KR102435865B1 - Reservoir management and predicting model using artificial intelligence attention layer - Google Patents

Abstract

The present invention relates to a reservoir water quantity change predicting system using an artificial intelligence attention method. The reservoir water quantity change predicting system comprises: a first data pre-processing module for receiving reservoir management data for a first reservoir and performing a pre-processing operation, a second data pre-processing module for receiving external data and performing a pre-processing operation; an input data generating module for receiving the output of the first data pre-processing module and the second data pre-processing module and generating first input data and second input data; and a time series predicting module including an encoder provided with the first input data, a first decoder provided with output from the encoder and the second input data, and a second decoder provided with output from the encoder and the second input data, wherein the second input data is composed of part of the first input data. Therefore, provided is a sales predicting system using time series analysis, wherein an accurate result can be derived using precisely configured input data.

Description

Reservoir management and predicting model using artificial intelligence attention layer

The present invention relates to a reservoir water management and prediction model using an artificial intelligence attention technique. Specifically, the present invention relates to a system and method for predicting the amount of change in low water quantity with high accuracy through the configuration of various input data.

Reservoirs are facilities to prevent natural disasters such as floods and droughts, and they not only prevent various natural disasters, but also have a profound impact on daily life by supplying agricultural and living water.

However, the reservoir is simply operated and managed based on human experience, and the decision on whether to open the floodgate is made only by subjective judgment without specific and objective criteria.

If the decision is wrong, the damage caused by it is very large, and in reality, it is managed only by skilled personnel with many experiences and know-how. However, if it is possible to accurately predict the amount of change in the future storage volume of the reservoir, it will be possible to operate the reservoir without the burden of manpower. In this case, since reduction of economic cost and time cost is expected, the development of a prediction system and method with high accuracy is required.

Registered Patent Publication No. 10-2128992

An object of the present invention is to provide a sales forecasting system using a time series analysis that accurately composes input data through correlation analysis on management data of a reservoir and various external data to derive accurate results.

In addition, another object of the present invention is to provide a sales forecasting method using a time-series analysis that accurately composes input data through correlation analysis on management data of a reservoir and various external data to derive accurate results.

The objects of the present invention are not limited to the above-mentioned objects, and other objects and advantages of the present invention not mentioned may be understood by the following description, and will be more clearly understood by the examples of the present invention. It will also be readily apparent that the objects and advantages of the present invention may be realized by the means and combinations thereof indicated in the appended claims.

The system for predicting the amount of change in the amount of storage according to some embodiments of the present invention for solving the above problems is a first data pre-processing module that receives the reservoir management data for the first reservoir and performs a pre-processing operation, and performs a pre-processing operation by receiving external data a second data preprocessing module, an input data generating module which receives the outputs of the first data preprocessing module and the second data preprocessing module to generate first input data and second input data, and an encoder to which the first input data is provided; , a time series prediction module comprising a first decoder provided with an output from the encoder and second input data, and a second decoder provided with an output from the encoder and second input data, wherein the second input data is a first input It consists of part of the data.

In some embodiments, the reservoir management data includes a first water quantity data of the first reservoir at a first time point, a first flow rate data of the first reservoir, and a first relating to a time point at which the first water quantity data and the first flow rate data are generated. Second data relating to the data generation time and the second water volume data of the first reservoir at a second time point different from the first time point, the second flow rate data of the first reservoir, and the second water volume data and the second flow rate data are generated The creation time may be included.

In some embodiments, the first data pre-processing module generates first storage water amount change data for the first reservoir, first accumulated flow rate data, and first sluice gate opening data by using the first water storage amount data and the first flow rate data and preprocessing the first stored water amount data, the first flow rate data, the first data generation time, the first stored water amount change data, the first accumulated flow rate data, and the first sluice gate opening data to generate the first preprocessing data.

In some embodiments, the external data includes first weather information data of a first area in which the first reservoir is located at a first time point, and a second time point at a second time point following the first time point and performing the prediction. The first weather information data of the first region and the first weather forecast data for a third time point after the second time point predicted at the second time point may be included.

In some embodiments, the first weather information includes temperature data for the first region, humidity data for the first region, sunlight data for the first region, and precipitation amount for the first region at the first time point and the second time point. data, wherein the first weather forecast data includes the highest temperature prediction data for the first area at the third time point predicted at the second time point, and the lowest temperature for the first area at the third time point predicted at the second time point It may include prediction data, average temperature prediction data for the first area at the third time point predicted at the second time point, and precipitation probability prediction data for the first area at the third time point predicted at the second time point.

In some embodiments, the first input data includes data at a first time point and data at a second time point that follows the first time point and is a time point at which prediction is performed, and the second input data includes data at the second time point data may be included.

In some embodiments, the first decoder uses the output from the encoder and the second input data to determine prediction data for the reservoir runoff of the first reservoir, and the second decoder receives the output from the encoder and the second input data. By using it, it is possible to determine the prediction data for the amount of stored water inflow of the first reservoir.

In some embodiments, the encoder further comprises an attention layer, the first decoder further uses the attention layer when determining the prediction data for the reservoir outflow of the first reservoir, and the second decoder is configured to: When determining prediction data for , an attention layer may be further used.

The method for predicting the amount of change in the amount of storage according to some embodiments of the present invention for solving the above problems includes the steps of receiving reservoir management data and external data, pre-processing the reservoir management data and external data, pre-processed reservoir management data and pre-processed external generating first input data for a first time point, a second time point following the first time point, and second input data for a second time point using the data, and providing the first input data to an encoder; , providing an output of the encoder and second input data to a first decoder; providing an output of the encoder and second input data to a second decoder; and generating the stored water inflow prediction data using the second decoder, and generating the stored water outflow prediction data by using the output of the encoder and the second input data.

In some embodiments, generating the stored water inflow prediction data may include further using an attention layer included in the encoder, and generating the stored water outflow prediction data may include further using an attention layer.

