Apparatus for predicting water demand based on artificial intelligence and method therefor

Abstract

A method for predicting water demand based on artificial intelligence comprises: a step in which a data processing unit prepares first input data, which is data for a first period from a reference point to a predetermined past point in time, and second input data, which is data for a second period from the reference point to a predetermined future point in time; a step in which a prediction unit analyzes the first input data through a first prediction model to derive first prediction data that predicts the solar power generation amount for the second period; a step in which the prediction unit analyzes the second input data through a second prediction model to derive second prediction data that predicts the solar power generation amount for the second period; and a step in which the prediction unit analyzes the first prediction data and the second prediction data through a third prediction model to derive third prediction data that predicts the solar power generation amount for the second period.

Description

Technology Field

[0001] The present invention relates to a technology for predicting water demand, and more specifically, to an apparatus for predicting water demand based on artificial intelligence and a method for the same. Background Technology

[0002] Water supply facilities are utilized by everyone living in the city and are vital infrastructure that sustains daily life and various urban activities; therefore, high-quality water must be stably provided to match the city's population concentration, improved living standards, and economic growth. To maintain this, water management operations—a comprehensive maintenance plan for the water supply based on drainage control—remain a critical task.

[0003] The system supporting water management operations is designed to efficiently supply safe water in terms of quantity, pressure, and quality by controlling and monitoring all water supply facilities from the water purification plant to the consumer; therefore, accurate forecasting of water demand on a daily or hourly basis is required. Prior art literature

[0004] Korean Patent Publication No. 2021-0049006 (Published May 4, 2021) The problem to be solved

[0005] The objective of the present invention is to provide an apparatus for predicting water demand based on artificial intelligence and a method for doing so. means of solving the problem

[0006] A method for predicting water demand according to a preferred embodiment of the present invention for achieving the above-described purpose comprises: a step in which a data processing unit classifies a plurality of consumers into a plurality of consumer groups based on the similarity of water usage data of a plurality of consumers; a step in which a model generation unit prepares water usage data of a plurality of consumers belonging to a consumer group as training data for each of the plurality of consumer groups; a step in which the model generation unit generates a plurality of prediction models corresponding to the plurality of consumer groups using the training data corresponding to the plurality of consumer groups; and a step in which a demand forecasting unit predicts the water usage of consumers using the plurality of prediction models.

[0007] The step of classifying into the above plurality of consumer groups further includes the step of the data processing unit deriving the cluster center of the plurality of consumer groups and the dynamic time warping similarity of each consumer as cluster similarity weights corresponding to the plurality of consumer groups.

[0008] The step of predicting the water usage of the above-mentioned consumers includes: a step of producing prediction data that predicts the water usage of consumers in a second period, which is a period from a reference point to a predetermined future point in time, through calculations on the water usage data of consumers in a first period, which is a period from a reference point to a predetermined past point in time, by a plurality of prediction models; a step in which the demand forecasting unit applies cluster similarity weights corresponding to the consumer group of the above-mentioned consumers to the prediction data produced by the prediction model corresponding to the consumer group to derive a weighted sum of the prediction data of the plurality of prediction models; and a step in which the demand forecasting unit predicts the water usage of consumers in the second period according to the weighted sum.

[0009] The step of deriving the cluster similarity weights described above is performed by the data processing unit

[0010] mathematical formula

[0011]

[0012] A cluster similarity weight is calculated according to the above, wherein W is a cluster similarity weight, X is a consumer, Y is a cluster center of a consumer group, DTW is a dynamic time warping similarity, f(X) is a consumer group to which consumer X belongs, and y is an index of a consumer group.

[0013] The step of classifying the plurality of consumers into a plurality of consumer groups includes the step of the data processing unit deriving a dynamic time warping similarity representing the similarity of water usage data between the plurality of consumers based on the warping path cost for water usage data between the plurality of consumers, and the step of the data processing unit classifying the plurality of consumers into a plurality of consumer groups according to the derived dynamic time warping similarity.

