KR102187327B1 - Dynamic management and control system for a building electric demand based on automated machine learning scheme - Google Patents

Abstract

The present invention relates to a system for optimal management and control of a load of a building based on automated machine learning and, more specifically, to a system for optimal management and control of the load of a building based on automated machine learning which probabilistically predicts the amount of generation and demand, plans an optimal schedule of an energy storage device based on a prediction result, and manages the load of the building according to a planned schedule.

Description

Building load optimal management and control system based on automatic machine learning {DYNAMIC MANAGEMENT AND CONTROL SYSTEM FOR A BUILDING ELECTRIC DEMAND BASED ON AUTOMATED MACHINE LEARNING SCHEME}

The present invention relates to a system for optimal management and control of building loads based on automatic machine learning, and more particularly, in a building using both power generation facilities, system facilities, and energy storage devices, probabilistic prediction of power generation and demand, and based on this The present invention relates to an automatic machine learning-based building load optimal management and control system that plans the optimal schedule of an energy storage device and manages the building power load according to the planned schedule.

As recent buildings become larger and more functional, various facilities such as air conditioning, sanitation, power, lighting, crime prevention, disaster prevention, and water treatment are installed in a complex manner, and various building facility management automation systems are proposed to detect and control these facilities. Has become.

The building facility management automation system controls the operation of field devices and field devices that make up various facilities such as a central control device that manages and controls the overall system, machine room, electrical room, and lighting equipment, and collects information of field devices. It consists of a remote control panel that analyzes and responds, and transmits the collected and analyzed information to the central control unit.

The remote control panel controls field devices under the control of a central control system, but can be operated independently so that the operation of the field devices can be independently controlled.

There is a need for a change to an intelligent building (IB) that can maximize productivity and maintain a pleasant office environment.

Korea has experienced a severe power shortage in recent years, and in accordance with demand management policies, measures are being taken to reduce power consumption at the national level as well as the private sector.

Accordingly, managing and reducing building energy, which accounts for a significant portion of the nation's total energy use, has become an increasingly important issue, and the introduction of the Building Energy Management System (BEMS) has become mandatory.

BEMS is an integrated system of measurement, control, management, and operation that provides an optimized energy management plan for buildings by monitoring energy use for efficient energy management and maintaining a pleasant indoor environment of buildings.

Building automation systems (BAS) that automate such intelligent buildings require real-time monitoring and control.

The building automation system (BAS) is a system for remotely and automatically controlling the operation of power equipment in a building such as lighting and heat source equipment, but is usually equipped with a simple interlocking logic or a predetermined schedule control logic (scenario).

Building automation systems (BAS) are often left in a scenario-based control state without a dedicated person, and there is a problem that the overall efficiency decreases over time due to full dependence on the initial scenario, and each building (installation) to be applied prior to control IOT sensor, heat source and power facility configuration, members, etc.) have the inconvenience of requiring verification and calibration.

In addition, the building automation system (BAS) is not equipped with an automated control algorithm, and thus is controlled by the subjective judgment of the manager. Therefore, there is a problem in that it is not possible to optimally control devices while saving energy automatically.

For example, sudden power failure, sudden change in power demand due to the addition of high-power devices, changes in demand power trends in regions and buildings by year, changes in users, addition of eco-friendly energy sources, and immediate dynamic response to environmental changes I have this difficult problem.

In Korean Patent Publication [10-2016-0022487], a maximum power demand management system is disclosed.

Korean Patent Publication [10-2016-0022487] (Publication date: March 2, 2016)

Accordingly, the present invention was conceived to solve the above-described problems, and an object of the present invention is to probabilistically predict the amount of power generation and demand in a building using both power generation facilities, system facilities, and energy storage devices, and It plans the optimal schedule of the energy storage device and provides an automatic machine learning-based building load optimal management and control system that manages the building power load according to the planned schedule.

Objects of the embodiments of the present invention are not limited to the above-mentioned objects, and other objects not mentioned will be clearly understood by those of ordinary skill in the art from the following description. .

The automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention for achieving the above object is a power generation amount probabilistically predicting the power generation amount of the self-power generation device 10 based on machine learning. A prediction unit 100; A demand forecasting unit 200 for probabilistically predicting the power demand of the demand facility 40 based on machine learning; Including the amount of electricity generated by the generation amount predicting unit 100, the amount of electric power demand predicted by the demand amount predicting unit 200, the existing peak load, the power rate plan, and the state value information of the power storage device 30 A power storage device 30 that can be minimized or a schedule generator 300 that generates an optimal schedule of a building load; And the self-powered device 10 and the power system 20 so that charging and discharging of the power storage device 30 is controlled according to the optimal schedule of the power storage device 30 generated by the schedule generator 300. It characterized in that it includes; a power control unit 500 for controlling the electrical connection between the power storage device 30 and the demand facility 40.

In addition, the demand forecasting unit 200 is characterized in that a plurality of methodologies selected from a time series model, an automatic pattern classification technique, and an artificial neural network model are dynamically combined according to time-specific accuracy to periodically generate a final prediction value for each predetermined time. do.

In addition, the demand forecasting unit 200 has a methodology for extracting a pattern with the highest similarity to the automatic load pattern classification through the K-mean method from a preset input element, and learning an artificial neural network through the corresponding input element and predicting through it. It is characterized in that a final prediction result is generated by dynamically collecting the methodologies according to prediction accuracy.

In addition, the demand forecasting unit 200 is the following equation,

는 최종 결과,

는 클러스터링모델의 결과,

는 인공신경망모델의 결과, A는 클러스터링모델의 과거 데이터의 정확도 B는 인공신경망모델의 과거 데이터의 정확도)(here,

Is the final result,

Is the result of the clustering model,

Is the result of the artificial neural network model, A is the accuracy of the past data of the clustering model, B is the accuracy of the past data of the artificial neural network model)

By, the clustering model and the artificial neural network model each have a weight according to the accuracy of the past data to derive the final result.

In addition, the schedule generation unit 300 is the following equation,

The charging/discharging schedule of the power storage device 30 is generated on the basis of.