1 is a conceptual diagram illustrating a system for predicting the amount of change in the amount of water stored using time series analysis according to some embodiments of the present invention.
2 is a block diagram for explaining in detail a storage change amount prediction server according to some embodiments of the present invention.
3 is a view for explaining a reservoir management data pre-processing module according to some embodiments of the present invention.
4 is a view for explaining an external data preprocessing module according to some embodiments of the present invention.
5 is a diagram for explaining a time series prediction module according to some embodiments of the present invention.
6 is a diagram for explaining an example of a method of generating a storage amount change amount prediction data using a time series prediction module according to some embodiments of the present invention.
7 is a diagram for explaining a training data set and a test data set according to some embodiments of the present invention.
8 is a graph of experimental results for explaining a difference between predicted data on the amount of change in the amount of stored water and the actual amount of change in the amount of stored water according to some embodiments of the present invention.
9 is a view for explaining a method of predicting a change in the amount of stored water according to some embodiments of the present invention.
FIG. 10 is a diagram for explaining an example of a method of generating a storage amount change amount prediction data using a time series prediction module according to some other embodiments of the present invention.
11 is a view for explaining a method for predicting a change in the amount of stored water according to another exemplary embodiment of the present invention.

Terms or words used in this specification and claims should not be construed as being limited to a general or dictionary meaning. In accordance with the principle that the inventor can define a term or concept of a word to best describe his/her invention, it should be interpreted as meaning and concept consistent with the technical idea of the present invention. In addition, the embodiments described in the present specification and the configurations shown in the drawings are only one embodiment in which the present invention is realized, and do not represent all the technical spirit of the present invention, so they can be replaced at the time of the present application. It should be understood that there may be various equivalents and modifications and applicable examples.

Terms such as first, second, A, B, etc. used in this specification and claims may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. The term 'and/or' includes a combination of a plurality of related listed items or any of a plurality of related listed items.

Terms used in this specification and claims are used only to describe specific embodiments, and are not intended to limit the present invention. The singular expression includes the plural expression unless the context clearly dictates otherwise. It should be understood that terms such as “comprise” or “have” in the present application do not preclude the possibility of addition or existence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification in advance. .

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

In addition, each configuration, process, process or method included in each embodiment of the present invention may be shared within a range that does not technically contradict each other.

Hereinafter, with reference to FIGS. 1 to 8 , a system for predicting the amount of change in water storage using time series analysis according to some embodiments of the present invention will be described.

1 is a conceptual diagram illustrating a system for predicting the amount of change in the amount of water stored using time series analysis according to some embodiments of the present invention.

Referring to FIG. 1, the storage water quantity change prediction system using time series analysis according to some embodiments of the present invention includes a storage water quantity change prediction server 200, and the storage water quantity change prediction server 200 is a reservoir 100 and an external institution ( 300) and data can be exchanged.

The reservoir 100 refers to an artificial repair facility that stores flowing water and uses it when water is needed. The reservoir 100 may supply/storage water by adjusting the sluice gate in preparation for natural disasters such as floods and droughts. In addition, the reservoir 100 may supply water by opening the sluice gate in a situation in which water is insufficient, such as agricultural water or living water.

The reservoir 100 may provide the reservoir management data D_Mng to the storage amount change prediction server 200 . In this case, the reservoir management data D_Mng may be time series data, that is, data listed in time order. In other words, the reservoir management data D_Mng may include data regarding a first time point and data regarding a second time point that follows the first time point.

Hereinafter, for convenience of explanation, the first time point is a past time point, the second time point is a reference point using the prediction system (that is, the prediction time point), and the third time point follows the second time point, and the prediction is performed. defined as a future point of view. However, the first viewpoint and the third viewpoint do not mean a single viewpoint and may include a plurality of viewpoints. In other words, the storage water quantity change prediction system according to some embodiments of the present invention can be viewed as a system for predicting data of a (plural) third time points by using the data of the (plural) first and second time points. .

The reservoir management data D_Mng may include water storage data and flow rate data of the reservoir 100 . The water volume data means data related to the amount of water stored in the reservoir 100, and the flow data means data related to the amount of water flowing in/out of the reservoir 100/to the reservoir 100.

The reservoir management data D_Mng may include water storage amount data and flow rate data for the first time point and the second time point, and information about the first time point and the second time point. However, embodiments are not limited thereto, and the reservoir management data D_Mng may further include at least one of storage amount change data for the first time point and the second time point, accumulated flow rate data, and data on whether the sluice gate is opened.

The reservoir 100 may provide the reservoir management data D_Mng to the storage amount change prediction server 200 periodically or aperiodically. The storage amount change prediction server 200 may receive the reservoir management data D_Mng periodically or aperiodically, and may predict the change amount of the storage water amount of the reservoir 100 . In addition, the storage amount change prediction server 200 may receive the reservoir management data D_Mng periodically or aperiodically, and adjust the internal parameter value by retraining the time series prediction module.

The external organization 300 may be an organization that generates data on various factors related to the amount of water stored in the reservoir 100 . In FIG. 1 , the external organ 300 is illustrated as a single organ, but the present embodiment is not limited thereto. That is, the external organ 300 may be one or two or more organs.

The external agency 300 may generate the external data D_Ext and transmit it to the storage amount change prediction server 200 . The external data D_Ext may include various types of data according to the type of the external organ 300 . The external data D_Ext may refer to various data that may affect the amount of water stored in the reservoir 100 . In some embodiments, the external agency 300 may be a meteorological agency, but embodiments are not limited thereto.

The external agency 300 may provide the external data D_Ext to the storage amount change prediction server 200 . In this case, the external data D_Ext may be time series data, that is, data arranged in time order. In other words, the external data D_Ext may include data regarding a first time point and data regarding a second time point that follows the first time point.

The external data D_Ext includes weather information data of the first time point and the second time point for the first area where the reservoir 100 is located, and the weather forecast data of the first area at the third time point predicted at the second time point. can do. For example, the weather information data may include temperature data of the first area, humidity data of the first area, sunlight data of the first area, and precipitation data of the first area. Also, for example, the weather forecast data includes the lowest temperature forecast data for the first region, the highest temperature forecast data for the first region, average temperature forecast data for the first region, and precipitation probability prediction data for the first region. may include However, embodiments are not limited thereto. For example, the weather information data may further include accumulated precipitation data of the first area.