[0014] The step of deriving dynamic time warping similarity includes the step of the data processing unit preparing water usage data of two different consumers; the step of the data processing unit arranging the water usage data in rows and columns to form a two-dimensional matrix in which the elements of the matrix consist of warping path costs; the step of the data processing unit deriving a path in which the sum of accumulated warping path costs on the matrix is ​​minimized; and the step of the data processing unit deriving dynamic time warping similarity by taking the square root of the warping path cost of the last element of the path in which the sum of warping path costs is minimized.

[0015] The step of forming the above matrix is ​​performed by the data processing unit

[0016] mathematical formula

[0017]

[0018] The warping path cost is calculated according to the above, wherein D is the warping path cost, x and y are water usage data of two different consumers, and i and j are time series indices of the water usage data.

[0019] The step of classifying into the plurality of consumer groups includes: an initialization step in which the data processing unit randomly sets a predetermined number of consumers among the plurality of consumers as the centers of an initial cluster; a classification step in which the data processing unit classifies the plurality of consumers into consumer groups having cluster centers where the dynamic time warping similarity between the consumers and the cluster centers is minimized for the plurality of consumers; a reselection step in which the data processing unit reselects the cluster centers to the consumers within each consumer group where the sum of squares of dynamic time warping similarity is minimized; a repetition step in which the data processing unit classifies the plurality of consumers into the plurality of consumer groups by repeating the classification step and the reselection step until the cluster centers are not changed according to the reselection; and an overall repetition step in which the data processing unit repeats the initialization step, the classification step, the resetting step, and the repetition step while resetting different predetermined numbers of consumers as the centers of the initial cluster so that the total sum of squares of dynamic time warping similarity between consumers belonging to the plurality of consumer groups is minimized.

[0020] An apparatus for predicting water demand according to a preferred embodiment of the present invention for achieving the purpose described above comprises: a data processing unit that classifies a plurality of consumers into a plurality of consumer groups based on the similarity of water usage data of a plurality of consumers; a model generation unit that prepares water usage data of a plurality of consumers belonging to a consumer group for each of the plurality of consumer groups as training data and generates a plurality of prediction models corresponding to the plurality of consumer groups using the training data corresponding to the plurality of consumer groups; and a demand prediction unit that predicts the water usage of consumers using the plurality of prediction models.

[0021] The above data processing unit is characterized by deriving the cluster center of a plurality of consumer groups and the dynamic time warping similarity of each consumer as cluster similarity weights corresponding to the plurality of consumer groups.

[0022] When a plurality of prediction models produce prediction data that predicts the water usage of consumers in a second period, which is a period from a reference point to a predetermined future point in time, through calculations on the water usage data of consumers in a first period, which is a period from a reference point to a predetermined past point in time, the demand forecasting unit applies cluster similarity weights corresponding to the consumer group of the consumers to the prediction data produced by the prediction model corresponding to the consumer group to derive a weighted sum of the prediction data of the plurality of prediction models, and predicts the water usage of consumers in the second period according to the weighted sum.

[0023] The above data processing unit is a mathematical formula

[0024]

[0025] A cluster similarity weight is calculated according to the above, wherein W is a cluster similarity weight, X is a consumer, Y is a cluster center of a consumer group, DTW is a dynamic time warping similarity, f(X) is a consumer group to which consumer X belongs, and y is an index of a consumer group.

[0026] The data processing unit above is characterized by deriving dynamic time warping similarity representing the similarity of water usage data between multiple consumers based on warping path costs for water usage data between multiple consumers, and classifying multiple consumers into multiple consumer groups according to the derived dynamic time warping similarity.

[0027] The data processing unit prepares water usage data of two different consumers, arranges the water usage data in rows and columns respectively to form a two-dimensional matrix in which the elements of the matrix are warping path costs, derives a path in which the sum of accumulated warping path costs is minimized on the matrix, and derives dynamic time warping similarity by taking the square root of the warping path cost of the last element of the path in which the sum of warping path costs is minimized.