In addition, the power control unit 500 is a driving characteristic alarm system that alarms when a pre-defined operation specific state occurs in the self-powered device 10, the power storage device 30, and the demand facility 40. It is characterized in that it generates an alarm according to.

In addition, the power control unit 500 acquires demand characteristics of power users and determines a Learning Model function using a data set collected by label processing characteristic items that become reference values. And, based on the value calculated by using this, it is characterized in that the electrical connection between the self-powered device 10, the power system 20, the power storage device 30, and the demand facility 40 is controlled.

In addition, the automatic machine learning-based building load optimal management and control system includes the power generation amount predicted by the power generation amount prediction unit 100, the power demand amount predicted by the demand amount prediction unit 200, and the schedule generation unit 300. An economical analysis unit 400 that calculates a change in the predicted power load according to the optimal schedule of the power storage device 30 generated by and applies a charge thereto to calculate economical efficiency; further comprising, the power storage device ( 30), the optimal schedule generation and economical calculation are repeatedly performed, and the power storage device 30 and the demand facility 40 control schedule candidates are continuously updated through an optimization technique, and a control schedule in which the lowest charge is induced is generated. It is characterized in that it controls the conventional power storage device 30 and the demand facility 40.

In addition, the automatic machine learning-based building load optimal management and control system includes a facility control unit 900 capable of controlling the demand facility 40, and the facility control unit 900 includes a machine field control unit, a power field control unit, and lighting. It characterized in that it comprises at least one selected from the field control unit, SI (System integration) field control unit, and BMS (building energy management system) field control unit.

According to the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention, the present invention is different from the existing peak load controller (fixed control) based on the existing simple logic (load cutoff when a specific load is close). By learning to predict the load and power generation, and by generating and controlling (dynamic control) the optimal energy storage device or load schedule that can suppress the peak load in real time based on this, there is an effect that it is possible to optimally manage the power load of the building. .

That is, in buildings that use both power generation facilities, system facilities and energy storage devices, by probabilistically predicting the amount of power generation and demand, planning the optimal schedule for energy storage devices based on this, and managing the building power load according to the planned schedule. In addition, the power-related factors that occur every moment are optimized in real time to maximize the efficiency of device use and increase the profits of general operators, while minimizing the country's electric energy cost.

In addition, the problem with the existing fixed control of the logic method is that the performance fluctuations are very severe depending on how much the reference value is set, and this reference value must be set by a person in the end and cannot reflect the changing building load situation. The invention predicts the future, dynamically generates the most optimal scenario, and continuously updates the prediction and scenario generation, so that it is possible to actively respond to changes in building demand patterns.

In addition, it is possible to flexibly respond to environmental changes through machine learning-based prediction algorithms.

In addition, the clustering model (K-mean technique) and the artificial neural network model are dynamically collected to generate the final prediction result, thereby correcting the variance error for each time period, thereby improving the performance of the prediction model.

In addition, the clustering model and the artificial neural network model are given weights according to the accuracy of the past data to derive the final result, thereby further improving the performance of the predictive model through error correction.

In addition, by using an objective function, a constraint condition, and an equation for calculating a battery cost, there is an effect of generating an optimal schedule for the power storage device.

1 is a conceptual diagram of an automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention.
2 is a flowchart showing an example of performing machine learning based on past data and predicting demand based on it.
3 is a conceptual diagram of an automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention in which an economical analysis unit is added to FIG. 1.
4 is a conceptual diagram of a building load optimal management and control system based on automatic machine learning according to an embodiment of the present invention in which a facility control unit is added to FIG. 1.

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to a specific embodiment, it should be understood to include all changes, equivalents, and substitutes included in the spirit and scope of the present invention.

When a component is referred to as being "connected" or "connected" to another component, it is understood that it may be directly connected or connected to the other component, but other components may exist in the middle. Should be.

On the other hand, when a component is referred to as being "directly connected" or "directly connected" to another component, it should be understood that there is no other component in the middle.

The terms used in the present specification are only used to describe specific embodiments and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In the present application, terms such as "comprise" or "have" are intended to designate the presence of features, numbers, processes, operations, components, parts, or combinations thereof described in the specification, but one or more other features. It is to be understood that the possibility of addition or presence of elements or numbers, processes, operations, components, parts, or combinations thereof is not preliminarily excluded.

Unless otherwise defined, all terms including technical or scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs. Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning of the related technology, and should not be interpreted as an ideal or excessively formal meaning unless explicitly defined in this application. Does not.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings. Prior to this, terms or words used in the specification and claims should not be construed as being limited to their usual or dictionary meanings, and the inventors appropriately explain the concept of terms in order to explain their own invention in the best way. Based on the principle that it can be defined, it should be interpreted as a meaning and concept consistent with the technical idea of the present invention. In addition, unless there are other definitions in the technical terms and scientific terms used, they have the meanings commonly understood by those of ordinary skill in the art to which this invention belongs, and the gist of the present invention in the following description and accompanying drawings Descriptions of known functions and configurations that may be unnecessarily obscure will be omitted. The drawings introduced below are provided as examples in order to sufficiently convey the spirit of the present invention to those skilled in the art. Accordingly, the present invention is not limited to the drawings presented below and may be embodied in other forms. In addition, the same reference numbers throughout the specification indicate the same elements. It should be noted that the same elements in the drawings are indicated by the same reference numerals wherever possible.

1 is a conceptual diagram of a system for optimal management and control of building loads based on automatic machine learning according to an embodiment of the present invention, and FIG. 2 is a flow chart showing an example of performing machine learning based on past data and predicting demand based on this 3 is a conceptual diagram of an automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention in which an economical analysis unit is added to FIG. 1, and FIG. 4 is an invention in which a facility control unit is added to FIG. A conceptual diagram of an automatic machine learning based building load optimal management and control system according to an embodiment of

The automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention optimizes in real time factors related to demand management and automatic building control related to electricity use and supply that occur every moment, so that equipment use efficiency In order to maximize the cost of electric energy and minimize the cost of electric energy in the country while maximizing the profits of the operator, the demand management and building automatic control system, self-generation equipment, and demand equipment are interconnected or cut off, and the optimal charging of power storage equipment is provided. It is a building power load optimal management and control system that creates and controls a discharge schedule.