As described above, the external data D_Ext includes the weather information data of the first time point and the second time point for the first area where the reservoir 100 is located, and the first area of the third time point predicted at the second time point. It may include weather forecast data. In other words, the external data D_Ext includes, for the first region, the temperature data at the first time point, the humidity data at the first time point, the amount of sunlight data at the first time point and the precipitation amount data at the first time point, the temperature data at the second time point, Humidity data at the second time point, sunlight data at the second time point, and precipitation data at the second time point, and at a third time point predicted at the second time point (eg, 1 day and/or 2 days after the second time point) The lowest temperature forecast data for the third time point predicted at the second time point, the highest temperature forecast data for the third time point predicted at the second time point, the average temperature forecast data for the third time point predicted at the second time point, and precipitation for the third time point predicted at the second time point Probability prediction data may be included.

The storage amount change prediction server 200 may receive the reservoir management data D_Mng and the external data D_Ext, and may generate the storage water amount change amount prediction data D_Es. The storage amount change prediction server 200 may transmit the storage water amount change amount prediction data D_Es to the reservoir 100 to provide assistance for the management of the reservoir 100 .

The storage amount change prediction server 200 includes a workstation, a data center, an internet data center (IDC), a direct attached storage (DAS) system, a storage area network (SAN) system, and a network attached storage (NAS) system. ) system and a RAID (redundant array of inexpensive disks, or redundant array of independent disks) system may be implemented as at least one, but the present embodiment is not limited thereto.

The storage amount change prediction server 200 may transmit data to the reservoir 100 through a network. The network may include a network based on a wired Internet technology, a wireless Internet technology, and a short-range communication technology. Wired Internet technology may include, for example, at least one of a local area network (LAN) and a wide area network (WAN).

Wireless Internet technology is, for example, wireless LAN (WLAN), DLNA (Digital Living Network Alliance), WiBro (Wireless Broadband: Wibro), Wimax (World Interoperability for Microwave Access: Wimax), HSDPA (High Speed Downlink Packet) Access), High Speed Uplink Packet Access (HSUPA), IEEE 802.16, Long Term Evolution (LTE), Long Term Evolution-Advanced (LTE-A), Wireless Mobile Broadband Service (WMBS) and 5G New Radio (NR) technology. However, the present embodiment is not limited thereto.

Short-range communication technologies include, for example, Bluetooth, Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra-Wideband (UWB), ZigBee, and Near Field Communication: At least one of NFC), Ultra Sound Communication (USC), Visible Light Communication (VLC), Wi-Fi, Wi-Fi Direct, and 5G NR (New Radio) may include. However, the present embodiment is not limited thereto.

The low water quantity change prediction server 200 communicating through the network may comply with technical standards and standard communication methods for mobile communication. For example, standard communication methods include Global System for Mobile communication (GSM), Code Division Multi Access (CDMA), Code Division Multi Access 2000 (CDMA2000), Enhanced Voice-Data Optimized or Enhanced Voice-Data Only (EV-DO). , at least one of Wideband CDMA (WCDMA), High Speed Downlink Packet Access (HSDPA), High Speed Uplink Packet Access (HSUPA), Long Term Evolution (LTE), Long Term Evolution-Advanced (LTEA), and 5G New Radio (NR). may include. However, the present embodiment is not limited thereto.

2 is a block diagram for explaining in detail a storage change amount prediction server according to some embodiments of the present invention.

Referring to FIG. 2 , the storage amount change prediction server 200 according to some embodiments of the present invention includes a reservoir management data preprocessing module 210 , an external data preprocessing module 220 , an input data generating module 230 , a time series prediction module 240 and a test module 250 .

The reservoir management data preprocessing module 210 may receive the reservoir management data D_Mng from the reservoir 100 . The reservoir management data pre-processing module 210 may perform a pre-processing operation on the reservoir management data D_Mng to generate the pre-processed reservoir management data D_pMng. For a more detailed description of the reservoir management data pre-processing module 210, see FIG. 3 further.

3 is a view for explaining a reservoir management data pre-processing module according to some embodiments of the present invention.

Referring to FIG. 3 , the reservoir management data pre-processing module 210 may receive the reservoir management data D_Mng and pre-process it to generate the pre-processed reservoir management data D_pMng.

Reservoir management data pre-processing module 210 includes a first data generation time pre-processing module 211, a storage amount data pre-processing module 212, a flow data pre-processing module 213, a storage amount change pre-processing module 214, an accumulated flow pre-processing module 215 ) and a sluice gate opening estimation module 216 .

As described above, the reservoir management data D_Mng may include storage water amount data, flow rate data for the reservoir 100 , and a first data generation time for a time point at which the stored water amount data and flow rate data are generated.

The first data generation time pre-processing module 211 may pre-process the first data generation time included in the reservoir management data D_Mng to convert it into a form of data that can be input to the time series prediction module 240 . The first data generation time refers to a time point at which water storage data and flow data included in the reservoir management data D_Mng are generated. According to some embodiments, the first data generation time preprocessing module 211 may preprocess the first data generation time to convert the first data generation time into a 1 X n-dimensional matrix form.

For example, the first data generation time may include month information indicating in which month the water storage data and flow data were generated, and week information indicating in what week of the month the storage water amount data and flow data were generated. In this case, the month information may be divided into 12 categories, and the main information may be divided into 5 categories. The first data generation time pre-processing module 211 may perform one-hot encoding on the month information and the week information included at the first data generation time point, and may represent it as a 1 X 17 matrix. For example, when the first data generation time means the second week of February, the first data generation time preprocessing module 211 sets the first data generation time to [0,1,0,0,0,0,0] ,0,0,0,0,0,0,1,0,0,0]. However, embodiments are not limited thereto, and the first data generation time pre-processing module 211 may convert the shape of the first data generation time point in various ways.

The water quantity data pre-processing module 212 may pre-process the water quantity data included in the reservoir management data D_Mng and convert it into a form of data that can be input to the time series prediction module 240 . For example, the low water quantity data preprocessing module 212 may normalize or scale the low water quantity data to convert it into a form of data that can be input to the time series prediction module 240 .