[0028] The above data processing unit is a mathematical formula

[0029]

[0030] The warping path cost is calculated according to the above, wherein D is the warping path cost, x and y are water usage data of two different consumers, and i and j are time series indices of the water usage data.

[0031] The data processing unit is characterized by performing an initialization process in which a predetermined number of consumers are randomly set as the centers of an initial cluster among a plurality of consumers; a classification process in which the plurality of consumers are classified into consumer groups having cluster centers where the dynamic time warping similarity between the consumers and the cluster centers is minimized for the plurality of consumers; a reselection process in which the data processing unit reselects the cluster centers to the consumers within each consumer group where the sum of squares of dynamic time warping similarity is minimized; a repetition process in which the plurality of consumers are classified into multiple consumer groups by repeating the classification step and the reselection step until the cluster centers are not changed according to the reselection; and an overall repetition process in which the initialization step, the classification step, the resetting step, and the repetition step are repeated while resetting different predetermined numbers of consumers as the centers of the initial cluster so that the total sum of the sums of squares of dynamic time warping similarity between consumers belonging to the multiple consumer groups is minimized. Effects of the invention

[0032] According to the present invention, patterns of water usage data from a plurality of consumers are classified by consumer group, and a plurality of prediction models are generated accordingly. Subsequently, water usage is predicted using a weighted sum in which cluster similarity weights are applied based on the similarity of the water usage data patterns to the prediction results using the plurality of prediction models. In other words, by generating prediction models using abundant data, the reliability of the water usage prediction results is not only high, but water usage can also be predicted precisely by considering the data characteristics of individual consumers through cluster similarity weights. Brief explanation of the drawing

[0033] FIG. 1 is a diagram illustrating the configuration of a system for predicting water demand based on artificial intelligence according to an embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of a device for predicting water demand based on artificial intelligence according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating a method for predicting water demand based on artificial intelligence according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating a method for deriving dynamic time warping similarity between a plurality of consumers according to an embodiment of the present invention. FIG. 5 is a diagram illustrating a matrix representing warping path costs according to an embodiment of the present invention. FIG. 6 is a diagram illustrating the dynamic time warping similarity between consumers according to an embodiment of the present invention. FIG. 7 is a flowchart illustrating a method for classifying a plurality of consumers into a plurality of consumer groups based on dynamic time warping similarity according to an embodiment of the present invention. FIG. 8 is a flowchart illustrating a method for generating a prediction model according to an embodiment of the present invention. FIG. 9 is a flowchart illustrating a method for predicting water usage of a consumer using a prediction model according to an embodiment of the present invention. FIG. 10 is an illustrative diagram for explaining a method of predicting water usage of a consumer using a prediction model according to an embodiment of the present invention. FIG. 11 is a drawing showing a computing device according to an embodiment of the present invention. Specific details for implementing the invention

[0034] Before the detailed description of the present invention, it should be understood that the embodiments described in this specification and the configurations illustrated in the drawings are merely the most preferred embodiments of the present invention and do not represent all of the technical ideas of the present invention, and that various equivalents and modifications that can replace them may exist at the time of filing this application.

[0035] Preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that identical components in the accompanying drawings are indicated by the same reference numerals whenever possible. Furthermore, detailed descriptions of known functions and configurations that could obscure the essence of the invention will be omitted. For the same reason, some components in the accompanying drawings may be exaggerated, omitted, or schematically depicted, and the size of each component does not entirely reflect its actual size.

[0036] In addition, the terms and words used in the specification and claims described below should not be interpreted as being limited to their ordinary or dictionary meanings, but should be interpreted in a meaning and concept consistent with the technical spirit of the invention, based on the principle that the inventor can appropriately define the concept of the terms to best describe his invention.

[0037] In particular, it should be noted that in the embodiments of the present invention, terms such as "first," "second," and "third" are used to indicate that they are different objects.