In addition, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention refers to a system having a function of managing a demand management and an automatic building control system. Demand management refers to controlling the operation state of whether all devices that accept and use electricity are electrically operated, and automatic building control is subdivided into a mechanical field, an electric power monitoring field, and a lighting control field, but one embodiment of the present invention In the following, examples were given focusing on the field of machinery.

However, the field of application of the automatic building control of the present invention is not limited to machines, and in the present invention, the building (BUILDING) and the building are used under the same concept, and the building automatic control management system is also applied to industrial facilities such as factories. Can be applied.

In addition, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention can remotely automatically control the operation of electric power equipment in the building, such as lighting and heat source equipment.

In addition, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention automatically generates in-building power device control values in real time to minimize power bills in the building, and optimizes devices by interlocking with the BAS. It can be controlled or mounted on the BAS.

In particular, in order to dynamically determine the optimal control schedule of devices whose current control values affect future controllable values due to storage capacity constraints, such as power storage devices (ESS), the power demand of buildings is predicted in real time. The main function can be the function of automatically generating the optimal schedule within the given physical constraints and updating it repeatedly in consideration of the contracted rate plan and interoperability with other devices.

To this end, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention interacts with the BAS to manage the energy facility environment of the building and collect operation data for efficient operation, For this, hardware (H/W) and software (S/W) interfaces may be provided.

In addition, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention includes data such as building power load and solar power generation for a certain period of the past, weather information for the period, days of the week, holidays, etc. It receives the data from and performs machine learning internally, and based on this, it has the function of predicting the power load and solar power generation amount for a specified period in the future. Also, based on the predicted value and the contracted power rate plan of the building, the optimal input schedule for control devices such as ESS is automatically generated. In addition, it is possible to provide a communication function and a S/W interface to interwork with the BAS by performing such machine learning, load prediction, and optimal schedule generation.

It is desirable to provide an interface based on a web XML service for interworking with the external BAS and external BAS, and through this, it is desirable to acquire data necessary for machine learning, prediction, and control value generation, and to provide feedback on the result. .

To this end, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention provides a web portal, a function that allows a user to perform a required function directly through a web page, and an XML-based Through messages, functions such as machine learning, prediction, and control value generation can be performed, and results can be received.

At this time, data necessary for command execution can also be transmitted (front end) together with the transmission of the REQUEST message. At this time, for security purposes, encrypted login information can be transmitted with each request.

The transmitted message is delivered to a web service that runs TEMS (Total Energy Management System) API, a core engine of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention that performs prediction and control functions. However, it is possible to create a result according to the execution result and give feedback to the BAS system through the web portal (Back End).

In order to efficiently manage the demand facility 40, the following management techniques can be used to predict and analyze the power generation of building loads and distributed resources, and this can be included in the machine learning analysis criteria for automatic building control. have.

As shown in FIG. 1, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention includes a power generation forecasting unit 100, a demand forecasting unit 200, a schedule generation unit 300, and It includes a power control unit 500.

The power generation predicting unit 100 predicts the power generation amount of the self-powered device 10 probabilistically based on machine learning.

The self-powered device 10 generates power using renewable energy such as solar heat, solar power, biomass, wind power, small power, geothermal heat, marine energy, and waste energy, and new energy such as fuel cell, liquefied coal gasification, and hydrogen energy. As a device, it refers to all power generation systems that produce electricity by the power generation method except for conventional thermal power generation, hydro power generation, and nuclear power generation.

When describing the self-powered device 10 using a solar cell as an example, the power generation predicting unit 100 predicts web-based weather without hardware and installation structure data such as module efficiency, rated output, and incident angle panel information of the solar panel. A hybrid prediction method that organically combines a correlation model and a neural network can be applied based on data and actual power generation data.

The demand forecasting unit 200 predicts the power demand of the demand facility 40 probabilistically based on machine learning.

The demand facility 40 includes household appliances, motors, pumps, fans, heat exchangers, boilers, etc., including power reception facilities and distribution facilities, and all types of electricity used by companies such as steelmaking, papermaking, thermal power generation, and nuclear power generation. Electric facilities or electric devices are called demand facilities.

The demand forecasting unit 200 may also apply a hybrid prediction method in which a correlation model and a neural network are organically combined based on the type of distribution facility in the premises and power consumption data, similar to the power generation forecasting unit 100.

That is, the generation amount prediction unit 100 and the demand amount prediction unit 200 predict probabilistically based on machine learning, and a hybrid prediction method in which a correlation model and a neural network are organically combined may be applied.

It goes without saying that the embodiments applicable to the demand forecasting unit 200 described below can also be applied to the power generation forecasting unit 100.

In general, electricity consumers can be classified into electricity used at home and electricity used by companies in terms of whether they only use electricity or supply electricity to the power system.In the present invention, simple power users and professional power users It will be strictly classified as follows.

A simple power user is a consumer who only receives and uses electricity from the power system, and refers to a person or institution that owns and operates only the demand facility 40 without the self-generation device 10 or the power storage device 30.

Professional power users receive and use electricity from the power system, supply electricity to the power system, and send it to their own demand facility (40) or power storage device (30) from their own self-generating device (10), or The electrical connection between the self-powered device 10, the power system 20, the power storage device 30, and the demand facility 40, which is also supplied from the power storage device 30 of the It refers to the person or organization that uses the function it controls.

In the present invention, a simple power user will be described using "household electricity customer" as an example, and a professional power user will be described by taking "enterprise power customer" as an example, but this is only an example for home/enterprise use, and actual simple power users and professional users The division of power users is not in homes and businesses/company, and is divided in terms of whether electricity users can send their own electricity to the power system.