The flow data pre-processing module 213 may pre-process the flow data included in the reservoir management data D_Mng, and may convert the flow data into a form of data that can be input to the time series prediction module 240 . For example, the flow data pre-processing module 213 may normalize or scale the flow data to convert it into a form of data that can be input to the time series prediction module 240 .

The amount of water change pre-processing module 214 may determine the amount of change data of the stored water by using the water flow data and the water storage data included in the reservoir management data D_Mng. However, embodiments are not limited thereto, and data on the amount of change in the amount of storage may be included in the reservoir management data D_Mng. The stored water quantity change pre-processing module 214 may pre-process the determined stored water quantity change quantity and convert it into a form of data that can be input to the time series prediction module 240 .

The accumulated flow rate preprocessing module 215 may determine the accumulated accumulated flow rate from initialization of the accumulated flow rate to a specific time point by using the flow rate data included in the reservoir management data D_Mng. For example, the accumulated flow rate pre-processing module 215 may accumulate all the flow rates from the beginning of the year to the first time point to determine the accumulated flow rate at the first time point. Similarly, the accumulated flow rate pre-processing module 215 may accumulate all the flow rates from the beginning of the year to the second time point to determine the accumulated flow rate at the second time point. Naturally, since the second time point follows the first time point, the accumulated flow rate at the second time point may be greater than or equal to the accumulated flow rate at the first time point. The accumulated flow rate pre-processing module 215 may pre-process the determined accumulated flow rate and convert it into a form of data that can be input to the time series prediction module 240 .

According to some embodiments, the accumulated flow rate may be used as very important data for predicting the change amount of the stored water. Reservoir 100 is typically determined by a contract for the total annual discharge amount, and establishes a discharge plan accordingly. Therefore, the cumulative flow, which means the total amount of discharge so far, can act as an important factor in making a decision on the discharge plan. Therefore, unlike the prior art, the stored water quantity change prediction system according to some embodiments of the present invention uses the accumulated flow data generated by the accumulated flow pre-processing module 215 to predict the stored water quantity change with higher accuracy. do.

The sluice gate opening estimating module 216 may estimate whether the sluice gate is opened by using the water storage amount data and the flow rate data included in the reservoir management data D_Mng, and generate sluice gate opening data. For example, the sluice gate opening estimating module 216 may estimate that the sluice gate is opened when the flow rate is greater than or equal to a specific standard by referring to the flow rate data. The sluice gate opening estimation module 216 may express whether the sluice gate is open as 0 or 1, but embodiments are not limited thereto.

In summary, the reservoir management data pre-processing module 210 according to some embodiments of the present invention pre-processes the reservoir management data (D_Mng) received from the reservoir 100, and the data that can be input to the time series prediction module 240 is can be converted into a form. In addition, the reservoir management data preprocessing module 210 may process the reservoir management data D_Mng to generate additional data. That is, the reservoir management data pre-processing module 210 may process and convert the reservoir management data D_Mng to generate the pre-processed reservoir management data D_pMng. The pre-processed reservoir management data D_pMng may include pre-processed time point data, pre-processed water storage data, pre-processed flow data, pre-processed storage change amount data, pre-processed accumulated flow data, and sluice gate opening data.

Although it has been described in this specification that the stored water amount change data, the accumulated flow rate data, and the sluice gate openness data are determined by the stored water amount change amount pre-processing module 214, the accumulated flow rate pre-processing module 215 and the sluice gate opening estimation module 216, respectively, the embodiment are not limited thereto. For example, at least one of the storage amount change data, the accumulated flow rate data, and the sluice gate open data may already be included in the reservoir management data D_Mng. For example, when the reservoir management data (D_Mng) includes data on the amount of change in the amount of water stored in the reservoir management data (D_Mng), the pre-processing module 214 for the amount of change in the amount of water stored in the reservoir management data (D_Mng) By pre-processing, it can be converted into the form of data that can be input to the time series prediction module 240 . Hereinafter, for convenience of description, data or pre-processed data are used interchangeably, but those of ordinary skill in the art will be able to easily understand what is intended to be described in this specification.

Referring back to FIG. 2 , the external data preprocessing module 220 may receive external data D_Ext from the external organization 300 . The external data preprocessing module 220 may perform a preprocessing operation on the external data D_Ext to generate the preprocessed external data D_pExt. For a more detailed description of the external data pre-processing module 220 , further reference is made to FIG. 4 .

4 is a view for explaining an external data preprocessing module according to some embodiments of the present invention.

Referring to FIG. 4 , the external data preprocessing module 220 may preprocess the external data D_Ext to generate the preprocessed external data D_pExt. The external data pre-processing module 220 may include a second data generation time pre-processing module 221 , a weather information data pre-processing module 222 , and a weather forecast data pre-processing module 223 .

Similar to the above-described first data generation time preprocessing module 211 , the second data generation time preprocessing module 221 pre-processes the second data generation time, which is information about the time when the external data D_Ext is generated, to time series It can be converted into the form of data that can be input to the prediction module 240 .

The weather information data includes temperature data for the first area where the reservoir 100 is located, humidity data for the first area, sunlight data for the first area, and precipitation for the first area at the first time point and the second time point. It may contain data. The weather forecast data includes the highest temperature prediction data for the first area at the third time point predicted at the second time point, the lowest temperature prediction data for the first area at the third time point, and the first area at the third time point. It may include the average temperature prediction data for the first area and the precipitation probability prediction data for the first area at the third time point. In other words, the weather information data includes data about a first time point that is a past time point and a second time point that is a current time point, and the weather forecast data includes data about a third time point predicted from the second time point that is the current time point. can do.

According to some embodiments, the weather information data pre-processing module 222 may generate accumulated precipitation data by using precipitation data for the first region. In other words, the meteorological information data pre-processing module 222 uses the precipitation data for the first region to accumulate the precipitation data from the initialization to a specific time point in the first region to generate the accumulated precipitation data at the time point. can For example, the weather information data pre-processing module 222 may accumulate all the precipitation data from the beginning of the year to the first time point to generate the accumulated precipitation data at the first time point. However, this is an example and the embodiments are not limited thereto. For example, the external data D_Ext may include accumulated precipitation data up to a specific point in time. In this case, the weather information data preprocessing module 222 may not need to generate the accumulated precipitation data.

