the target power consumption As shown in Fig.

In more detail, the LS-SVR Updater is used to extract and use parameters that minimize the error. That is, in the present invention, the parameters σ and C of the LS-SVR are optimized to minimize the error between the power consumption realized by the LS-SVR, the model according to the temperature change, and the actual power consumption data Can be applied.

The function blocks (constituent elements) included in the indicator corresponding to the instruction unit 340 of FIG. 12 will be described with reference to FIG.

The weight W is a user-defined tuning parameter for tuning the SR-Index Reference, and is a weight parameter for reflecting the real-time price trend and state in the SR-Index reference. The value of Weight (W) is a value reflected in the reference of SR-Index. It can be used as variables such as KPI (Key Performance Indicator).

Bias B is a user tuning parameter for tuning the SR-Index Reference. By applying bias to the result of considering the real-time price, the SR-Index Reference (Level) of the &lt; / RTI &gt; The bias value can be tuned by taking into account the amount of "available accumulated energy" of the object to which the estimator is applied and the total amount of "peak power" that can be lowered by using the accumulated energy amount. Bias (B) can be used as a parameter for optimization by lowering the peak power with the value of "the amount of accumulated energy available (ice storage capacity, battery, etc.)" for each national station.

That is, the weight W and the bias B may be tuning parameters of the estimator to adjust the SR-Index Reference result value. Weight (W) and Bias (B) vary from country to country and can be used by adjusting the basic data for optimization to the optimum (optimum) and adjusting the value according to the KPI index. The tuning parameters W and B can be designed to be automatically changed through an indicator (expected power consumption curve) to be derived in the future.

Consequently, the value O_SRIR output by the instruction unit 340 can be expressed by the following equation.

W (weighted) × (real-time price) * (1/24) * (final predicted power consumption per day) + B (bias)

The O_SRIR indicates a reference level of the current (or future specific) SR-Index, that is, a steady state level of the SR-Index, and the unit of the level may be, for example, KW or KWh. An example of a graph of O_SRIR is shown in the lower right portion of FIG.

The above-described present invention can be applied to the Smart Energy Management System (SEMS) shown in FIG. SEMS is a centralized management system that predicts usage information of energy through algorithms including LS-SVR, creates guidelines for optimal use, and generates control commands.

The electric power management system for communication (EEMS) in Fig. 9 is a system for managing communication water, power distribution and power supply facilities requiring supply of water, power distribution, The Building Automation System (BAS) is an intelligent system for existing building facility management systems such as HVAC (HVAC).

The Smart Energy Management System (SEMS) performs power demand forecasting for the day through regression analysis with information such as past power usage, weather forecast, current temperature, and specific date, -Index (Level) It is possible to determine 9 steps of ± 1 ~ 4. Smart Energy Management System (SEMS) notifies EEMS and BAS of the SR-Index (Index (index or level), Step, Predictor, Indicator) and EEMS and BAS can perform optimization control according to SR-Index .

Fig. 14 is a view for explaining another embodiment of the guidance unit 340 in Fig.

Referring to FIG. 14, input information O_E_P is a predicted power demand (consumption amount) O_E_P of a target day derived through the estimator of FIG. 13, and the unit may be KWh in units of 15 minutes.

The input information Object_E is set as a target power amount (used power amount) to a value equal to or less than the amount of power that can be accumulated through a charging facility in a building (building), and the unit may be KWh. The input information, Accuracy, is the power quantity accuracy when controlling the facility through the management system.

The input information p_price is the hourly electricity rate of the target day, the electricity rate (real-time electricity rate) per 15 minutes interval of the target day (target day), and the unit may be (won).

The mode information, which is input information, is mode decision information. When the power demand forecast value O_E_P of the target day exceeds the level 4 of the SR-Index, the mode information is mode 1 (maximum power demand minimization mode) , And mode 0 (mode of minimizing the electricity charge) if it is not exceeded.