In addition, the present invention relates to a technology in which a professional power user receives electricity from the power system 20 or supplies electricity to the power system 20 but efficiently and economically supplies/accepts it.

The demand characteristics of simple power users are the domestic environmental factors (hereinafter referred to as "household simple environmental factors"), household appliances environmental factors (hereinafter referred to as "household simple environmental factors") and household electricity bill system. It depends.

The demand characteristics of professional power users are the environmental factors (hereinafter referred to as "company-oriented professional environmental factors") and operating factors (hereinafter referred to as "facility professional operation factors") and the company's electricity. It depends on the fare system.

Corporate professional environmental factors include items related to employees, such as quality of manpower, environment in which manpower is faced, and treatment of manpower, accounting items such as sales, net income, and debt, and items related to sales such as research and development, product competitiveness, and company and product awareness. There are. Equipment-specific professional operation factors include equipment durability items such as prices of production and building management equipment, their aging, their precision, and deterioration due to their operation, and real-time operation characteristics items such as voltage, current, and resistance.

The demand forecasting unit 200 may probabilistically predict the power demand by preparing a machine learning analysis criterion for automatic building control.

The real-time facility characteristic state value of the demand facility 40 refers to a state value of a sensor that is not attached or attached to the demand facility 40 but is input in real time from the measurement and monitoring sensors that monitor the operation of the facility. do.

Like the real-time facility characteristic state value of the demand facility 40, the real-time facility characteristic state value of the power storage device 30 and the self-power generation device 10 is also the power storage device 30 and the self-power generation device 10 It refers to the state value of a sensor that is not attached to or attached to, but is input in real time from measurement and monitoring sensors that monitor the operation of the facility.

The schedule generation unit 300 is the amount of generation generated by the generation amount prediction unit 100, the amount of power demand predicted by the demand amount prediction unit 200, the existing peak load, power plan and state value information of the power storage device 30 Including to generate the power storage device 30 or the optimal schedule of the building load so that the building electricity bill can be minimized.

That is, when the power storage device 30 is not installed, an optimal schedule for the load of the building can be generated, and when the power storage device 30 is installed, an optimal schedule for the power storage device 30 can be generated, and power storage When the device 30 is installed, an optimal schedule for both the power storage device 30 and the building load can be generated.

The power storage device 30 is also called an ESS (Energy Storage System), and refers to a device that converts electrical energy into a form of chemical energy, stores it, and generates electricity when needed, meaning that it can be charged several times, " It also refers to the use of the name "rechargeable battery". Commonly used storage batteries include lead storage batteries, nickel-cadmium batteries (NiCd), nickel-metal hydride batteries (Ni-MH), lithium ion batteries (Li-ion), and lithium ion polymer batteries (Li-ion polymer). In the present invention, it is not limited to these batteries, but it means all things that can store or release electrical energy.

The building load refers to devices that consume energy in a building, such as a power load, a heat load, and a gas load.

Power load refers to a load using power as an energy source, heat load refers to a load using a heat source as an energy source, and gas load refers to a load using gas as an energy source.

Examples of the power load may be lighting, air conditioners, heaters, and the like, and the power load may be controlled by ON/OFF control and brightness control (dimming).

The power plan can be largely divided into a real-time power supply rate system and a real-time supply and demand rate system.

The real-time power supply and supply and demand rate system of the power system 20 is a general commercial electricity supplier, that is, an electricity rate system determined by a company such as KEPCO, that is, the rate system that customers pay for electricity use, referred to as the electricity supply and demand rate system. , Electricity charge system that sells electricity through facilities owned by customers, such as solar or wind power, to commercial electricity providers is called a power supply fee system.

If this rate system changes in real time, it becomes a real-time power supply and supply rate system.

The real-time power supply fee system is a system in which the electricity rate system supplies electricity to customers (the same meaning as consumers) in the electricity system, and the customer pays the electricity rate ?? In other words, KEPCO supplies electricity to homes, factories, and buildings, and KEPCO receives the price.

The real-time supply and demand rate system is a rate system in which customers receive electricity rates from KEPCO by reverse transmission of electricity from homes, buildings, and factories to the power system.

Here, the term “real-time” means that the electricity rate system is not a rate system within a specific period such as monthly, weekly, or daily, but varies by the time when electricity is supplied, so that the electricity rate is the time for receiving electricity (supplied from the power system to the customer, or It refers to the time of receiving electricity in the tariff system, which is determined differently according to the real-time reverse transmission.

The optimal schedule of the power storage device 30 may be performed through a getESSschedule function.

Performs the above prediction function internally to generate predicted load data, and optimal schedule based on meta information provided from external systems such as BAS (existing peak load value, current SOC of power plan and ESS, status value information such as capacity, etc.) Can be created.

The optimal schedule of the power storage device 30 considers both aspects of the building's peak load and the rate plan, and generates a schedule in which the building's electricity bill can be minimized based on the predicted power demand (load) and power generation information.

This can also be requested in XML format and the results can be fed back.

For this request, since prediction is performed internally, the model must be created in advance, otherwise an error code can be returned.

The power control unit 500 is the self-powered device 10 to control charging and discharging of the power storage device 30 according to the optimal schedule of the power storage device 30 generated by the schedule generation unit 300, It controls the electrical connection between the power system 20, the power storage device 30, and the demand facility 40.

The power system 20 refers to an electricity supply general system that supplies electricity by using an electric line that supplies electricity to electricity customers who use electrical equipment such as homes, buildings, and industries, and is usually a high-voltage transmission system and a low-voltage distribution system. consist of. In more detail, the electricity generated from the power plant passes through the transmission network and enters from the substation to the distribution network, and from the distribution network to homes, buildings, and industries.It is related to the supply of electric wires and electricity to the power grid, excluding electricity inside homes, buildings, and industries. The devices are collectively called the power system.

The power control unit 500 connects two or more devices selected from among the temporary power generation device 10, the power system 20, the power storage device 30, and the demand facility 40 so that electricity is communicated with each other or It refers to a facility that blocks the electric wire from passing through. The facility to connect or cut off uses a commercial circuit breaker, but a relay and a PLC (Programmable Logic Controller) or DDC (Direct Digital Controller) are connected to Control can be made possible.

