Method For Generating Composite Drought Index Based On Artificial Intelligence Technology And A Computer-Readable Recording Medium On Which The Program To Perform The Method Is Recorded

Abstract

The present invention relates to a method for generating a composite drought index based on artificial intelligence technology and a computer-readable recording medium having a program for performing the same. More specifically, the invention relates to a method for generating a composite drought index based on artificial intelligence technology and a computer-readable recording medium having a program for performing the same, comprising an input weight extraction step in which input weights for input variables, namely a meteorological index, a hydrological index, and an agricultural index, are respectively extracted using an autoencoder, and an index generation step in which a composite drought index reflecting meteorological, hydrological, and agricultural causes is generated using the input weights for each index.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence technology-based method for generating a composite hydrological index to integrate various drought indices used in the water resources field, considering the form and use of water resources, into a single hydrological factor, and a computer-readable recording medium having a program for performing the same. Background Technology

[0002] Drought is caused by a lack of precipitation over an extended period. Due to recent climate change, the frequency of droughts is increasing. To mitigate the frequency of droughts, various indicators are used in drought monitoring, assessment, and forecasting. Generally, the causes of drought are classified into three basic categories: meteorological, hydrological, and agricultural.

[0003] First, meteorological drought is the result of a lack of precipitation, and indices including the Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), Palmer Drought Severity Index (PDSI), and Z-index are used to monitor and evaluate meteorological drought.

[0004] Next, hydrological drought is associated with reduced river flow and insufficient reservoir storage, and indices including the Standardized Reservoir Supply Index (SRSI), Palmer Hydrological Drought Index (PHDI), Streamflow Drought Index (SDI), and Surface Water Supply Index (SWSI) are used to monitor and evaluate hydrological drought.

[0005] Next, agricultural drought is caused by a decrease in soil moisture, and indices including the Soil Moisture Index (SMI) and the Standardized Soil Moisture Index (SSI) are used to monitor and evaluate agricultural drought.

[0006] As mentioned above, conventional drought has been expressed using an index for a single cause; however, since drought-stricken regions can be affected by two or more causes, the Aggregate Drought Index (ADI), Combined Drought Index (CDI), and Multivariate Standardized Drought Index (MSDI) have been proposed.

[0007] Conventional composite indices were derived through Principal Component Analysis (PCA). Principal Component Analysis (PCA) is a technique that reduces high-dimensional data to low-dimensional data, and it linearly transforms the data into a new coordinate system such that when the data is mapped to a single axis, the axis with the largest variance is the first principal component and the axis with the second largest variance is the second principal component.

[0008] However, since composite indices are influenced by various factors, there are technical limitations in accurately analyzing and predicting the causes of drought using linear interpretation alone. Therefore, non-linear analysis methods are urgently needed in this technical field to accurately analyze and predict the complex causes of drought. Prior art literature

[0009] Korean Registered Patent Publication No. 10-1954570 Korean Registered Patent Publication No. 10-1651747 The problem to be solved

[0010] The present invention aims to solve the aforementioned problems by providing a method for generating a complex drought index based on artificial intelligence technology to accurately analyze and predict the causes of non-linearly complex drought occurrences, and a computer-readable recording medium having a program for performing the same.

[0011] The technical problems that the present invention aims to solve are not limited to those mentioned above, and other unmentioned technical problems can be clearly understood by those skilled in the art from the description of the present invention. means of solving the problem

[0012] To achieve the above objective, the artificial intelligence technology-based composite drought index generation method of the present invention provides: an input weight extraction step in which an autoencoder is utilized by at least one processor to extract input weights for input variables, namely a meteorological index, a hydrological index, and an agricultural index; and an index generation step in which the input weights for each index are utilized by at least one processor to generate a composite drought index that reflects all meteorological causes, hydrological causes, and agricultural causes.

[0013] To achieve the above objective, the present invention provides a computer-readable recording medium having a program recorded thereon for performing the artificial intelligence technology-based complex drought index generation method of the present invention. Effects of the invention

[0014] As described above, according to the present invention, an autoencoder is utilized to generate a composite drought index that reflects meteorological, hydrological, and agricultural causes, thereby enabling the accurate analysis and prediction of complex non-linear causes of drought.

[0015] In addition, the present invention has the effect of rapidly generating a composite drought index by improving the model performance of the autoencoder through not estimating or extracting output weights and output variables that are not used at all in the process of generating the index.