[0038] First, the configuration of a system for predicting water demand based on artificial intelligence according to an embodiment of the present invention will be described. FIG. 1 is a diagram illustrating the configuration of a system for predicting water demand based on artificial intelligence according to an embodiment of the present invention. FIG. 2 is a diagram illustrating the configuration of a device for predicting water demand based on artificial intelligence according to an embodiment of the present invention. Referring to FIG. 1, the prediction system includes a prediction device (10), a plurality of smart meters (20), a plurality of consumers (30), and a weather server (40).

[0039] The prediction device (10) is intended to collect water usage data of a consumer (30) through a smart meter (20), which is an IoT (Internet of Things) device, and to predict the consumer's water usage for a second period by analyzing the collected water usage data for a first period using multiple prediction models. The first period refers to a period from a reference point to a predetermined past point in time. The second period refers to a period from a reference point to a predetermined future point in time. Here, the reference point may be the point in time (present) when a process for predicting water demand using a prediction model is performed. The weather server (40) is a server operated by the Korea Meteorological Administration, and the prediction device (10) can receive weather observation data such as temperature, humidity, and precipitation, weather forecast data, numerical weather prediction model data, etc., from the weather server (40) and use them for prediction. Specifically, the weather observation data is data provided by the Korea Meteorological Administration and may include weather observation data observed through equipment such as ASOS (Automated Surface Observing System) and AWS (Automatic Weather Station). Weather forecast data refers to data provided by the Korea Meteorological Administration and may include weather forecast data that predicts weather conditions. Numerical weather prediction model data may include data such as temperature, solar radiation, and cloud cover predicted through numerical weather prediction models, such as the Global Weather Forecasting Model (GDAPS) and Local Weather Forecasting Model (LDAPS).

[0040] The prediction model is a learning model (Deep Learning Model) and can have structures such as a Recurrent Neural Network (RNN), Transformer, or Linear. The prediction model includes multiple interconnected layers (or modules), and these layers (or modules) consist of multiple operations. Additionally, these layers (or modules) are connected by weights. That is, the output resulting from the operation of a single layer (or module) is input to the next layer with weights applied. The first, second, and third prediction models derive an output by performing multiple operations on the input in which weights are applied between the multiple layers (or modules). In other words, they perform multiple operations connected by weights between the multiple layers (or modules). These operations of the prediction model are referred to as 'weighted operations'. Specifically, multiple layers (or modules) may include a Convolution Layer (CL), a Deconvolution Layer (DL), a Pooling Layer (PL), a Fully-Connected Layer (FL), an RNN memory cell, etc. The Convolution Layer (CL) performs convolution operations and operations based on an activation function. Additionally, the Deconvolution Layer (DL) performs deconvolution operations and operations based on an activation function. The Pooling Layer (PL) performs downsampling or upsampling operations. The Fully-Connected Layer (FL) performs operations based on an activation function on multiple input values ​​from the previous layer. The RNN memory cell performs operations based on an activation function. Here, examples of activation functions include sigmoid, hyperbolic tangent (tanh), exponential linear unit (ELU), rectified linear unit (ReLU), leaky reLU, maxout, minout, and softmax.

[0041] Referring to FIG. 2, a prediction device (10) according to an embodiment of the present invention includes a data processing unit (100), a model generation unit (200), and a demand forecasting unit (300).

[0042] The data processing unit (100) is basically intended to collect water usage data of a consumer (30) through a smart meter (20). The data processing unit (100) may provide water usage data for a first period and a second period to the model generation unit (200) as training data. Alternatively, the data processing unit (100) may provide water usage data for the first period to the demand forecasting unit (300) for predicting water usage for the second period. Basically, the water usage data is time-series data and is based on values ​​of water usage measured over a certain period. The water usage data may be data obtained by extracting statistical features from values ​​of water usage measured over a certain period and normalizing them. According to one embodiment, the water usage data may be water usage statistics (average value, median value, etc.) for a predetermined unit period. For example, the water usage data may be water usage statistics for an hourly prediction model (0 to 23 hours). For example, water usage data may include time-lag values ​​of water usage, moving average values, hourly water usage statistics, etc. As another example, water usage data may be for a daily prediction model and may be daily (1 to 31 days) or weekly (Sun to Sat) water usage statistics. For example, water usage data may include daily water usage sum values, daily maximum / minimum temperature values, nighttime maximum / minimum temperature values, and daily / weekly water usage statistics. Additionally, the data processing unit (100) may classify multiple consumers into multiple consumer groups based on the similarity of water usage data of multiple consumers. Furthermore, the data processing unit (100) may receive weather observation data, weather forecast data, numerical weather prediction model data, etc. for a first period from the weather server (40).