The power control unit 500 is from the power system 20 to the demand facility 40, the power system 20 to the power storage device 30, the self-generation device 10 to the demand facility 40, self-powered From the device 10 to the power storage device 30, from the self-generation device 10 to the power system 20, from the power storage device 30 to the demand facility 40, or from the power storage device 30 The system 20 may be connected to move electricity or may perform a blocking function to prevent electricity.

The power control unit 500 includes a data processing system using real-time input values for demand characteristics of simple power users and demand characteristics of professional power users or values through machine learning, and the power system 20 Through a real-time analysis in which the power supply and supply fee system and the facility characteristic state values of the power storage device 30 and the self-powered device 10 are interlinked, an instruction to determine the connection target of the control facility for automatic control of the building I can.

Therefore, the power control unit 500 determines a control criterion for controlling the electrical connection command required for automatic building control in various directions, such as a power storage device, a power machine field, a power monitoring field, and a lighting control field, and commands accordingly. Function can be performed.

Interlocking the real-time facility characteristic status values of the power storage device 30 and the self-powered device 10 with the real-time power supply and supply and demand rate system of the power system 20 means that electricity bills and sales used by customers This means that the electric charge is linked with the state of the facility owned by the user to determine the driving mode.

The demand forecasting unit 200 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention applies a plurality of methodologies selected from a time series model, an automatic pattern classification technique, and an artificial neural network model to the accuracy of each time period. Accordingly, it may be characterized in that it is dynamically combined to periodically generate a final predicted value for each predetermined time.

The demand forecasting unit 200 may dynamically combine a time series model, an automatic pattern classification technique, and a neural network model according to time-specific accuracy to periodically generate a final predicted value for each predetermined time.

In particular, it is possible to perform automated machine learning based on building power load data and periodically update predictive models based on this.

As shown in Fig. 2, the demand prediction unit 200 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention has the most similarity to automatic load pattern classification through the K-mean technique. A methodology for extracting a high pattern from a preset input element and a methodology for learning an artificial neural network through the corresponding input element and predicting through it are dynamically aggregated according to prediction accuracy to generate a final prediction result.

At this time, error correction is also performed in the predicted value using the K-mean method, error correction is also performed in the predicted value using artificial neural network learning, and the result of comparing and analyzing the predicted value using the K-mean method and the predicted value using the artificial neural network learning. The final predicted value can be calculated by performing error correction.

This can improve coping ability for various power consumption variables through 3-step error correction.

The K-mean technique is also referred to as a K-Means algorithm, and is a technique used in machine learning (machine learning) and data mining, and is a representative unsupervised learning.

Unsupervised learning means that a certain outcome must not be predicted, and the computer finds a certain answer by itself.

For example, "I want to classify men and women" cannot be unsupervised learning because there is already a purpose and a value, but if a computer classifies data by itself and realizes the difference in characteristics between men and women, It becomes unsupervised learning.

K-Means selects a center value and performs classification using the distance between the center value and other data.

In the next execution, a more centrally located center value is selected, classified, and this process is repeated. When the classification is no longer possible, the task is terminated.

The K-mean method is an analysis method that uses the distance or similarity between given observations when there is no prior information about a population or category.The entire data is grouped into several groups and the characteristics of each group are identified, thereby improving the overall structure of the data. This is an analysis method that is intended to help you understand.

The artificial neural network is a total of computing systems implemented based on a neural network of a human or animal brain, and is one of the detailed methodologies of machine learning, and is in the form of a network in which several neurons, which are neurons, are connected. It is classified into several types according to its structure and function, and the most common artificial neural network is a multilayer perceptron with multiple hidden layers between one input layer and an output layer.

The K-mean technique, which is a clustering technique, classifies the pattern by day, so there is a limitation in giving changes within the same cluster. (Low flexibility)

Therefore, a technique that can correct the variance error by time is needed, and for this purpose, high flexibility based on artificial neural networks was performed. , It was confirmed that the performance according to the correction model was significantly improved.

As the TestBed, the Living-Lab environment in which the researchers live was applied, and the development and verification of correction functions for unusual matters occurring in real time in the TestBed were completed.

FIG. 2 is an example of a methodology for performing machine learning based on past data and predicting demand based on it.In this example, automatic load pattern classification through K-mean technique and the pattern with the highest similarity are inputted, such as a wake-up date. This is an example of a model that dynamically collects a methodology for extracting from an element and a method for learning an artificial neural network through the corresponding input element and predicting it according to the prediction accuracy. However, in the methodology, a regression model or other prediction methodology can be used in various ways, so the method is not limited to the methodology described above.

The demand forecasting unit 200 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention is represented by the following equation,

는 최종 결과,

는 클러스터링모델의 결과,

는 인공신경망모델의 결과, A는 클러스터링모델의 과거 데이터의 정확도 B는 인공신경망모델의 과거 데이터의 정확도)(here,

Is the final result,

Is the result of the clustering model,

Is the result of the artificial neural network model, A is the accuracy of the past data of the clustering model, B is the accuracy of the past data of the artificial neural network model)

Thus, the clustering model and the artificial neural network model can be characterized by assigning weights according to the accuracy of past data to derive the final result (see FIG. 2).

At this time, the accuracy of past data can be performed by setting a specific period.

For example, the final result can be derived by assigning weights according to the accuracy of the past week for each of the clustering model and the neural network model.

The weighted equation is a result of comparing and analyzing a predicted value using the K-mean technique and a predicted value using learning an artificial neural network, and applying error correction to the value.

The schedule generator 300 of the system for optimal management and control of building loads based on automatic machine learning according to an embodiment of the present invention includes the following equation,

Based on the power storage device 30 may be characterized in that it generates a charge/discharge schedule.

The optimization function for generating the charge/discharge schedule of the power storage device 30 is composed of “objective function” and “constrained condition”.

The objective function is a function representing the operating goal to be maximized or minimized.