[0016] The effects of the present invention are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art from the detailed description and claims. Brief explanation of the drawing

[0017] Figure 1 is a flowchart of the artificial intelligence technology-based composite drought index generation method of the present invention. FIG. 2 is a drawing showing meteorological data (a), hydrological data (b) and agricultural data (c) according to an embodiment of the present invention. FIG. 3 is a graph showing time series data for meteorological indices, hydrological indices, and agricultural indices according to an embodiment of the present invention. FIG. 4 is a structural diagram of an autoencoder according to an embodiment of the present invention. FIG. 5 is a detailed flowchart of a method for generating a composite drought index based on artificial intelligence technology according to an embodiment of the present invention. FIG. 6 is a detailed flowchart of a method for generating a composite drought index based on artificial intelligence technology according to this embodiment of the present invention. FIG. 7 is a detailed flowchart of a method for generating a composite drought index based on artificial intelligence technology according to the third embodiment of the present invention. FIG. 8 is a detailed flowchart of a method for generating a composite drought index based on artificial intelligence technology according to a factual example of the present invention. Specific details for implementing the invention

[0018] The terms used in this specification have been selected based on currently widely used general terms whenever possible, taking into account their functions in the present invention; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, terms used in this invention should be defined not merely by their names, but based on their meanings and the overall content of the invention.

[0019] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by those skilled in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant technology, and should not be interpreted in an ideal or overly formal sense unless explicitly defined in this application.

[0020] Hereinafter, embodiments according to the present invention will be described in detail with reference to the attached drawings. First, the present invention includes a computer-readable recording medium (120) on which a program for performing a method for generating a complex drought index based on artificial intelligence technology is recorded. The recording medium (120) may be, for example, a CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM, etc. Furthermore, the method for generating a complex drought index based on artificial intelligence technology according to the present invention may be implemented by at least one processor (110) within a computer device (100) reading the recording medium (120).

[0021] Referring to FIG. 1, the artificial intelligence technology-based composite drought index generation method of the present invention utilizes an autoencoder by at least one processor (110) to input weights (w) for input variables (x), namely a meteorological index (x1), a hydrological index (x2), and an agricultural index (x3). Ii An input weight extraction step (S100) in which ) are each extracted, and by the at least one processor (110), an input weight (w) for each index Ii It includes an index generation step (S200) in which a composite drought index (ACDI) reflecting meteorological, hydrological, and agricultural causes is generated using ).

[0022] According to one embodiment of the present invention, the meteorological index is the Standardized Precipitation Index (SPI), and the hydrological index and agricultural index are the Standardized Reservoir Supply Index (SRSI). SPI, SRSI(A), and SRSI(H) are conventional single drought indices.

[0023] Referring to an embodiment of FIG. 2, an agricultural index can be derived by inputting agricultural data measured from a region of interest, such as South Korea, into a pre-established algorithm. The agricultural data may include reservoir data for detailed regions within the region of interest or data missing. A hydrological index can be derived by inputting hydrological data measured from a region of interest, such as South Korea, into a pre-established algorithm. The hydrological data may include river flow, agricultural reservoir flow, water supply dam flow, multi-purpose dam flow, and data missing for detailed regions within the region of interest. A meteorological index can be derived by inputting meteorological data measured from a region of interest, such as South Korea, into a pre-established algorithm. The meteorological data may include rainfall observed from ASOS, a synoptic meteorological observation instrument, for detailed regions within the region of interest, and basin-averaged rainfall using the Thiessen method.

[0024] Looking at an embodiment of FIG. 3, changes in SRSI (A) and SRSI (H) can be observed over the time range from 2000 to 2020. Also, changes in SPI expressed using different cumulative months of 1 month, 3 months, 6 months, 9 months, and 12 months can be observed over the time range from 2000 to 2020. That is, in the input weight extraction step (S100), the meteorological index (x1), hydrological index (x2), and agricultural index (x3), which are the time series data described above, can be input at the same time point as the input variable (x).

[0025] Referring to an embodiment of FIG. 4, the autoencoder mentioned in the present invention is an artificial intelligence neural network utilizing unsupervised learning and may include an encoder and a decoder. Here, the encoder includes an input layer and a hidden layer, and the decoder includes a hidden layer and an output layer, so that the encoder and the decoder have a structure symmetrical with respect to the hidden layer. That is, the number of input variables input to the encoder and the number of output variables output from the decoder are the same. In addition, the number of hidden layers may be multiple, but according to an embodiment of the present invention, it may be one.

[0026] Accordingly, in the input weight extraction step (S100), meteorological indices (x1), hydrological indices (x2), and agricultural indices (x3) may be input as input variables (x) of the input layer within the encoder. Furthermore, in the input weight extraction step (S100), input weights (w) for each index are used in the encoding process where each index is mapped to a low-dimensional hidden layer and dimensionality is reduced. I1 , w I2 , w I3 Each of ) can be extracted.

[0027] A conventional autoencoder has an input variable (x) input to the encoder and an output variable ( Unsupervised learning is performed in the direction of minimizing the difference between ), that is, the loss function of [Equation 1] below.