The buffer information (Buf), which is input information, is a buffer (%) to reflect the storage efficiency and the like, and to prepare for the loss due to the uncertainty of the situation or the unexpected instantaneous use Buffer information.

Indicator_out, which is output information through the output unit, is used as the SR-Index of the day at the same time as the optimum power demand amount from 00:00 to 24:00 on the target day, and the result can be derived at 00:00 every day.

The maximum power demand minimization calculation unit 410 included in the indicator corresponding to the instruction unit 340 receives (analyzes) the O_E_P, Object_E, Accuracy, and p_price information input through the input unit 405, If the forecasted power demand of the work exceeds the level 4 of the SR-index, the usage guidelines (usage pattern) of the building facilities can be generated so that the maximum power of the building equipment is below the level 4 of the SR-index. That is, the calculation unit for minimizing the maximum power demand can generate an indicator for the mode 1 that minimizes the maximum power demand aiming at the maximum peak suppression.

The power amount fee minimization calculation unit 415 included in the indicator corresponding to the instruction unit 340 receives the inputted O_E_P, Object_E, Accuracy, and p_price information so that the predicted power demand value of the target day is the level ) Less than 4 (less than) can generate usage instructions for building equipment to minimize power charges for building equipment. That is, the calculation unit for minimizing the amount of electricity charges can generate an indicator for the mode 0 that minimizes the amount of electricity using the energy stored in the expensive period.

According to the mode information, any one of the output value of the maximum power demand minimization calculation unit and the output value of the power amount charge minimization calculation unit may be selected and output as Indicator_out which is output information.

The power storage efficiency maximization calculation unit 420 included in the indicator corresponding to the instruction unit 340 receives the input O_E_P, Object_E, Accuracy, and p_price information to maximize the storage efficiency (charge efficiency) of the power storage facility (Eg, temperature) can be calculated and output. The output value may be output as Indicator_out which is output information together with a storage ratio ratio to the target power amount.

FIG. 14 is a block diagram illustrating a method of generating an intelligent energy consumption guidance according to an embodiment of the present invention. The intelligent energy consumption guideline generation method is applied to a building energy consumption guideline generating indicator 400 and the building energy consumption guideline generating indicator 400 includes an input unit (MUX) 405, a mode selection unit 410, a charge minimization calculation unit 415, a power storage efficiency maximization calculation unit 420, a selection unit, and an output unit MUX can do.

FIG. 15 is a view for explaining an embodiment of the SR-Index control method using the output value of the guidance unit 340 of FIG.

Referring to FIG. 15, as shown in the left drawing of FIG. 15, when it is necessary to perform distributed control for a building facility as a result of predicting a power consumption amount of a future specific day (for example, level) of 4 or less and the power charges of the building facility need to be minimized), the indicators included in the power usage guidance generating apparatus 300 of the building facility may be used as the indicators of the building facilities SR-Index The SR-Index control can be performed on the control target.

The level 4 can be determined by a day peak power determination method (or Golden Pointing Algorithm) in a building facility. The level 4 (the reference peak power amount) is calculated by taking into consideration the power cost based and the power amount based reduction factors for the daily power consumption data in the building facility for a specific period before the future specific day, Calculating a total energy cost curve in the building facility by calculating a curve (power consumption reduction value) and reflecting the basic rate reduction factor, the generation cost factor, and the electricity rate reduction factor And using the total energy cost curve. The power cost based reduction factor may mean a reduction of power cost by means of a dispersion indication or the like shown in the lower right part of FIG. 10, and the reduction factor based on the power amount may be behavior based scheduling control or upper, It can mean reduction of electric power by pattern control such as water tank pattern scheduling control.

The method of determining the reference peak power using the total energy cost curve may comprise calculating the base peak power in the building facility using the power usage reduction curve (using the daily power usage or the peak power amount indicated in the power usage reduction curve) Calculating a power generation cost curve at a building facility determined based on a daily power generation amount of a generator that operates to reduce the base rate using the basic charge saving curve, And calculating a usage charge reduction curve in the building facility, the usage charge reduction curve being determined according to the amount of usage and the daily generation amount, and using the basic charge reduction curve, the generation cost curve, and the usage charge reduction curve Cost curve, and usage charge reduction curve) at the future specific day It can be a process for determining a peak power basis. The power generation cost may mean a cost determined by the amount of oil consumption and the unit price of oil according to the operation time of the generator.