For example, maximize-profit, odds, etc. | Minimization-can be cost, loss, power loss, error, etc.

Formulas for generating the charge/discharge schedule of the power storage device 30 are for generating a charge/discharge schedule for optimal operation of the power storage device 30.

), ②피크 부하치와 연동되는 전력기본요금(

), ③ESS 운전패에 따른 배터리 열화 비용(

), ④제어 가능 부하의 제어에 따른 사용자의 불편 비용, ⑤불필요한 충ㆍ방전을 억제(

)하고 ESS 충방전을 가능하면 평활화하기 위한 가상의 비용 항으로 구성된다.The objective function is expressed as the sum of several cost factors. Energy cost according to the amount of electricity used under the ① Time-of-Use (TOU), where the unit price of electricity varies by time.

), ② Basic electricity charge linked to peak load (

), ③ Battery deterioration cost due to ESS operation loss (

), ④ Controllable user's inconvenience cost according to the control of load, ⑤ suppress unnecessary charging and discharging (

), and it consists of a virtual cost term to smooth the ESS charge and discharge if possible.

)과 건물의 나머지 부하에 대한 예측치(

)의 합이 건물의 순 부하가 되므로 여기에 시간대별 전력요금 단가(

)를 곱한 것이 스케줄 구간(T) 내에서의 총 전력사용량에 따른 요금이 된다. Here, ① Time-by-time rate plan is a rate plan (real-time rate plan) in which the unit price of electricity by time slot is different. It is a rate plan that is applied in most commercial buildings in Korea.

) And the predicted value (

The sum of) becomes the net load of the building.

) Multiplied by the total power consumption within the schedule section (T).

)와 미래 스케줄 구간(T) 안에서 발생할 순 부하(Net load)의 최대치(

) 중 큰 값에 연동되어 다음 요금 갱신 기간 동안의 기본전력 요금을 나타낸다. ② Base rate according to peak load is the maximum peak value that has already occurred in the past during the base rate update period (usually 1 year).

) And the maximum value of the net load that will occur within the future schedule period (T) (

), which is linked to the larger value, and indicates the basic power rate for the next rate renewal period.

③ The deterioration cost according to the ESS operation pattern is a part for costing the depreciation of the battery according to the charge/discharge pattern of the ESS, especially the state-of-charge (SOC) determined by this optimization. It is expressed based on an equation modeling the battery deterioration cost according to the unit energy charge/discharge of.

As for the deterioration cost according to the ESS operation pattern, as described in the previously registered patent related to the deterioration density function (KR20160030643A), the depreciation triangulation according to the battery cycle can be cost.

④ Considering user inconvenience caused by on/off control of a controllable load as a cost, it reduces power bills during control and minimizes user inconvenience.

). 이를 통해 총 비용이 동일한 ESS 스케줄 후보 중에서 배터리에 가해지는 전류량을 작게하고 배터리의 감가상각을 최소화할 수 있는 스케줄이 최종적으로 선택되는 효과가 있다. 단, ①~③의 단위는 직접적 금액인 반면 ④, ⑤ 는 실제적 금액이 아닌 가상의 비용 항이므로 정의하기 위해 각 항들의 우선 순위를

과 같이 정규화된 가중치를 부여하여, ①~④ 이 무조건 우선시 되고, ①~④ 부분이 동일한 복수 스케줄이 존재하는 경우만 ⑤ 항에 따라 최종 스케줄이 결정되도록 하였다.Lastly, ⑤ unnecessary charge/discharge charge/discharge suppression clause is added by squared ESS charge/discharge amount for each time period in order to suppress unnecessary charge/discharge of ESS as much as possible when there are multiple ESS schedules with the same total cost of ①~③. It is a part of costing. For example, if the first hour is charged with 10 kWh and the remaining hour is charged with 20 kWh and the two hours are uniformly charged with 15 kWh, as a result, the total charging energy is the same as 30 kWh. Due to this ④ section, the latter schedule of charging two hours with 15kWh is selected (

). This has the effect of finally selecting a schedule capable of reducing the amount of current applied to the battery and minimizing the depreciation of the battery from among ESS schedule candidates with the same total cost. However, the units of ①~③ are direct amounts, whereas ④ and ⑤ are hypothetical cost terms rather than actual amounts.

As shown in Figure 5, normalized weights are given so that ①~④ are given priority unconditionally, and the final schedule is decided according to item ⑤ only when there are multiple schedules with the same part ①~④.

The constraint conditions are for setting conditions that must be satisfied by various solutions that can become operational targets. In the formula for generating the charge/discharge schedule of the power storage device 30, the constraints of the power storage device 30 are as follows: Can be set together.

The amount of remaining energy by time of the battery is desirable when the SOC is between 0% and 100%, and it can be set as Equation (2) because it is also necessary to calculate the SOC change amount according to time-based charging and discharging.

The physical usable range of the battery can be set as Equation (3) because the amount of power that can be discharged or charged is limited depending on the performance of the connected power converter.

Equation (4) represents the maximum charging (discharging) power limit of the battery,

Equation (5) is an equation for calculating the battery deterioration cost according to the SOC change,

A and b used in Equation (5) represent battery dependent parameters that can be estimated by experiment, and the following equation

(Here, the ACC function refers to a function that obtains the number of cycles that can be achieved at a specific DOD (Depth of Discharge) value, and DOD represents the depth of discharge, and the depth of discharge is the capacity of a fully charged (full charged) battery as 100%. It is a value (%) indicating how much discharge was discharged.

The degree of deterioration of the battery varies depending on how much charge and discharge is performed in which SOC section. Therefore, it is desirable to calculate the parameters a and b based on the big data of the battery characteristic test.

The power control unit 500 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention is predefined in the self-power generation device 10, the power storage device 30, and the demand facility 40 When a specific driving condition occurs, an alarm may be generated according to a driving characteristic alarm system that alarms the relevant matter.