The weather information data pre-processing module 222 may pre-process the weather information data included in the external data D_Ext and convert it into a form of data that can be input to the time series prediction module 240 . For example, the weather information data pre-processing module 222 may normalize or scale the weather information data to convert it into a form of data that can be input to the time series prediction module 240 .

The weather forecast data pre-processing module 223 may pre-process the weather forecast data included in the external data D_Ext and convert it into a form of data that can be input to the time series prediction module 240 . For example, the weather forecast data pre-processing module 223 may normalize or scale the weather forecast data to convert it into a form of data that can be input to the time series prediction module 240 .

Referring back to FIG. 2 , the input data generating module 230 may receive the preprocessed reservoir management data D_pMng and the preprocessed external data D_pExt. The input data generation module 230 may generate the first input data D_In1 and the second input data D_In2 by using the pre-processed reservoir management data D_pMng and the pre-processed external data D_pExt.

The input data generation module 230 compares the preprocessed first data generation time and the preprocessed second data generation time to match the preprocessed reservoir management data (D_pMng) and the preprocessed external data (D_pExt) time points. have. Next, the first input data D_In1 may be generated by arranging the pre-processed reservoir management data D_pMng and the pre-processed external data D_pExt according to a predetermined arrangement at the same time point. Also, the input data generation module 230 may generate the second input data D_In2 by using some of the first input data D_In1 . In other words, the second input data D_In2 may be a part of the first input data D_In1 .

The first input data D_In1 may include data for a plurality of time points, and the second input data D_In2 may include data for a specific time point. According to some embodiments, the first input data D_In1 may include at least a portion of the preprocessed reservoir management data D_pMng and the preprocessed external data D_pExt for the first time point and the second time point. In addition, the second input data D_In2 may include partial data of the pre-processed reservoir management data D_pMng for the second time point.

For example, the first input data D_In1 includes information about a first time point (a first data generation time or a second data generation time), and temperature data of a first area where the reservoir 100 is located at the first time point. , humidity data of the first area at the first time point, the amount of sunshine data of the first area at the first time point, precipitation data of the first area at the first time point, accumulated precipitation data of the first area at the first time point, It may include storage water amount data at the first time point, flow rate data at the first time point, storage water amount change data at the first time point, accumulated flow data at the first time point, and data on whether a sluice gate is opened at the first time point.

In addition, the first input data D_In1 includes information on the second time point (the first data generation time or the second data generation time), the temperature data of the first area at the second time point, and the first area at the second time point. Humidity data of the first area at the second time point, the precipitation amount data of the first area at the second time point, the accumulated precipitation data of the first area at the second time point, the third time point predicted at the second time point of the highest temperature prediction data, the lowest temperature prediction data at the third time point predicted at the second time point, the average temperature prediction data at the third time point predicted at the second time point, and the precipitation probability prediction data at the third time point predicted at the second time point , stored water amount data at the second time point, flow rate data at the second time point, storage water quantity change data at the second time point, accumulated flow data at the second time point, and data on whether the sluice gate is opened at the second time point.

In other words, the first input data D_In1 may include reservoir management data D_Mng for a first time point that is a past time point and weather information data for the first time point. In addition, the first input data D_In1 may include reservoir management data D_Mng for the second time point that is the current time point, weather information data for the second time point, and weather forecast data for the second time point.

The second input data D_In2 may include data for the second time point. For example, the second input data D_In2 may include low water amount data at the second time point. The first input data D_In1 and the second input data D_In2 may be provided to the time series prediction module 240 .

For a more detailed description of the time series prediction module 240 , further reference is made to FIG. 5 .

5 is a diagram for explaining a time series prediction module according to some embodiments of the present invention.

Referring to FIG. 5 , the time series prediction module 240 may include an encoder 241 , a first decoder 242 , and a second decoder 243 . For reference, the time series prediction module 240 may use an LSTM-based prediction model and a ReLU function as an activation function, but embodiments are not limited thereto.

The encoder 241 may receive the first input data D_In1 . The encoder 241 may generate the context vector CV by using the first input data D_In1 . The context vector CV generated by the encoder 241 may be provided to the first decoder 242 and the second decoder 243 , respectively. In other words, the context vector CV generated by the encoder 241 may be provided to the first decoder 242 , and the same context vector CV may also be provided to the second decoder 243 .

The first decoder 242 may receive the above-described context vector CV and the second input data D_In2 . The first decoder 242 may generate the stored water inflow prediction data Es_In of the reservoir 100 for the third time point by using the context vector CV and the second input data D_In2 .

Similarly, the second decoder 243 may receive the context vector CV and the second input data D_In2 . The second decoder 243 may use the context vector CV and the second input data D_In2 to generate the stored water outflow prediction data Es_Out of the reservoir 100 for the third time point.

The time series prediction module 240 uses the stored water inflow prediction data Es_In generated by the first decoder 242 and the stored water outflow prediction data Es_Out generated by the second decoder 243 to predict the amount of change in the storage water quantity D_Es ) can be created. For a more detailed description, refer further to FIG. 6 .

6 is a diagram for explaining an example of a method of generating a storage amount change amount prediction data using a time series prediction module according to some embodiments of the present invention. In FIG. 6 , it is assumed that the first input data D_In1 includes data for s time points, and it is assumed that data for k future time points are predicted based on the time point t. In other words, the time points t-s-1 to t-1 are the first time points, the time points t are the second time points, and the time points t+1 to t+k are the third time points.

Referring to FIG. 6 , the encoder 241 may include s first prediction module cells 241_1 , 241_2 , ..., 241_s. For convenience of description, the s first prediction module cells 241_1, 241_2, ..., 241_s included in the encoder 241 are 1-1 prediction module cells 241_1 and 1-2 prediction module cells. (241_2), … , expressed as the 1-s prediction module cell 241_s.

