KR20210097613A - Power consumption prediction system for optimization of demand management resources - Google Patents

Abstract

A power consumption prediction system for optimization of demand management resources for sequencing batch reactors (SBR) process is provided. According to an embodiment of the present invention, a power consumption prediction system includes: a collecting part for collecting power data for a sequencing batch reactors (SBR) process; and a processor for analyzing the characteristics of power consumption based on the collected power data and generating a machine learning-based power consumption prediction model. Accordingly, it is possible to respond to national peak demand management by predicting the electricity consumption of a wastewater treatment facility.

Description

Power consumption prediction system for optimization of demand management resources

The present invention relates to a power usage prediction system, and more particularly, to a power usage prediction system for optimizing demand management resources for a Sequencing Batch Reactors (SBR) process.

The operation of conventional wastewater treatment facilities aims to satisfy the legal standards of effluent quality rather than energy reduction.

However, as of the end of 2015, the amount of electricity used in the actual wastewater treatment facilities is 4,900GWh, which is 1% of the domestic electricity consumption. It is a large industry, and the power consumption of the industry is also on the rise due to the policy stance that the standards for discharged water quality are strengthened.

Therefore, it is required to find a way to respond to the national peak demand management by predicting the power consumption of the wastewater treatment facility, which is such an energy-consuming facility.

The present invention has been devised to solve the above problems, and an object of the present invention is to provide a power usage prediction system for optimizing demand management resources that can respond to national peak demand management through power usage prediction of a wastewater treatment facility. is in

According to an embodiment of the present invention for achieving the above object, a system for predicting power consumption includes a collecting unit for collecting power data for a Sequencing Batch Reactors (SBR) process; and a processor that analyzes the characteristics of power consumption based on the collected power data and generates a machine learning-based power usage prediction model.

In addition, the processor may include a data pre-processing module for classifying data characteristics and performing data pre-processing to remove outlier data.

And the data preprocessing module can perform directional statistics work that replaces time with a direction vector that follows von Mises distribution for machine learning learning of time-dependent and periodic power data.

In addition, the data preprocessing module may perform an outlier removal operation of removing outliers by using a unit time average or standard deviation of power data.

And the data preprocessing module obtains continuous power data or restores lost power data using at least one interpolation method of linear interpolation and spline interpolation to perform an interpolation operation. can

In addition, when the interpolation operation is performed, the data preprocessing module may perform a time synchronization operation of the power data using a re-sampling technique.

In addition, the processor may generate a power usage prediction model by removing a variable that reflects a past state in current data from a recurrent neural network (RNN).

On the other hand, according to another embodiment of the present invention, a method for predicting power consumption includes: collecting, by a power usage prediction system, power data for a Sequencing Batch Reactors (SBR) process; analyzing, by the power usage prediction system, characteristics of power usage based on the collected power data; and generating, by the power usage prediction system, a machine learning-based power usage prediction model using the analysis result.

As described above, according to the embodiments of the present invention, it is possible to cope with the national peak demand management by predicting the power consumption of the wastewater treatment facility.

1 is a view provided for the description of a power usage prediction system for optimizing demand management resources according to an embodiment of the present invention;
2 is a diagram illustrating an electric power usage pattern for the SBR processing process;
3 is a view provided for the description of the sewage outlier removal operation;
4 is a diagram illustrating the temperature data of the Meteorological Agency before the Interpolation operation;
5 is a diagram illustrating a result of performing an interpolation operation on the temperature data of the Meteorological Agency of FIG. 4 ;
6 is a diagram illustrating a power consumption prediction model of the SBR water treatment process;
7 is a diagram illustrating a model training data set;
8 is a diagram illustrating an example of a power usage pattern prediction result, and
9 is a flowchart provided to explain a method of predicting power consumption for optimizing demand management resources according to an embodiment of the present invention.

Hereinafter, the present invention will be described in more detail with reference to the drawings.

1 is a diagram provided to explain a power usage prediction system for optimizing demand management resources according to an embodiment of the present invention (hereinafter, collectively referred to as a 'power usage prediction system').

