1. Introduction

Direct Load Control(DLC) program is an advanced agreement between a power utility and customers to control some customers’ appliances (e.g., air conditioners and water heaters). It is an incentive-based demand response system in which the utility provides the affected customer with financial incentives. After participation in the program, the utility can reschedule or turn on/off the appliances of customers using remote control switches

The existing DLC program has the following two kind of problems in view point of its functionality.

First, the utility turn on/off customers’ appliances without considering the customer's inconvenience. that causes a lot of customers are reluctant to participate in the DLC program, therefore, the DLC program participation rate is not increasing these days.

Second, it does not have a functionality to analyze customers’ appliances power consumption pattern according to the season of year, the day of the week, the special days of year(ex : New Year’s day, Thanks giving day). And it also does not have a functionality to analyze load consumption pattern according to the day weather(ex : temperature, humidity, illuminance). If the utility can select customers’ appliances to be turned on/off by considering power consuming pattern during specific time period according to the above, it can avoid customers’ inconvenience when they participate in the DLC program.

For example, if the utility turn off the customers’ appliances which is unnecessary to be turned on during a specific time(ex: light off during sunny day, air conditioner off during windy day) and it turns on appliances which is needed to be turned on in a specific time(ex : light on during cloudy day). Many customers are will to participate in DLC program.

In this paper, the author develops weather based intelligent demand controller to analyze correlation between weather condition and customers energy consumption and to predict customers energy consumption under the specific weather condition and special day in order to attract more customers are willing to participate in the DLC program. Presented demand controller includes two functionalities.

One is collecting weather data from weather sensors which is located in outside of building. Another is predicting load consuming pattern by using deep neural networks. The developed demand controller shows its effectiveness.

2. The Design of Intelligent Demand Controller

Short term load forecasting techniques have been presented to analyze the pattern of loads consumption until now. In this paper, Deep neural networks is applied to analyze and predict each customers’ appliances power consuming pattern.

2.2 The Overall Solution Process for Intelligent Demand Controller

In Fig. 1 shows overall architecture for the intelligent demand controller

그림 1 지능형 최대수요전력시스템 구성

Fig. 1 Overall architecture for demand controller

■ Step 1 : Outside standing weather sensors send a weather data( illuminance, temperature, humidity, wind speed) to the intelligent demand controller

■ Step 2 : Intelligent demand controller collects on-line weather data from Korea Meteorological Administration

■ Step 3 : By using deep neural networks, intelligent demand controller analyzes the correlation between the customer energy consumption pattern and weather condition. and it predicts the energy consumption in one hour by using weather data from Korea Meteorological Administration.

■ Step 4 : Customers set the target power value of demand controller to the predicted energy consumption value obtained from deep neural networks outcome. That can reduces customers’ inconvenience during demand controller is executed to be turn on/off the customers load.

■ Step 5 : When intelligent demand controller gets peak control signals from the utility, it turns on/off customers’ load according to the priority which is defined by deep neural networks.

Owing to including power forecasting and load priority functions according to the weather and customer’s preference, Proposed intelligent demand controller can reduces customers inconvenience when they participate in DLC program.

3. The Development of Intelligent Demand Controller

3.1 The Development of hardware

Proposed intelligent demand controller collects weather data using two ways. the one is collecting on-line weather data from Korea Meteorological Administration. Another is collecting data from stand-alone weather sensors located in the outside of customer building.

In Fig. 2 shows the prototype components of Intelligent demand controller

그림 2 지능형 최대수요전력시스템 구성요소

Fig. 2 The components of demand controller

In Fig. 2, Stand-alone weather sensors sends weather data to data acquisition device via RF signal and data acquisition device data are transmitted to intelligent demand controller via internet.

3.2 Man-Machine Interface (MMI)

Presented MMI is developed by Window 6.0 CE based C# software and 8 inch touch graphic LCD is used to display variable information about demand controller

3.2.1 Main display

In Fig. 3 shows Target power, estimated power, base power, current power. and the expected power is displayed graphically through an deep neural networks prediction according to the current temperature.

그림 3 MMI 초기 화면

Fig. 3 MMI for main display

3.2.2 Network settings

그림 4 네트웍 셋팅

Fig. 4 Display for network settings

Network settings include not only communication setting between demand controller and stand-alone weather sensors, but also reporting and various event during demand controller operation.

3.2.3 Memory settings

Memory settings are not only directly related to the speed of the system, but also allow real-time voltage and current values ​​to be stored for a specific period of time. In Fig. 5 shows a display for memory settings

그림 5 메모리 셋팅

Fig. 5 Display for memory settings

3.2.4 Target power value settings

그림 6 목표전력 셋팅

Fig. 6 Target power value settings

In Fig. 6, the target power value is determined after deep learning for customers energy consumption prediction is finished. During deep neural netwoks learning, seasonal and monthly energy consumption and external weather data are used.

그림 7 일반 셋팅

Fig. 7 General setting

When target power value is determined, customers set the priority of loads which is turned On/Off for the purpose of meeting the target power value for minimizing the bill.

3.2.5 General settings

In Fig. 7 shows controller software and memory upgrade for the controller general function.

3.2.6 Daily, monthly, Yearly Report

그림 8 일보, 월보, 년보 화면

Fig. 8 Display for daily, monthly, yearly report

In Fig. 8 shows customer energy consumption for year, month and day.

3.2.7 Deep neural networks learning for predicting customer energy consumption

By using deep neural networks learning algorithm, the correlation between customer energy consumption and weather condition is analyzed, Customers energy consumption according to temperature, humidity, iIlluminance from outside weather sensors are used to input data of neural network algorithm. And on-line weather data

그림 9 딥러닝을 이용한 부하예측 프로세스

Fig. 9 Energy consumption prediction process

from Korea Meteorological Administration are also used to input data. Therefore, intelligent demand controller can predict customers energy consumption in 15 minutes. If predicted power value exceeds target power value, demand controller reschedules the priority of customer’s loads.

In Fig. 9 shows a prediction process for customers energy consumption

그림 10 딥러닝 초기 화면

Fig. 10 Initial MMI for deep neural networks learning

그림 11 딥러닝 처리 과정

Fig. 11 Deep learning processing

As shown in Fig. 11, numbers of neural network learning are 5000 times, and input layer 3, hidden layer 1 is set to 24, and hidden layer 2 is set to 8. All data used in the above program is based on 15-minute power demand and weather data stored in a real time, and the weights of each neuron in the neural network are stored and used for power prediction. If the predicted power through the above process is greater than the target power values, priority based maximum demand power control is performed, and if it is smaller than the target values, regular power data adjustment control is performed.

In korea, DLC program uses 15 minutes accumulated power consumption. Therefore, in this paper, deep neural networks learning uses 15 minutes demanded temperature, humidity, iIlluminance and the special day power consumption value.

4. Conclusion

Direct Load Control program is an incentive-based demand response system in which the utilities provide the affected customer with a financial incentive in return for turn on/off the appliances of customers using remote control switches regardless of customers preference. Even though utilities should meet the customers satisfaction of energy consumption, the exisiting DLC program can not meet the this kind of satisfaction because they always restricts customers right for turn on/off loads whenever they want. In this paper, the author develops weather based intelligent demand controller to reduce the inconvenience of customer side in order to increase DLC program participant. the developed intelligent demand controller includes software for predicting energy consuming pattern according to weather condition by using deep neural networks and it reduces the inconvenience of customer participant by rescheduling the priority which is prioritized by customers. Prototype intelligent DLC controller is presented to validate this paper and it shows its effectiveness.