The first prediction module cells 241_1, 241_2, ..., 241_s may be LSTM cells. However, embodiments are not limited thereto, and the first prediction module cells 241_1, 241_2, ..., 241_s may be configured as cells for various time series prediction modules such as RNN cells and GRU cells in addition to LSTM cells.

The 1-1 prediction module cell 241_1 may be provided with the first input data D_In1 at time t-s-1. In addition, the 1-2 prediction module cell 241_2 may be provided with the first input data D_In1 at time t-s. Similarly, the 1-s-th prediction module cell 241_s may be provided with the first input data D_In1 at time t. In FIG. 6, x means data related to the reservoir management data D_Mng, and x* means data related to the external data D_Ext, but embodiments are not limited thereto. The output from the 1-1 prediction module cell 241_1 may be provided to the 1-2 prediction module cell 241_2 . Similarly, the 1-s-th prediction module cell 241_s may receive the output of the 1-(s-1) prediction module cell. Outputs of the first prediction module cells 241_1, 241_2, ..., 241_s may include a hidden state vector. In addition, the first prediction module cells 241_1, 241_2, ..., 241_s may output a cell state vector for minimizing the gradient loss phenomenon, but embodiments are not limited thereto. It will be described that one prediction module cell 241_1, 241_2, ..., 241_s outputs a hidden state vector.

The 1-s prediction module cell 241_s generates a context vector CV using the hidden state vector provided from the 1-(s-1) prediction module cell and the first input data D_In1 at time t. can do. The context vector CV may be provided to the first decoder 242 and the second decoder 243 , respectively.

In summary, for each of the s first prediction module cells 241_1 , 241_2 , ..., 241_s, s pieces of first input data D_In1 may be provided according to a time series. Each output of the first prediction module cells 241_1, 241_2, ..., 241_s is transmitted to the following first prediction module cells 241_1, 241_2, ..., 241_s, and finally 1-s prediction An output of the module cell 241_s may be provided to the first decoder 242 and the second decoder 243 as a context vector CV.

The first decoder 242 may include k second prediction module cells 242_1, 242_2, ..., 242_k. For convenience of description, k second prediction module cells 242_1, 242_2, ..., 242_k included in the first decoder 242 are 2-1 prediction module cells 242_1 and 2-2 prediction. It is expressed as the module cell 242_2, ..., the 2-k-th prediction module cell 242_k. Similarly, the second decoder 243 may include k third prediction module cells 243_1, 243_2, ..., 243_k. For convenience of description, the k third prediction module cells 243_1, 243_2, ..., 243_k included in the second decoder 243 are defined as the 3-1 prediction module cell 243_1 and the 3-2 prediction. It is expressed as the module cell 243_2, ..., the 3-k-th prediction module cell 243_k.

The second prediction module cells 242_1, 242_2, ..., 242_k and the third prediction module cells 243_1, 243_2, ..., 243_k may be LSTM cells. However, embodiments are not limited thereto, and the second prediction module cells 242_1, 242_2, ..., 242_k and the third prediction module cells 243_1, 243_2, ..., 243_k are RNN cells in addition to the LSTM cells. and cells for various time series prediction modules such as GRU cells.

The 2-1 prediction module cell 242_1 may receive the context vector CV and the second input data D_In2. The second input data D_In2 may refer to data on the amount of water stored in the reservoir 100 at time t (second time point). The 2-1 prediction module cell 242_1 uses the context vector CV and the second input data D_In2 to generate the stored water inflow prediction data Es_In of the reservoir 100 at time t+1. have. The stored water inflow prediction data Es_In may be expressed as a positive number.

Similarly, the 3-1 prediction module cell 243_1 may receive the context vector CV and the second input data D_In2. The second input data D_In2 may mean data of the amount of water stored in the reservoir 100 at time t. The 3-1 prediction module cell 243_1 uses the context vector CV and the second input data D_In2 to generate the storage water runoff prediction data Es_Out of the reservoir 100 at time t+1. have. The stored water runoff prediction data Es_Out may be expressed as a negative number.

The time series prediction module 240 includes the stored water inflow prediction data Es_In at time t+1 generated in the 2-1 prediction module cell 242_1 and the time t+ generated in the 3-1 prediction module cell 243_1. By using (eg, summing up) the storage water runoff prediction data Es_Out at 1, the storage water quantity change prediction data D_Es at time t+1 may be generated.

Meanwhile, the 2-1 prediction module cell 242_1 may generate a hidden state vector using the context vector CV and the second input data D_In2 and provide it to the 2-2 prediction module cell 242_2. have. Also, the storage amount change prediction data D_Es at time t+1 may be provided to the 2-2 prediction module cell 242_2. The 2-2 prediction module cell 242_2 uses the hidden state vector provided from the 2-1 prediction module cell 242_1 and the low water quantity change prediction data D_Es at the time t+1, at the time t+2 It is possible to generate the stored water inflow prediction data Es_In. Similarly, the 3-2 prediction module cell 243_2 uses the hidden state vector provided from the 3-1 prediction module cell 243_1 and the low quantity change amount prediction data D_Es at time t+1, It is possible to generate the stored water runoff prediction data Es_Out at time t+2. The time series prediction module 240 includes the stored water inflow prediction data Es_In at time t+2 generated in the 2-2 prediction module cell 242_2, and the time t generated in the 3-2 prediction module cell 243_2. By using the storage water runoff prediction data Es_Out at +2, the storage water amount change prediction data D_Es at time t+2 may be generated. In a similar manner, the 2-k-th prediction module cell 242_k and the 3-k-th prediction module cell 243_k may generate the low water quantity change amount prediction data D_Es at time t+k.

As described above, the time series prediction module 240 according to some embodiments includes preprocessed reservoir management data (D_pMng) at first and second time points, preprocessed weather information data at first and second time points, and Using the pre-processed weather forecast data predicted at the second time point, the amount of change in the amount of water stored in the reservoir 100 may be predicted.

Referring again to FIG. 2 , the test module 250 may receive the low water quantity change amount prediction data D_Es from the time series prediction module 240 . Also, the test module 250 may receive the reservoir management data D_Mng. The test module 250 may generate the error rate information Re by comparing the storage amount change amount prediction data D_Es and the storage water amount change amount data included in the reservoir management data D_Mng.