The power usage prediction system according to this embodiment is provided to respond to the national peak demand management through the power usage prediction of the wastewater treatment facility.

Specifically, in order to provide a demand management service of a wastewater treatment facility using the power usage prediction system according to the present embodiment, it is essential to predict the power usage during the peak time when the demand response is issued.

To this end, the present power usage prediction system may include a collection unit 110 , a storage unit 120 , and a processor 130 .

The collection unit 110 is provided to collect power data for the SBR process.

Specifically, the collection unit 110 may include a sensor for collecting power data or a communication means capable of receiving data from a separate sensor.

For example, the collection unit 110 may collect the power consumption of the pilot facility for designing a predictive model, and in this case, the pattern for the SBR water treatment process is as illustrated in FIG. 2 .

The power data illustrated in FIG. 2 utilizes an instantaneous power value, and the unit of data is watts and was stored once every 10 minutes.

At this time, the accumulated power value is not suitable as training data for the prediction model because the change in power consumption is not accurately reflected because the minimum unit of data is kilowatts. Therefore, it is preferable that the processor predicts the instantaneous power pattern through the prediction model and inversely calculates the accumulated power consumption per hour based on the prediction model.

The storage unit 120 is a storage medium for storing programs and data necessary for the operation of the processor 130 .

The processor 130 may analyze the characteristics of power consumption based on the collected power data and generate a machine learning-based power usage prediction model that can be used as an index for calculating the reduction amount.

In addition, the processor 130, after deriving the prediction result for the power consumption using the machine learning-based power usage prediction model, when the demand response reduction is issued, based on the power reduction prediction information according to the normal power usage and the response scenario It is possible to determine whether the demand response participation is possible.

Here, the machine learning-based power use prediction model may be implemented using a Python-based Keras library, or the like.

In general, sewage treatment facilities exist on a regional basis, and each treatment facility consists of a mixed environment with a plurality of treatment plants and processes. The prediction system constitutes a pilot facility for the SBR treatment process, which is mainly used in small sewage treatment facilities of less than 5 tons of existing processes, and collects and analyzes power data for the SBR process with this target, and analyzes the characteristics of the SBR treatment process. Implement a power use prediction model for

Meanwhile, the processor 130 performs data pre-processing to effectively classify data characteristics and remove outlier data because pre-processing of input data is very important to improve the performance of the power usage prediction model. It may include a data pre-processing module to perform.

Specifically, the data preprocessing module can perform directional statistics work that replaces time with a direction vector following the von Mises distribution for machine learning learning of time-dependent and periodic power data.

Here, the power data of the wastewater treatment facility used for machine learning has a characteristic that the power data is repeated based on the HRT (Hydraulic Retention Time).

In particular, the HRT of SBR is 6 hours and power data is repeated based on 24 hours. In order to learn time-dependent and periodic power data because it has a characteristic characteristic, time is replaced with a direction vector following the von Mises distribution.

In this case, the von Mises distribution is a continuous probability distribution on a circle showing an approximation close to the wrapped normal distribution, which is a circular analogue of the normal distribution in probability theory and directional statistics.

The data preprocessing module uses von Mises distribution to increase the amount of data in the process of substitution with a direction vector, but the final performance is improved and the characteristics of periodic data can be included.

In addition, the data preprocessing module may perform an outlier removal operation of removing outlier data using a unit time average or standard deviation of power data.

3 is a diagram illustrating sewage temperature data of a water treatment process. The sewage temperature data may show outliers different from the actual temperature as shown in the figure above depending on whether or not chemicals are supplied for microbial treatment. At this time, if the error outlier caused by the machine operation is removed, a normal temperature value can be obtained as shown in the figure below. The data preprocessing module may remove outliers as illustrated in FIG. 3 .

For example, when the data preprocessing module exceeds the normal value range determined by the maximum and minimum values of data corresponding to within 250 to 300% of the standard deviation of the power data, it may be determined as outlier data and removed.

Here, the outlier data includes thermal noise of electronic equipment among data acquired by various environments and paths and various sensors, an error in the human collection process, an error in the measurement process, an error in the network transmission process, and an error in the sampling process. It refers to data that has abnormal extreme values caused by various causes, such as data, or values that deviate from the normal distribution without validity.