[0028]

[0029] Here, L AE is the loss function value of a conventional autoencoder, x is the input variable, and is an output variable.

[0030] However, the present invention does not use any output variables in the process of generating the index. The output weight (w) output during the decoding process. Oi ) and output variables( i If estimation and extraction are performed even though ) is not used at all, the model performance of the autoencoder of the present invention may be degraded.

[0031] To solve this problem, the input weight extraction step (S100) comprises input weights (w) for each exponent that minimize the difference between the input variable (x) and the hidden variable (z) estimated from the encoder within the autoencoder, i.e., the loss function of [Equation 2] below, so that the decoder within the autoencoder does not operate. I1 , w I2 , w I3 It is characterized by the fact that each ) is extracted.

[0032]

[0033] Here, L AE is the loss function value of the autoencoder of the present invention, x is the input variable, and z is the hidden variable. At this time, the hidden variable (z) can be substituted as it can be calculated using the following [Equation 3].

[0034]

[0035] Here, W I is the input weight (w) for each exponent I1 , w I2 , w I3 ) and, b I is the input bias parameter, and σ is the activation function.

[0036] In other words, the input weight extraction step (S100) is an input weight (w) such that the loss function of [Equation 2], in which the hidden variable (z) is replaced by [Equation 3], is minimized. I1 , w I2 , w I3 ) and bias parameter (b IEach of ) can be extracted.

[0037] Therefore, the present invention has a significant effect of improving the model performance of an autoencoder and rapidly extracting and generating the input weights to be extracted and the composite drought index to be generated by not estimating or extracting output variables that are not used at all in the process of generating an index without operating a decoder.

[0038] Next, according to an embodiment of FIG. 5, the index generation step (S200) comprises input weights (w) for each index to a linear function. I1 , w I2 , w I3 By substituting ), the hidden variable (z) of the hidden layer within the autoencoder can be computed (S201). And, the index generation step (S200) is characterized in that the hidden variable (z) computed by a linear function is generated as the composite drought index (S203).

[0039] The linear function mentioned in the present invention is the activation function (σ) of [Equation 3] above, and may be a sigmoid function or a ReLU (Rectified Linear Unit) function. Therefore, it is related to the Principal Component Analysis (PCA) used to derive a composite exponent in that the hidden variable (z) of the autoencoder corresponds to a principal component (PC).

[0040] In addition, when the linear function is a sigmoid function, the hidden variable (z) calculated from the sigmoid function exists in the range of [0 and 1]. Since the sigmoid function has already operated as a cumulative distribution function when calculating the hidden variable (z), the index generation step (S200) can omit a separate transformation process to fit the hidden variable (z) to a probability distribution and has the significant effect of generating the hidden variable (z) as the composite drought index.

[0041] According to one embodiment of FIG. 6, the index generation step (S200) comprises input weights (w) for each index to a hyperbolic tangent (tanh) function. I1 , w I2 , w I3 By substituting ), the hidden variable (z) of the hidden layer within the autoencoder can be computed (S204). And the index generation step (S200) is characterized in that the hidden variable (z) computed by the hyperbolic tangent (tanh) function is generated as the composite drought index (S206).

[0042] The hyperbolic tangent (tanh) function mentioned in the present invention is the activation function (σ) of [Equation 3] above, and is as shown in [Equation 4] below.

[0043]

[0044] Here, W I is the input weight (w) for each exponent I1 , w I2 , w I3 ) and, b I is the input bias parameter. The range of the hyperbolic tangent (tanh) function is [-1,1].

[0045] Meanwhile, the above-mentioned index generation step (S200) is characterized by the fact that the range of the hidden variable (z) calculated from the hyperbolic tangent (tanh) function is transformed using the following [Mathematical Formula 5] (S207).

[0046]

[0047] Here, Φ -1is the inverse function of the Cumulative Distribution Function (CDF) of the standard normal variable, and z is a hidden variable calculated from the hyperbolic tangent (tanh) function. And (z+1) / 2 in [Equation 5] above is a formula applied to make the hidden variable (z) calculated from the hyperbolic tangent (tanh) function into a non-negative probability distribution function in the range [0,1] from the range [-1,1].

[0048] Accordingly, the index generation step (S200) utilizes the hyperbolic tangent (tanh) function, thereby enabling the generation of a composite drought index capable of expressing drought conditions with extremely negative values, and the generated composite drought index can be produced as a non-negative value, similar to other composite indices. Accordingly, the present invention has a significant effect in that a composite drought index capable of expressing drought conditions more accurately and precisely within a numerical range can be generated.

[0049] Next, the input weight extraction step (S100) comprises input weights (w) for each exponent so that the following [Equation 6] holds. I1 , w I2 , w I3 It is characterized by including a weight conversion step (S110) in which ) is converted.