The operation-specific state is a case in which the operation state is not desirable, and is input according to the operation of the device, and refers to the electric power generation device (referred to as the self-generation device 10, the power storage device 30, and the demand facility 40). It is classified into values that can be evaluated immediately, such as voltage and current, and values by analysis, such as the efficiency of interlocking devices based on them, and they are judged to be inferior by comparing them with the theoretical values of other demanding facilities or reference values through machine learning. Tell the situation.

The power control unit 500 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention acquires demand characteristics of power users, and collects the characteristic items that become reference values by labeling them. A learning model function is determined using a data set, and the self-powered device 10, the power system 20, the power storage device 30, and It may be characterized in that it controls the electrical connection between the demand equipment (40).

The machine learning analysis criterion for automatic building control is a technique that automatically acquires the demand characteristics of simple power users and the demand characteristics of professional power users and analyzes them by machine learning.In the present invention, the power data processing system is used. It can also be expressed as follows.

The machine learning analysis criterion for automatic building control is a learning model using a data set collected by label processing characteristic items that become reference values through machine learning for automatic building control (Machine Learning). Learning Model) function can be determined and processed based on the calculated value.

The real-time facility characteristic status value is a company-specific environmental factor that handles the demand characteristics of a professional power user at the management level of the company, and machinery, electricity, and control devices such as mechanical facilities required for the operation of buildings or factories or facilities required for production. It is classified and managed by the same facility professional operation factor, of which it refers to the real-time state value of the facility professional operation factor, that is, a value input in real time in the present invention.

Probabilistic building load power prediction that predicts the power generation amount of the self-powered device 10 probabilistically can be determined according to two functions: model build (eg, setDemandModel and setPVModel) and load prediction (ie, getDemandForecast and getPVForecast).

The model creation functions (setDemandModel and setPVModel) receive load history data as input to build a load prediction model, perform training, and create a model.

To this end, a historical data file consisting of data time metadata (e.g. year, month, day, hour, quarter) and weather information such as peak temperature, humidity and cloud cover, and historical energy data (e.g. PV power or building energy) You can make a request in XML format by configuring it in the given format.

The generated prediction model may be stored in the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention for future load prediction.

Results such as success or failure of this work can also be fed back through XML.

Generating a predictive model takes a significant amount of computation time, so it is usually desirable to perform it once a day.

After model creation, load and solar power generation can be predicted by functions such as getDemandForecast and getPVForecast.

If a prediction function is called without model generation, an error code can be returned.

When calling the prediction function, meta information such as the forecast request period and weather information for prediction, load and generation data for the previous few days must be transmitted together, and based on this, prediction and error correction for the period are performed and the final result is returned in XML format. can do.

As shown in Figure 3, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention is the power generation predicted by the power generation amount prediction unit 100, the demand amount prediction unit 200 An economical analysis unit that calculates an economic feasibility by calculating the predicted power load fluctuation value according to the predicted power demand amount and the optimal schedule of the power storage device 30 generated by the schedule generator 300 and applying a fee thereto ( 400),

The power storage device 30 and the demand facility 40 control schedule candidates are continuously updated through an optimization technique by repeatedly performing the optimal schedule generation and economic calculation of the power storage device 30, and the lowest charge is induced. It may be characterized by generating a control schedule to control the final power storage device 30 and the demand facility 40.

In the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention, control inputs of each power device (demand facility 40) and energy storage device (ESS) (power storage device 30) In determining, the operation information of each device is collected and the instantaneous load (lighting, monitor, etc.) that affects the instantaneous load only at the time of operation according to the operation characteristics of the load devices, and a task-type load with an operation sequence according to the state value (washing machine) , Dishwasher, etc.), thermal loads (air conditioners, heaters, etc.) that have a time-series dependence such as thermal energy, and the optimal values of operation and energy input and output are determined according to the characteristics of each operation, such as genetic algorithm and particle swarm optimization. An algorithm for controlling the electrical connection may be generated using a methodology or an optimization methodology such as linear programming or dynamic programming, and control may be performed through the method.

In applying the above method, the power cost of the building is calculated and optimal control of the power equipment is performed based on this. In this case, more than two aspects of the rate plan are usually considered at the same time.

In particular, it is possible to consider a time-based rate system in which the power unit price varies depending on time, and a peak rate system in which the basic rate is determined in conjunction with the highest peak value among the 15-minute power loads throughout the year.

To this end, it performs machine learning-based building load prediction and solar cell generation amount prediction, and calculates a change in the predicted power load according to the control schedule candidates of the generated power storage device 30 and demand facility 40, By applying, it is possible to calculate economic feasibility through the economic feasibility analysis unit 400.

Through an optimization technique that repeatedly performs this, the power storage device 30 and the demand facility 40 control schedule candidates are continuously updated, and a control schedule in which the lowest charge is induced is generated, and the final power storage device 30 and the demand facility The input value of (40) can be determined.

Such load prediction and schedule generation may be repeatedly performed periodically, and it is preferable to control the power storage device 30 and the demand facility 40 by using the most recent result.

As shown in Fig. 4, the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention includes a facility control unit 900 capable of controlling the demand facility 40, and the facility control unit 900 is characterized in that it comprises at least one selected from a machine field control unit, a power field control unit, a lighting field control unit, a SI (System integration) field control unit, and a BEMS (building energy management system) field control unit. can do.

The facility control unit 900 is based on the energy use characteristics of a building such as a machine field control unit, a power field control unit, a lighting field control unit, a SI (System integration) field control unit, and a BEMS (building energy management system) field control unit. It refers to performing a function of supplying or blocking electricity to a device that uses energy by collecting connection terminals according to the classified electricity usage characteristics into groups, and controlling them.

The machine field control unit is an assembly of terminals that control things such as AHU, EHP, temperature adjustment, and direct load control.

The power field control unit is an assembly of terminals such as switchgear, generator, demand control, and sequential control.

The lighting field control unit is an assembly of terminals such as general lights and dimming lights.

The SI (System Integration) field control unit is a collection of system integration related terminals.

The BEMS (building energy management system) field control unit is a collection of Building Energy Management System terminals.