4 is a diagram illustrating the temperature data of the Korea Meteorological Agency before the interpolation operation, and FIG. 5 is a diagram illustrating the result of performing the interpolation operation on the temperature data of the Meteorological Agency of FIG. 4 .

As illustrated in FIG. 4 , the power data collected through the collecting unit has a state value at a specific time as a discrete time value. These values may be equally spaced, but they may also be unequally spaced or have no data.

The data preprocessing module may perform an interpolation operation by using the interpolation technique to obtain continuous data or to restore lost data as illustrated in FIG. 5 by using these data.

That is, the data pre-processing module obtains continuous power data or restores lost power data using at least one interpolation method of a linear interpolation method and a spline interpolation method. will do

를 보간하기 위하여 인접 함수값

,

을 함수 전개하여, 이용하면 하기의 수식 1과 같이 표현된다.Linear interpolation is a method of linearly calculating according to a straight line distance to estimate a value located between the end points when the values are given.

Adjacent function values to interpolate

,

It is expressed as Equation 1 below when the function is expanded and used.

Formula 1)

Spline interpolation is a method that uses a cubic polynomial to estimate a value between a given point.

The data pre-processing module can obtain continuous power data by applying the spline technique to reduce the oscillation occurring at the end of the data.

4 to 5 show the temperature data of the Korea Meteorological Administration.

The temperature data illustrated in FIG. 4 is transmitted at unequal intervals during discrete time. Therefore, a step phenomenon occurs and a slight difference from the actual temperature pattern occurs.

The data pre-processing module may pre-process such data to converge discrete data into continuous data as illustrated in FIG. 5 when linear and cubic spline interpolation are applied by interpolation of these data.

In addition, when the interpolation operation is performed, the data pre-processing module may perform a time synchronization operation of power data using a re-sampling technique.

Since the collecting unit collects data through various sensors and acquisition paths, the collection time is different for each data, which requires time synchronization for each data.

Therefore, when the interpolation operation is performed for time synchronization of the collected data, the data preprocessing module may perform the time synchronization operation of the power data using the re-sampling technique.

On the other hand, the SBR treatment process is a batch-type water treatment process, not a continuous water treatment process, in which the sequential treatment process of inflow-stirring-aeration-precipitation-discharge is driven.

That is, since the power usage pattern shows a regular pattern according to the processing order, it can be determined as time series data. However, it was confirmed that the predictive model designed and judged with time series data had very low predictive performance. This means that the power usage pattern of the SBR process is not conditional probability data in which the current power usage is affected by the past power usage.

Accordingly, the processor 130 according to the present embodiment is implemented to generate a power use prediction model by removing a variable reflecting a past state in current data from a recurrent neural network (RNN) as illustrated in FIG. 6 .

Specifically, for model learning and prediction, a data set configuration is required.

In general, the SBR treatment process is a 6-hour HRT process and has a structure in which 4 cycles are repeated during one day.

In the model designed for this, it is desirable to configure the system so that the power consumption data of the past 24 hours 4 cycle process can be input and the power usage amount of the next 1 cycle process can be predicted.

Since power data is stored once every 10 minutes, a data set was created to have 144 24-hour input data and 36 correct answer data.

The data set consists of three types of training data, test data, and verification data, as illustrated in FIG. 7 . 7 is a diagram illustrating a part of a model training data set.

Specifically, the processor 130 trains a model designed with the generated data set, and each training parameter uses a batch size of 32 and a mean square error, and the optimization function may be set to Adam.

In addition, machine learning is performed for 800 epochs, and prediction results as shown in FIG. 8 can be derived by performing prediction on actual power consumption using a model that has completed machine learning.

9 is a flowchart provided to explain a method for estimating power usage for optimizing demand management resources (hereinafter, collectively referred to as a 'power usage prediction method') according to an embodiment of the present invention.

Referring to FIG. 9 , the power usage prediction method is executed by the power usage prediction system, a collection step of collecting power data for the SBR process (S910), and analyzing the characteristics of power usage based on the collected power data It may be composed of an analysis step (S920) and a generation step (S930) of generating a machine learning-based power use prediction model using the analysis result.