[0043] The model generation unit (200) is intended to generate a prediction model, which is a learning model (Deep Learning Model) according to an embodiment of the present invention, through learning. The model generation unit (200) prepares water usage data of multiple consumers belonging to a group of consumers as learning data for each group of consumers through the data processing unit (100), and can generate multiple prediction models corresponding to the groups of consumers using the learning data corresponding to the groups of consumers.

[0044] The demand forecasting unit (300) is intended to predict water demand using multiple forecasting models. When the demand forecasting unit (300) inputs water usage data of consumers in a first period, which is a period from a reference point to a predetermined past point in time, into multiple forecasting models, the multiple forecasting models can produce forecasting data that predicts the water usage of consumers in a second period, which is a period from a reference point to a predetermined future point in time, through calculations on the water usage data of consumers in the first period, that is, weight calculations. Then, the demand forecasting unit (300) applies cluster similarity weights corresponding to the consumer group of consumers to the forecasting data produced by the forecasting models corresponding to the consumer group to derive a weighted sum of the forecasting data of the multiple forecasting models, and can predict the water usage of consumers in the second period according to the weighted sum.

[0045] The specific operation of the forecasting device (10), which includes the aforementioned data processing unit (100), model generation unit (200), and demand forecasting unit (300), will be described in more detail below.

[0046] Next, a method for predicting water demand based on artificial intelligence according to an embodiment of the present invention will be described. FIG. 3 is a flowchart illustrating a method for predicting water demand based on artificial intelligence according to an embodiment of the present invention.

[0047] The data processing unit (100) can collect water usage data for multiple consumers (30) for a certain period or longer through multiple smart meters (20) in step S110. Basically, the water usage data is time-series data and is based on values ​​of water usage measured for a certain period or longer. The water usage data may be data obtained by extracting statistical features from values ​​of water usage measured for a certain period or longer and normalizing them. According to one embodiment, the water usage data may be water usage statistics (average value, median value, etc.) for a predetermined unit period. For example, the water usage data may be water usage statistics by hour (0 to 23 hours) for an hourly prediction model. As another example, the water usage data may be water usage statistics by day (1 to 31 days) or by day of the week (Sun to Sat) for a daily prediction model.

[0048] When water usage data of multiple consumers (30) is collected, the data processing unit (100) classifies the multiple consumers (30) into multiple consumer groups based on the similarity of the water usage data of the multiple consumers (30). To this end, in step S120, the data processing unit (100) derives dynamic time warping similarity, which represents the similarity of the water usage data between the multiple consumers (30), based on the warping path cost for the water usage data between the multiple consumers (30). Then, in step S130, the data processing unit (100) classifies the multiple consumers (30) into multiple consumer groups according to the previously derived dynamic time warping similarity. In this step S130, the data processing unit (100) classifies the multiple consumers (30) into multiple consumer groups, each having a clustering center, through clustering. Accordingly, the data processing unit (100) can derive the cluster center of the multiple consumer groups and the dynamic time warping similarity of each consumer as cluster similarity weights corresponding to the multiple consumer groups.

[0049] When multiple consumers (30) are classified into multiple consumer groups, the model generation unit (200) prepares water usage data of multiple consumers (30) belonging to each consumer group as training data in step S140. For example, assuming there are two consumer groups, water usage data of multiple consumers (30) belonging to the first consumer group can be prepared as first training data, and water usage data of multiple consumers (30) belonging to the second consumer group can be prepared as second training data.