The characteristics of these terminal assemblies can easily implement the present invention by referring to the contents in the technical literature related to commercial electric equipment or energy-using equipment.

The facility control unit 900 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention may be characterized in that it includes one of a sequential control method, an ON/OFF control method, or a combination of both. .

The sequential control method is a control that gradually changes the input and output signals of the sensor, such as temperature or illumination, and the ON/OFF control method is to change the number of lights to be lit by one or several. This is a method of determining and controlling the quantity of moving equipment.

Hereinafter, an embodiment of controlling the temperature for cooling and heating, air volume for cooling and heating, and illuminance of a lamp in a sequential control method, and for managing a load in an ON/OFF control method for a cooling/heater, air conditioner, or lamp will be described as follows.

When a random signal is received from the facility control unit 900 of the automatic machine learning-based building load optimal management and control system according to an embodiment of the present invention, the load installed in the energy use space in the automatic control program is then adjusted for cooling, heating and lighting load Energy load can be reduced by using the same method.

Adjustment of the heating and cooling load is performed by the 1st to 3rd methods described later, but after the 1st adjustment, the 2nd and 3rd methods can be applied sequentially or separately set.

1st plan-temperature reset control: a plan to manage the power load by resetting the temperature of the air conditioner or air conditioner

Second plan-Air volume output control control: A plan to manage power load by adjusting the output of motors such as air conditioners

3rd plan-ON/OFF control: A plan to manage power load by sequentially turning off air conditioners, air conditioners, and various devices in a set order

If there is an illuminance control function, the light load can be adjusted by adjusting the current illuminance, controlling the electric energy individually or by circuit, or by adjusting both.

The present invention is not limited to the above-described embodiments, and of course, various modifications can be made without departing from the gist of the present invention as claimed in the claims.

10: self-powered device 20: power system
30: power storage device 40: demand facility
100: power generation forecasting unit 200: demand forecasting unit
300: schedule generation unit 400: economic analysis unit
500: power control unit 900: facility control unit

Claims (10)

A power generation amount prediction unit 100 for probabilistically predicting the power generation amount of the self-powered device 10 based on machine learning;
A demand forecasting unit 200 for probabilistically predicting the power demand of the demand facility 40 based on machine learning;
Including the amount of electricity generated by the generation amount predicting unit 100, the amount of electric power demand predicted by the demand amount predicting unit 200, the existing peak load, the power rate plan, and the state value information of the power storage device 30 A power storage device 30 that can be minimized or a schedule generator 300 that generates an optimal schedule of a building load; And
The self-powered device 10, the power system 20, and the power storage device 30 to control charging and discharging of the power storage device 30 according to the optimal schedule of the power storage device 30 generated by the schedule generator 300 A power control unit 500 for controlling an electrical connection between the power storage device 30 and the demand facility 40;
Including,
The schedule generation unit 300
The following expression,

Based on the automatic machine learning-based building load optimal management and control system, characterized in that generating the charge / discharge schedule of the power storage device (30).
The method of claim 1,
The demand forecasting unit 200 is
Optimal management of building load based on automatic machine learning, characterized in that the final predicted value is periodically generated by a fixed time period by dynamically combining a plurality of methodologies selected from time series model, automatic pattern classification technique, and artificial neural network model according to timely accuracy. And control system.
The method of claim 2,
The demand forecasting unit 200 is
Automatic load pattern classification through K-mean method and methodology for extracting patterns with the highest similarity from preset input elements, and methodology for learning artificial neural networks through corresponding input elements and predicting through them are dynamically collected according to prediction accuracy Building load optimal management and control system based on automatic machine learning, characterized in that to generate a final prediction result.
The method of claim 2,
The demand forecasting unit 200 is
The following expression,

(here,

Is the final result,

Is the result of the clustering model,

Is the result of the artificial neural network model, A is the accuracy of the past data of the clustering model, B is the accuracy of the past data of the artificial neural network model)
By, the automatic machine learning-based building load optimal management and control system, characterized in that the weight according to the accuracy of the past data of each of the clustering model and the artificial neural network model is assigned to derive the final result.
delete The method of claim 1,
The power control unit 500
In the case of occurrence of a pre-defined operation specific state in the self-powered device 10, the power storage device 30, and the demand facility 40, an alarm is generated according to an operation characteristic alarm system that alerts the relevant matter. Building load optimal management and control system based on automatic machine learning.
The method of claim 1,
The power control unit 500
A learning model function is determined using a data set collected by acquiring the power user's demand characteristics, label processing characteristic items that become reference values, and calculating the calculated value using this. Based on the automatic machine learning-based building load optimal management and control system, characterized in that the electrical connection between the self-powered device 10, the power system 20, the power storage device 30, and the demand facility 40 is controlled. .
The method of claim 1,
The automatic machine learning-based building load optimal management and control system
Estimated according to the amount of generation predicted by the generation amount prediction unit 100, the amount of power demand predicted by the demand amount prediction unit 200, and the optimal schedule of the power storage device 30 generated by the schedule generation unit 300 An economic feasibility analysis unit 400 for calculating economic feasibility by calculating a fluctuation value of the generated power load and applying a charge thereto;
It further includes,
The power storage device 30 and the demand facility 40 control schedule candidates are continuously updated through an optimization technique by repeatedly performing the optimal schedule generation and economic calculation of the power storage device 30, and the lowest charge is induced. Automatic machine learning-based building load optimal management and control system, characterized in that by generating a control schedule to control the final power storage device (30) and demand equipment (40).
The method of claim 1,
The automatic machine learning-based building load optimal management and control system
It includes a facility control unit 900 capable of controlling the demand facility 40,
The facility control unit 900
Automatic machine learning, characterized in that it comprises at least one selected from a machine field control unit, a power field control unit, a lighting field control unit, a SI (System integration) field control unit, and a BEMS (building energy management system) field control unit. Based building load optimal management and control system.
The method of claim 9,
The facility control unit 900
Automatic machine learning-based building load optimal management and control system, characterized in that it comprises one of a sequential control method, ON / OFF control method, or both.

Images