The error rate information Re is for determining whether the time series prediction module 240 operates properly, that is, whether training is performed well. If the error rate information Re is greater than or equal to a predetermined criterion, the time series prediction module 240 may determine the overfitting or underfitting state and train the time series prediction module 240 again. In order to determine whether the time series prediction module 240 is overfitting or underfitting, data used when training the time series prediction module 240 and data for testing the training state of the time series prediction module 240 are separated should be It will be described with further reference to FIG. 7 .

7 is a diagram for explaining a training data set and a test data set according to some embodiments of the present invention.

Referring to FIG. 7 , the reservoir management data (D_Mng) may include training reservoir management data (D_Mng) from time t1 to time t2, and reservoir management data (D_Mng) for testing from time t2 to time t3. . In addition, the external data D_Ext may include external training data D_Ext from time t1 to time t2 and external data D_Ext for testing from time t2 to time t3.

The time series prediction module 240 according to some embodiments of the present invention may optimize internal parameters using the training reservoir management data D_Mng and the training external data D_Ext. After the training of the time series prediction module 240 is primarily completed, the time series prediction module 240 uses the reservoir management data for testing (D_Mng) and the external data for testing (D_Ext) to train the time series prediction module 240 state can be tested. In this case, the test module 250 may calculate the error rate information Re by comparing the stored water quantity change amount prediction data D_Es generated by the time series prediction module 240 with the actual stored water quantity change amount data. When the error rate information Re calculated by the test module 250 exceeds a predetermined criterion, the time series prediction module 240 may be trained again to adjust the parameters.

8 is a graph of experimental results for explaining a difference between predicted data on the amount of change in the amount of stored water and the actual amount of change in the amount of stored water according to some embodiments of the present invention.

Referring to FIG. 8 , the portion indicated by the red line is the actual amount of change in water storage, and the portion indicated by the blue line is the predicted data of the amount of change in the water storage determined using the system for predicting the amount of change in the storage according to some embodiments of the present invention. As shown in FIG. 8 , it can be seen that the error rate is very small between the stored water quantity change prediction data and the actual stored water quantity change amount. represents

9 is a view for explaining a method of predicting a change in the amount of stored water according to some embodiments of the present invention. For convenience of description, content overlapping or similar to those described above will be omitted or simply described.

Referring to FIG. 9 , reservoir management data D_Mng and external data D_Ext may be provided ( S100 ). Subsequently, the reservoir management data D_Mng and the external data D_Ext may be pre-processed (S110). First input data D_In1 and second input data D_In2 may be generated using the pre-processed reservoir management data D_pMng and the pre-processed external data D_pExt ( S120 ).

The first input data D_In1 may be provided to the encoder 241 ( S130 ). The encoder 241 may generate the context vector CV by using the first input data D_In1 . The generated context vector CV may be provided to the first decoder 242 and the second decoder 243 , respectively. Also, the second input data D_In2 may be provided to the first decoder 242 and the second decoder 243, respectively (S140). The first decoder 242 may generate the stored water inflow prediction data Es_In by using the context vector CV and the second input data D_In2 ( S150 ). Also, the second decoder 243 may generate the stored water runoff prediction data Es_Out by using the context vector CV and the second input data D_In2 ( S160 ). By using the stored water inflow prediction data Es_In and the stored water outflow prediction data Es_Out, the stored water quantity change prediction data D_Es may be generated ( S170 ).

FIG. 10 is a diagram for explaining an example of a method of generating a storage amount change amount prediction data using a time series prediction module according to some other embodiments of the present invention. For convenience of description, content overlapping or similar to those described above will be omitted or simply described.

Referring to FIG. 10 , the encoder 241 may further include an attention layer AL. The attention layer AL includes information on the hidden state vectors of the plurality of first prediction module cells 241_1, 241_2, ..., 241_s. When generating the stored water inflow prediction data Es_In, the first decoder 242 may refer to the attention layer AL. Also, the second decoder 243 may refer to the attention layer AL when generating the stored water runoff prediction data Es_Out. The attention layer (AL) is a layer for further referencing a related part of the entire hidden state vector of the encoder 241 at every time point when the first decoder 242 and the second decoder 243 generate prediction data. , accuracy of prediction may be improved by introducing an attention layer (AL).

11 is a view for explaining a method for predicting a change in the amount of stored water according to another exemplary embodiment of the present invention. For convenience of description, content overlapping or similar to those described above will be omitted or simply described.

Referring to FIG. 11 , reservoir management data D_Mng and external data D_Ext may be provided ( S200 ). Subsequently, the reservoir management data D_Mng and the external data D_Ext may be pre-processed ( S210 ). The first input data D_In1 and the second input data D_In2 may be generated using the pre-processed reservoir management data D_pMng and the pre-processed external data D_pExt ( S220 ).

The first input data D_In1 may be provided to the encoder 241 ( S230 ). The encoder 241 may generate the context vector CV by using the first input data D_In1 . The generated context vector CV may be provided to the first decoder 242 and the second decoder 243 , respectively. Also, the second input data D_In2 may be provided to the first decoder 242 and the second decoder 243, respectively (S240). The first decoder 242 may generate the stored water inflow prediction data Es_In by using the context vector CV, the second input data D_In2, and the attention layer AL (S250). Also, the second decoder 243 may generate the stored water outflow prediction data Es_Out by using the context vector CV, the second input data D_In2, and the attention layer AL (S260). By using the stored water inflow prediction data Es_In and the stored water outflow prediction data Es_Out, the stored water quantity change prediction data D_Es may be generated ( S270 ).

The above description is merely illustrative of the technical idea of this embodiment, and various modifications and variations will be possible without departing from the essential characteristics of the present embodiment by those of ordinary skill in the art to which this embodiment belongs. Accordingly, the present embodiments are intended to explain rather than limit the technical spirit of the present embodiment, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the claims below, and all technical ideas within the scope equivalent thereto should be construed as being included in the scope of the present embodiment.