[0050] The model generation unit (200) generates multiple prediction models corresponding to multiple consumer groups using training data corresponding to multiple consumer groups in step S150. For example, assuming there are two consumer groups, a first prediction model can be generated using first training data corresponding to the first consumer group, and a second prediction model can be generated using second training data corresponding to the second consumer group.

[0051] The demand forecasting unit (300) predicts the water usage of the consumer using a plurality of prediction models generated earlier (S150) in step S160. At this time, the demand forecasting unit (300) predicts the water usage of the consumer through a weighted sum with cluster similarity weights applied.

[0052] Then, steps S120 through S160 described above will be explained in more detail. First, a method for deriving dynamic time warping similarity, which represents the similarity of water usage data between multiple consumers (30), will be explained. FIG. 4 is a flowchart for explaining a method for deriving dynamic time warping similarity between multiple consumers according to an embodiment of the present invention. FIG. 5 is a diagram for explaining a matrix representing warping path costs according to an embodiment of the present invention. FIG. 6 is a diagram for explaining dynamic time warping similarity between consumers according to an embodiment of the present invention. To emphasize, FIG. 4 through 6 are for a detailed explanation of step S120.

[0053] Referring to FIG. 4, the data processing unit (100) prepares water usage data for two different consumers in step S210. As previously described, the water usage data is time-series data, and the lengths of the time-series data may not be the same. For example, referring to FIG. 5, it is assumed that the water usage data of the first consumer (A) is {1, 2, 4, 5, 3, 1} and the water usage data of the second consumer (B) is {2, 4, 5, 3, 2, 1}.

[0054] In step S220, the data processing unit (100) arranges the water usage data in rows and columns to form a two-dimensional matrix in which the elements of the matrix consist of warping path costs. An example of such a matrix is ​​illustrated in FIG. 5. In step S220, the data processing unit (100) derives the warping path costs according to the following mathematical formula 1.

[0055]

[0056] Here, D represents the warping path cost. Also, x and y are water usage data of two different consumers, and i and j are time series indices of the water usage data.

[0057] The data processing unit (100) derives the path in which the sum of the accumulated warping path costs is minimized in the matrix, as shown in the path (indicated by hatching) of FIG. 5, at step S230. Then, at step S240, the data processing unit (100) derives dynamic time warping similarity, which represents the similarity between the water usage data of two consumers, by taking the square root of the warping path cost of the last element of the path in which the sum of the warping path costs is minimized.

[0058] The data processing unit (100) can derive dynamic time warping similarity between multiple consumers by applying the method of steps S210 to S240 described above to all consumers. For example, assuming that all consumers are a first consumer (A), a second consumer (B), a third consumer (C), and a fourth consumer (D), dynamic time warping similarity between multiple consumers can be derived as shown in FIG. 6. According to the example in FIG. 6, it can be seen that the dynamic time warping similarity (DTW) between the first consumer (A) and the second consumer (B) is 1, the dynamic time warping similarity (DTW) between the first consumer (A) and the third consumer (C) is 9, and the dynamic time warping similarity (DTW) between the first consumer (A) and the fourth consumer (D) is 10.

[0059] Then, a method for classifying multiple consumers into multiple consumer groups based on dynamic time warping similarity between multiple consumers as shown in FIG. 6 will be described. FIG. 7 is a flowchart for explaining a method for classifying multiple consumers into multiple consumer groups based on dynamic time warping similarity according to an embodiment of the present invention. To emphasize, FIG. 7 is for a detailed explanation of step S130.

[0060] Referring to FIGS. 6 and FIGS. 7, the data processing unit (100) sets a predetermined number (K) of consumers randomly among a plurality of consumers as the centers of the initial cluster in step S310. For example, referring to FIG. 6, it is assumed that K=2 and the first consumer (A) and the second consumer (B) are set as the centers of the initial cluster.