A method for determining a reference peak power amount at a future future date using the base charge reduction curve, the electricity generation cost curve, and the usage charge reduction curve may comprise: using the base charge reduction curve, the electricity generation cost curve, Calculating a total energy cost curve in the building facility and determining a power value corresponding to the minimum energy cost on the total energy cost curve as the reference peak power amount.

The basic charge saving curve may be a curve for a basic charge change amount according to a decrease in the daily electric power consumption during the specific period, and the electric power generation cost curve may be a curve for the generator fuel oil change amount according to the daily electricity generation amount during the specific period The curve for the change in the cost of the generator due to the decrease in the daily power consumption during the period, and the curve for the usage charge reduction is a curve for the change in the usage charge according to the variation in the daily power generation cost during the specific period A curve for the amount of change in the usage charge according to the daily power consumption for a specific period using the difference between the usage fee corresponding to the daily power consumption and the generation cost). The specific period may be one year prior to the future specific date.

In another embodiment of the present invention, the method of determining the peak power on the same day is a method of determining a peak power reduction during a predetermined period (for example, from a day before to one day before a day on the same day) Calculating a base charge reduction curve (a base charge reduction value) according to the peak power reduction curve according to the peak power reduction during the predetermined period, Calculating a power generation cost curve generated by at least one or more generators provided in the building operating to reduce peak power for the predetermined period of time, Obtain a building energy cost curve from the curve, the utilization rate reduction curve, and the power generation cost curve sum , And the power value corresponding to the minimum value of the energy cost of building the curve can be determined by the day of peak power. The peak power on the same day may be a power corresponding to level 4.

16 is a view for explaining an intelligent energy consumption predicting method according to another embodiment of the present invention.

16, a building energy consumption predicting apparatus 500 for managing power usage and maximizing distribution efficiency of a building includes a least square support vector machine unit (LS-SVR Part) 502 and a compensator part 504). The intelligent energy consumption predicting method may be applied to the building energy consumption predicting apparatus and may perform an operation similar to the operation (operation or configuration) of the demand predicting unit 310 described with reference to Figs. 12 and 13.

)와 Kernel 함수로 사용된 Gaussian Function의 Width Parameter인 σ의 조정(tuning) 절차가 필요할 수 있다. 상기 감마(

)는 전술한 비용계수(C)의 1/2일 수 있다.The least squares support vector machine part (LS-SVR Part) 502 can be used to obtain a first-order prediction model of a building energy consumption predicting device, and it can calculate the first- A function of deriving a prediction model can be performed. In the intelligent energy consumption forecasting method, the penalty constant gamma (), which has an important influence on the degree of error of the LS-SVR prediction model construction

) And the tuning procedure of the Width parameter of the Gaussian function used in the kernel function may be necessary. The gamma (

) May be 1/2 of the cost coefficient C described above.

A genetic algorithm (GA), a DEA, an immune algorithm (IA), a particle swarm optimization (PSO), a genetic algorithm (a kind of difference evolution algorithm), and a golden partition search algorithm GSS-RCGA, or a golden split search and real coding genetic algorithm), or multiple golden split search algorithms (Multi-GSS) which can be a genetic technique (a kind of difference evolution algorithm). The genetic algorithm can be a technique of generating an initial population, evaluating fitness according to an objective function, selecting a fitness object (dominant), and generating a new object through a crossover and a mutation process.

The GSS-RCGA can be realized through the fusion of RCGA (Real-coded genetic algorithm) and GSS (Golden Section Search) based on the genetic fact that hybrid crossing is more advantageous in natural selection among crossing populations than hybridization, The GSS-RCGA procedure may be a different technique of RCGA, crossover, mutation, and fitness evaluation. The convergence rate and the convergence rate of the GSS-RCGA are superior to those of the Multi-GSS among the four global optimization techniques (Multi-GSS, GSS-RCGA, RCGA, and DEA).