[0050]

[0051] Here, w Ii is the input weight for each index. Since the relationship between input weights cannot be a negative value, all input weights converted from the weight conversion step (S110) are positive.

[0052] Referring to an embodiment of FIG. 7, the input weight conversion step (S110) comprises an input weight (w) for each index. I1 , w I2 , w I3) can be transformed so that the above [Mathematical Equation 6] holds. Subsequently, in the index generation step (S200), the transformed input weights for each index are substituted into a linear function, thereby allowing the hidden variable (z) of the hidden layer within the autoencoder to be calculated (S202). Then, in the index generation step (S200), the hidden variable (z) calculated by the linear function can be generated as a composite drought index (S203).

[0053] Referring to an embodiment of FIG. 8, the input weight conversion step (S110) comprises an input weight (w) for each index. I1 , w I2 , w I3 ) can be transformed so that the above [Mathematical Equation 6] holds. Subsequently, in the index generation step (S200), the transformed input weights for each index are substituted into the hyperbolic tangent function, thereby allowing the hidden variable (z) of the hidden layer within the autoencoder to be calculated (S205). Then, in the index generation step (S200), the hidden variable (z) calculated with the hyperbolic tangent function can be generated as a composite drought index (S206).

[0054] The model structure of the autoencoder is not related to the physical characteristics of the input variable (x). However, according to the present invention, the input weight (w) for each exponent from the input weight conversion step (S110) I1 , w I2 , w I3 ) is transformed proportionally, and at the same time, the input weight (w) for each exponent I1 , w I2 , w I3 By converting the sum of ) to 1, there is a significant effect of being able to reflect the physical characteristics of the input variable (x) in the autoencoder of the present invention.

[0055] The embodiments may be implemented by hardware, software, firmware, middleware, microcode, a hardware description language, or any combination thereof. Where implemented by software, firmware, middleware, or microcode, program code or code segments that perform the necessary tasks may be stored on a computer-readable storage medium and executed by one or more processors.

[0056] Furthermore, aspects of the subject matter described herein may be described in the general context of computer-executable instructions, such as program modules or components executed by a computer. Generally, program modules or components include routines, programs, objects, and data structures that perform specific tasks or implement specific data types. The aspects of the subject matter described herein may be implemented in distributed computing environments where tasks are performed by remote processing devices linked through a communication network. In a distributed computing environment, program modules may be located on both local and remote computer storage media, including memory storage devices.

[0057] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results may be achieved even if the described techniques are performed in a different order than described, and / or the components of the system, structure, device, circuit, etc. described are combined or assembled in a form different from the described method, or are replaced or substituted by other components or equivalents.

[0058] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below. Explanation of the symbols

[0059] 100.. Computer device 110.. at least one processor 120.. Recording media S100.. Input weight extraction step S110.. Input weight conversion step S200.. Index generation stage

Claims

Claim 1 A method for generating a composite drought index based on artificial intelligence technology, comprising: an input weight extraction step in which an autoencoder is utilized by at least one processor to extract input weights for input variables, namely a meteorological index, a hydrological index, and an agricultural index; and an index generation step in which the input weights for each index are utilized by at least one processor to generate a composite drought index that reflects all meteorological causes, hydrological causes, and agricultural causes; wherein the input weight extraction step is characterized in that input weights for each index are extracted such that the difference between the input variables and the hidden variables estimated from the encoder within the autoencoder is minimized so that the decoder within the autoencoder does not operate. Claim 2 A method for generating a composite drought index based on artificial intelligence technology according to claim 1, wherein the index generation step comprises calculating a hidden variable of a hidden layer within an autoencoder by substituting an input weight for each index into a linear function, and generating the hidden variable calculated by the linear function as the composite drought index. Claim 3 A method for generating a composite drought index based on artificial intelligence technology according to claim 1, wherein the index generation step comprises calculating a hidden variable of a hidden layer within the autoencoder by substituting an input weight for each index into a hyperbolic tangent (tanh) function, and generating the hidden variable calculated by the hyperbolic tangent (tanh) function as the composite drought index. Claim 4 A method for generating a composite drought index based on artificial intelligence technology according to claim 1, wherein the input weight extraction step comprises an input weight conversion step in which the input weight for each index is converted so that the following [Equation 6] is satisfied. [Equation 6] Here, w Ii is the input weight for each index. Claim 5 A method for generating a composite drought index based on artificial intelligence technology according to claim 1, characterized in that the meteorological index is the Standardized Precipitation Index (SPI), and the hydrological index and agricultural index are the Standardized Reservoir Supply Index (SRSI). Claim 6 A computer-readable recording medium having a program recorded thereon for performing a method for generating a complex drought index based on artificial intelligence technology according to any one of paragraphs 1 to 5. Claim 7 delete

Images