In the collection step S910 , the power usage prediction system may collect power data through various sensors and acquisition paths.

In the analysis step ( S920 ), the power usage prediction system may effectively classify data characteristics and perform data preprocessing to remove outlier data.

In the generating step ( S930 ), the power usage prediction system may generate a power usage prediction model by removing a variable that reflects a past state in current data from a recurrent neural network (RNN).

Through this, it is possible to predict the electricity consumption of the wastewater treatment facility and respond to the national peak demand management by using the prediction result.

On the other hand, it goes without saying that the technical idea of the present invention can also be applied to a computer-readable recording medium containing a computer program for performing the functions of the apparatus and method according to the present embodiment. In addition, the technical ideas according to various embodiments of the present invention may be implemented in the form of computer-readable codes recorded on a computer-readable recording medium. The computer-readable recording medium may be any data storage device readable by the computer and capable of storing data. For example, the computer-readable recording medium may be a ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, hard disk drive, or the like. In addition, the computer-readable code or program stored in the computer-readable recording medium may be transmitted through a network connected between computers.

In addition, although preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and the technical field to which the present invention belongs without departing from the gist of the present invention as claimed in the claims In addition, various modifications may be made by those of ordinary skill in the art, and these modifications should not be individually understood from the technical spirit or perspective of the present invention.

110: collection unit
120: storage
130: processor

Claims (8)

a collection unit that collects power data for the Sequencing Batch Reactors (SBR) process; and
A processor that analyzes the characteristics of power consumption based on the collected power data and generates a machine learning-based power usage prediction model;
The processor is
A data pre-processing module for classifying data characteristics and performing data pre-processing to remove outlier data;
Data preprocessing module,
Based on the periodicity in which the power data for the SBR process appears at 4 cycle intervals, substituting the power data with a direction vector following a specific distribution,
The power usage prediction model is,
Power usage prediction system, characterized in that it is a model for receiving power usage for 4 cycles of the SBR process and predicting power usage for the next 1 cycle.
The method according to claim 1,
Data preprocessing module,
For machine learning learning of time-dependent and periodic power data, a power usage prediction system characterized by performing Directional Statistics work that replaces time with a direction vector following the von Mises distribution.
3. The method according to claim 2,
Data preprocessing module,
Power usage prediction system, characterized in that performing an outlier removal operation of removing outlier data using a unit time average or standard deviation of power data.
4. The method according to claim 3,
Data preprocessing module,
Power consumption, characterized in that by using at least one interpolation method of linear interpolation and spline interpolation to obtain continuous power data or perform an interpolation operation to restore lost power data prediction system.
5. The method according to claim 4,
Data preprocessing module,
When the interpolation operation is performed, the power usage prediction system, characterized in that the time synchronization operation of the power data using a re-sampling technique.
The method according to claim 1,
The processor is
A power usage prediction system, characterized in that by removing a variable that reflects a past state in current data from a Recurrent Neural Network (RNN), a power usage prediction model is generated.
The method according to claim 1,
The processor is
After deriving the prediction result for the power consumption during the peak time when the demand response is issued by using the generated electricity usage prediction model, when the demand response reduction is issued, the demand response is based on the power consumption forecast in normal times and the reduction power forecast information according to the response scenario Power usage prediction system, characterized in that it is determined whether participation is possible.
Collecting, by the power usage prediction system, power data for a Sequencing Batch Reactors (SBR) process;
analyzing, by the power usage prediction system, characteristics of power usage based on the collected power data; and
Including, by the power usage prediction system, generating a machine learning-based power usage prediction model using the analysis result;
The analysis step is
Classify data characteristics, perform data preprocessing to remove outlier data,
Based on the periodicity in which the power data for the SBR process appears at 4 cycle intervals, substituting the power data with a direction vector following a specific distribution,
The power usage prediction model is,
Power usage prediction method, characterized in that it is a model for receiving power usage for 4 cycles of the SBR process and predicting power usage for the next 1 cycle.

Images