The Multi-GSS is a method that extends the search level of the GSS method and uses the GSS method in a two-dimensional search space or more. In the Multi-GSS using the multiple GSS, Can be implemented to find the global optimum in the search space. In a Multi-GSS, one GSS is passed through three GSSs in the course of a generation, and the number of generations of each GSS can have one-third of the number of individuals N set. Multi-GSS can be the best among the four global optimization techniques (Multi-GSS, GSS-RCGA, RCGA, DEA) and convergence rate of data. GSS-RCGA → Multi-GSS → RCGA → DEA may be listed in the order of lower RMSE (Root Mean Square Error) between the power demand of the target day and the predictive model.

The two global optimization methods using GSS (Golden Section Search) are as follows. The first method, GSS-RCGA, is a hybrid method of GSS and RCGA (Real-coded genetic algorithm). The second method, Multi-GSS, And can be extended to a wide area search method through the configuration. The GSS-RCGA is referred to as JS Kim, JH Jeon, and H. Heo, Hybrid DSO-GA-based sensorless optimal control strategy for wind turbine generators, Journal of Mechanical Science and Technology 27 (2) (2013) 549-556 It is possible.

Both of these methods have advantages of complexity of computation, reduction of operation load, and reduction of the rigorous modeling process of the target system, since they do not use differential information and differential procedure for searching for an optimal point. The speed and the quality of the solution can have excellent advantages.

The least squares support vector machine part (LS-SVR Part) 502 includes a first least squares support vector machine portion (LS-SVR) 505, a second least squares support vector machine portion (LS-SVR) A first Golden Section 515, a second Golden Section 520, and a Real-time Updater 525. The first Golden Section 515, the second Golden Section 520,

The first least squares support vector machine 505 computes the least squares support vector machine regression algorithm LS-SVR (E-SVR) for the historical power consumption of the building equipment (e.g., E_1, E_2, E_3, E_4) ) To generate the current predicted power consumption of the building facility. Each of E_1, E_2, E_3, and E_4 may be power consumption data five days ago, power consumption data four days ago, power consumption data three days ago, and power consumption data two days ago. In another embodiment of the present invention, the historical power consumption may be one year of data prior to the future specific date.

The second least squares support vector machine 510 computes a minimum squared support vector E_T for the current estimated power consumption amount and the current power consumption E_T of the building equipment, The predicted power consumption of future specific days (immediately after the current power consumption of the building facility) using the machine regression algorithm (LS-SVR) (predicted power consumption Or a time-consuming power consumption pattern (O_E_P_NO_Temp). E_T may be power consumption data of a target day.

The first golden section section 515 may perform a real coding derivation and a golden partition search algorithm (GSS-RCGA) or a multiple golden search algorithm (ESS) on the generated current predicted power consumption amount and the past power consumption amount E_5 (C, sigma) used in the LS-SVR of the first least squares support vector machine section can be updated by applying (applying) the partial search algorithm (Multi-GSS). The tuning parameter may be calculated by calculating an error value between the current predicted power consumption amount (or the past predicted power consumption amount of the current power consumption amount one day before) and the past power consumption amount E_5 of the current power consumption amount one day before, and minimizing the error value have.

The second golden section 520 may perform a real coding derivation and a golden partition search algorithm (GSS-RCGA) on the generated predicted power consumption of the future specific day and the current power consumption (E_T) Or the tuning parameter (C, sigma) used in the LS-SVR of the second least squares support vector machine section may be updated using a multiple golden partition search algorithm (Multi-GSS). The tuning parameter may calculate an error value between the current predicted power consumption amount and the current power consumption amount E_T, and minimize the error value.