Claims (10)

Water storage amount data for the first reservoir measured at a first time point, water storage amount data for the first reservoir measured at a second time point subsequent to the first time point, and flow rate for the first reservoir measured at the first time point Data, flow data for the first reservoir measured at the second time point, and information on the first time point and the second time point are provided, and the water storage amount data for the first reservoir measured at the first time point, and The storage amount change data at the first time point is generated using the flow rate data for the first reservoir measured at the first time point, and the water storage amount data for the first reservoir measured at the second time point and the second time point are used. The storage amount change data at the second time point is generated using the flow rate data for the first reservoir measured at the first time point, and the flow rate data for the first reservoir measured at the first time point is used to generate the data from the initialization. Generates accumulated flow data up to a first time point, and generates accumulated flow data from the initialization to the second time point by using the flow data for the first reservoir measured at the second time point, and the first time point The sluice gate opening estimation data estimated at the first time point is generated by using the water storage amount data for the first reservoir measured in and flow data data for the first reservoir measured at the first time point, The sluice gate opening estimation data estimated at the second time point is generated using the measured water storage amount data for the first reservoir and the flow rate data for the first reservoir measured at the second time point, and measured at the first time point the amount of water stored in the first reservoir, the amount of water stored in the first reservoir measured at the second time point, the flow rate data for the first reservoir measured at the first time point, and the amount of water measured at the second time point Flow rate data for the first reservoir, storage water amount change data at the first time point, storage water quantity change data at the second time point amount data, accumulated flow rate data up to the first time point, accumulated flow rate data up to the second time point, sluice opening estimation data estimated at the first time point, sluice opening estimation data estimated at the second time point, and the second time point; a first data pre-processing module that performs a pre-processing operation on information about a first time point and the second time point;
The temperature data of the first reservoir at the first time point, the temperature data of the first reservoir at the second time point, the humidity data of the first reservoir at the first time point, the second time point at the second time point 1 humidity data of the reservoir, the amount of sunshine data of the first reservoir at the first time point, the amount of sunshine data of the first reservoir at the second time point, the precipitation data of the first reservoir at the first time point, the first The precipitation amount data of the first reservoir at two time points, the maximum temperature prediction data for the first reservoir at a third time point that is later than the second time point predicted at the second time point, and the predicted data at the second time point Minimum temperature prediction data for the first reservoir at a third time point, average temperature prediction data for the first reservoir at the third time point predicted at the second time point, and the third predicted at the second time point Precipitation probability prediction data for the first reservoir at a time point is provided, and accumulation of the first reservoir from the initialization to the first time point using the precipitation amount data for the first reservoir at the first time point generating precipitation data, and generating accumulated precipitation data for the first reservoir from the initialization to the second time point using the precipitation data for the first reservoir at the second time point, and at the first time point of the temperature data of the first reservoir, the temperature data of the first reservoir at the second time point, the humidity data of the first reservoir at the first time point, and the humidity data of the first reservoir at the second time point , the sunshine amount data of the first reservoir at the first time point, the sunshine amount data of the first reservoir at the second time point, the precipitation amount data of the first reservoir at the first time point, the said at the second time point Precipitation amount data of the first reservoir, maximum temperature prediction data for the first reservoir at the third time point predicted at the second time point, and phase predicted at the second time point The lowest temperature prediction data for the first reservoir at the third time point, the average temperature prediction data for the first reservoir at the third time point predicted at the second time point, and the second predicted data at the second time point Preprocessing operation on the precipitation probability prediction data for the first reservoir at three time points, the accumulated precipitation data for the first reservoir up to the first time point, and the accumulated precipitation data for the first reservoir up to the second time point a second data pre-processing module that performs;
an input data generating module receiving the outputs of the first data preprocessing module and the second data preprocessing module and generating first input data and second input data; and
A time series comprising an encoder provided with the first input data, a first decoder provided with an output from the encoder and the second input data, and a second decoder provided with an output from the encoder and the second input data a prediction module;
The time series prediction module predicts the amount of change in the amount of storage in the first reservoir using the output of the first decoder and the output of the second decoder,
The first input data may include water storage data for the first reservoir measured at the first time point, water storage data data for the first reservoir measured at the second time point, and the first reservoir measured at the first time point. flow rate data, flow rate data for the first reservoir measured at the second time point, storage water quantity change data at the first time point, storage water quantity change data data at the second time point, accumulated flow data up to the first time point, the Cumulative flow data up to a second time point, sluice opening estimation data estimated at the first time point, sluice opening estimation data estimated at the second time point, temperature data of the first reservoir at the first time point, and the second The temperature data of the first reservoir at the time point, the humidity data of the first reservoir at the first time point, the humidity data of the first reservoir at the second time point, the humidity data of the first reservoir at the first time point sunshine amount data, sunshine amount data of the first reservoir at the second time point, precipitation amount data of the first reservoir at the first time point, precipitation amount data of the first reservoir at the second time point, at the second time point The predicted highest temperature prediction data for the first reservoir at the third time point, the lowest temperature prediction data for the first reservoir at the third time point predicted at the second time point, and the predicted second time point The average temperature prediction data for the first reservoir at the third time point, the precipitation probability prediction data for the first reservoir at the third time point predicted at the second time point, and the first time up to the first time point including accumulated precipitation data for the reservoir and accumulated precipitation data for the first reservoir up to the second time point,
The second input data may include water storage amount data for the first reservoir measured at the second time point, flow rate data for the first reservoir measured at the second time point, water storage amount change data at the second time point, and the second input data. Cumulative flow data up to two time points, sluice opening estimation data estimated at the second time point, temperature data of the first reservoir at the second time point, humidity data of the first reservoir at the second time point, and the second time point Sunlight data of the first reservoir at two time points, precipitation data of the first reservoir at the second time point, maximum temperature prediction data for the first reservoir at the third time point predicted at the second time point, The lowest temperature prediction data for the first reservoir at the third time point predicted at the second time point, the average temperature prediction data for the first reservoir at the third time point predicted at the second time point, and the second time point Precipitation probability prediction data for the first reservoir at the third time point predicted at the second time point and the accumulated precipitation data for the first reservoir up to the second time point,
Low water volume change prediction system.
delete delete delete delete delete delete delete delete delete

Images