The compensating unit 504 applies a least squares support vector machine regression algorithm LS-SVR and a difference evolution algorithm DEA to the past power consumption of the building facility and the past temperature outside the building facility outside the building temperature (The amount of change in the predicted power consumption amount depending on the time and temperature during the day of the building equipment used in the future on a specific day) according to the temperature of the future specific day. Further, the compensating unit may reflect (for example, sum) the predicted power consumption change amount to the predicted power consumption amount of the future specific day generated by the second least squares supporting vector machine unit, (O_E_P) can be output (generated).

Wherein the model update unit included in the compensation unit measures a difference value between an actual power consumption amount of one week before the future specific date and a predicted power consumption amount of one week before a future specific date, A dynamic model can be generated that minimizes the value multiplied by the difference between the 7-day average outside the building and the outside temperature of the building one week before the future, two weeks before the week. The dynamic model unit included in the compensation unit may apply a difference value between a temperature average of 7 days before a future specific day and a predicted temperature of a future specific day to the dynamic model so as to output a power consumption change amount according to a temperature of a future specific day ). Each of T_1 to T_5 inputted to the compensating unit 504 is a temperature outside the building five days ago, four days before the building, three days before the building, two days before the building, and one day before the building .

The operation (operation or configuration) of the compensator described above is similar to the operation of the estimator described with reference to Figs. 12 and 13, and therefore, the description of the operation (function) .

In another embodiment of the present invention, the compensation unit may calculate a minimum square support vector machine for the power consumption of the building equipment according to the temperature before the specific day in the future, or the power consumption of the building equipment according to the humidity before the specific day in the future, A regression algorithm (LS-SVR) and a differential evolution algorithm (DEA) can be used to generate a predicted power consumption variation according to the temperature of future specific days. Further, the compensating unit may reflect (for example, sum) the predicted power consumption change amount to the predicted power consumption amount of the future specific day generated by the second least squares supporting vector machine unit, (O_E_P) can be output (generated).

A method for generating a predicted power consumption variation according to a temperature of a future specific day using a power plant quantity data of a building facility according to a temperature before a specific day in the future by the LS-SVR and the difference evolution algorithm, Each of the power amount data for the previous one day is multiplied by a function value including a temperature corresponding to a predicted temperature for a specific future date and the power amount data for a future specific day, And a method of generating a predicted power consumption variation according to the temperature by obtaining an average value.

The operation (function or configuration) of the compensator described above is similar to that of the estimator described with reference to Figs. 1, 2, and 4, so that the description of the operation (function) An operation description can be referred to.

The least squares support vector machine (LS-SVR or LS-SVM) described above will be described as follows. The LS-SVM can be used to create a line function that can represent a non-linear function.

The LS-SVM can be modeled using the following Equation (1-1) to estimate a nonlinear function representing the exact correlation between the input data and the output data. The above equations can be expressed in terms of a weight vector, a basis function, and a bias (Bias).

[Mathematical expression 1-1]

The LS-SVM is to accurately estimate [Equation 1-1] when a training data set is given. Therefore, when the formula is changed from the expression (1-1) to the expression using the basis function, the output data can be defined by the following expression (1-2).

[Equation 1-2]

The loss function of the normalized least-squares method consists of minimization problem through w, b, and e _i and is expressed by Equation 1-3 below.

[Equation (1-3)

Through the conditional part of [Equation 1-3], it can be seen that the LS-SVM uses the inequality constraint of the SVM instead of the equation condition unlike the existing SVM.

)는 0 보다 큰 페널티(penalty) 상수 또는 정규화 상수로서, bias-variance trade-off(학습 오차)를 제어하기 위해 사용된다.The gamma of (Equation 1-3)

) Is a penalty constant or normalization constant that is greater than zero and is used to control the bias-variance trade-off.

In order to solve the minimization problem with the condition as shown in [Equation 1-3], a minimization problem with a constraint condition can be efficiently solved by applying a Lagrange multiplier to [Equation 1-3] -4].

[Equation (1-4)

은 Lagrange multiplier이다. Karush-Kuhn-Tucker(KKT) 조건에 의해, [수학식 1-4]가 최적이 되기 위한 First-order 조건은 아래의 [수학식 1-5]와 같다.In Equation 1-4,

Is a Lagrange multiplier. According to the Karush-Kuhn-Tucker (KKT) condition, the first-order condition for optimizing [Equation 1-4] is given by Equation 1-5 below.

[Mathematical expression 1-5]

The components or parts used in the present embodiment may be software such as a task, a class, a subroutine, a process, an object, an execution thread, a program, a field programmable gate array (FPGA) Or an application-specific integrated circuit (ASIC), or a combination of the above software and hardware. The components or parts may be included in a computer-readable storage medium, or a part of the components may be distributed to a plurality of computers.

As described above, the embodiments have been disclosed in the drawings and specification. Although specific terms are used herein, they are used for the purpose of describing the present invention only and are not used to limit the scope of the present invention described in the claims or the claims. It is therefore to be understood by those skilled in the art that various modifications and equivalent embodiments are possible in light of the present invention. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

105: Demand forecasting section
110: Instructions section
205: Demand Forecast Phase
210: instructions generation step
310:
315: Support vector machine part
320: Dynamic model part
325:
330: Support vector machine update unit
335: Model update section
340: Instructions section
345:
350:
355:

Claims (5)

In an intelligent energy consumption predicting method,
(a) The first least squares support vector machine included in the building energy consumption predicting device uses the least squares support vector machine regression algorithm (LS-SVR) for the past power consumption of the building equipment Generating a current predicted power consumption amount of the building facility; And
(b) a second least squares support vector machine of the building energy consumption predicting device calculates a least squares support vector machine regression for the current predicted power consumption amount and the current power consumption amount of the building equipment, Generating a predicted power consumption amount for a future specific day after the current power consumption amount of the building equipment using an algorithm (LS-SVR)
The method comprising the steps of: The intelligent energy consumption prediction method according to claim 1,
(c) a first golden section of the building energy consumption predicting apparatus calculates a real-valued coding genetic and golden partition search algorithm (GSS-RCGA) for the generated current predicted power consumption and the past power consumption of the current power consumption one day ago, Updating a tuning parameter used in the LS-SVR of the first least squares supporting vector machine using a multiple golden partition search algorithm (Multi-GSS); And
(GSS-RCGA) or multiple gold (GSS-RCGA) for the current power consumption; and (d) a second golden section of the building energy consumption prediction device, Further comprising updating the tuning parameters used in the LS-SVR of the second least squares support vector machine section using a partitioned search algorithm (Multi-GSS). The intelligent energy consumption prediction method according to claim 1,
(c) a compensating unit of the building energy consumption predicting apparatus calculates a least squares support vector machine regression algorithm (LS-SVR) and a difference evolving algorithm (LS-SVR) for the past power consumption of the building equipment and a past temperature before the present temperature outside the building equipment Generating a predicted power consumption variation according to a temperature of the future specific day by applying an algorithm (DEA); And
(d) a compensating unit of the building energy consumption predicting apparatus calculates a final predicted power consumption amount of the building equipment by reflecting a predicted power consumption change amount generated in the step (c) to a predicted power consumption amount of a future specific day generated in the step (b) And outputting the power consumption. The method of claim 1, wherein the step (c)
(c1) the model update unit included in the compensation unit measures a difference value between an actual power consumption amount of one week before the future specific date and a predicted power consumption amount of one week before the future specific date, A dynamic model is generated that minimizes a value obtained by multiplying the 7-day average of the outside temperature of the building included one week before and one week before the future specific day by the difference between the outside temperature of the building one week before the future specific day ; And
(c2) the dynamic model unit included in the compensation unit applies a difference value between the temperature average of 7 days before the future specific date and the predicted temperature of the future specific day to the dynamic model, And outputting a change amount of the power consumption amount. The method according to claim 1,
Wherein the historical power consumption is one year of data prior to the future specific day.

Images