[0061] Next, the data processing unit (100) classifies the plurality of consumers into groups of consumers having a cluster center with the minimum dynamic time warping similarity between the consumer and the cluster center for the plurality of consumers in step S320. For example, referring to FIG. 6, the first consumer group may be the first consumer (A) and the second consumer group may be the second to fourth consumers (B, C, D) according to the dynamic time warping similarity.

[0062] Next, in step S330, the data processing unit (100) reselects the cluster center for each consumer group as the consumer with the minimum sum of squares of dynamic time warping similarity within the consumer group. For example, referring to FIG. 6, the cluster center is not changed because there is only one consumer in the first consumer group. In the second consumer group, the sum of squares of dynamic time warping similarity of the second consumer (B) is And, the sum of squares of the dynamic time warping similarity of the third consumer (C) is And, the sum of squares of the dynamic time warping similarity of the fourth consumer (D) is This is the case. Accordingly, the center of the cluster may be changed from the second consumer (B) to the third consumer (C).

[0063] The data processing unit (100) determines whether the cluster center changes due to the re-selection in the preceding step (S330) at step S340. Accordingly, if the cluster center changes, the aforementioned steps S320 through S340 are repeated. On the other hand, if the cluster center does not change, the process proceeds to step S350.

[0064] For example, referring to FIG. 6, if the cluster center is changed to the first consumer (A) and the third consumer (C) according to reselection, repeating step S320 changes the first consumer group to the first consumer (A) and the second consumer (B), and the second consumer group can become the third consumer (C) and the fourth consumer (D). In this state, even if steps S330 and S340 are repeated, the consumer with the minimum sum of squares of dynamic time warping similarity is not changed, so proceed to step S350.

[0065] The data processing unit (100) calculates the total sum of squared dynamic time warping similarities between consumers belonging to a plurality of consumer groups in step S350. For example, referring to FIG. 6, when the first consumer group is the first consumer (A) and the second consumer (B), and the second consumer group is the third consumer (C) and the fourth consumer (D), the total sum of squared dynamic time warping similarities is 1 for the first consumer group, 4 for the second consumer group, and the total sum of squared dynamic time warping similarities is 5.

[0066] Next, the data processing unit (100) determines in step S360 whether the total sum of squared dynamic time warping similarities between consumers belonging to a plurality of consumer groups is at a minimum. If the result of the determination is that the total sum of squared dynamic time warping similarities is not at a minimum, the data processing unit (100) proceeds to step S370 to reset the center of the initial cluster and repeats the aforementioned steps S320 through S360.

[0067] On the other hand, if the total sum of the squares of the dynamic time warping similarity is at its minimum, the data processing unit (100) proceeds to step S380 to complete the clustering. In this way, the consumer group at the time the clustering is completed becomes the final consumer group.

[0068] Next, in step S390, the data processing unit (100) calculates the cluster center of a plurality of consumer groups and the dynamic time warping similarity of each consumer as a cluster similarity weight corresponding to the plurality of consumer groups. At this time, the data processing unit (100) can calculate the cluster similarity weight according to the following mathematical formula 2.

[0069]

[0070] Here, W represents the cluster similarity weight. X represents a consumer, and Y represents the cluster center of consumer group Y. Accordingly, W(X, Y) represents the cluster similarity weight of consumer X for consumer group Y.

[0071] DTW represents dynamic time warping similarity. f(X) represents the consumer group to which consumer X belongs. y is the index of the consumer group. represents the dynamic time-warping similarity between consumer X and the cluster center of the consumer group to which consumer X belongs. represents the dynamic time-warping similarity of consumer X with the cluster center of consumer group Y. And, represents the dynamic time-warping similarity between consumer X and the cluster center of consumer group y.

[0072] Next, the aforementioned steps S140 and S150 will be explained in more detail. FIG. 8 is a flowchart illustrating a method for generating a prediction model according to an embodiment of the present invention. To emphasize, FIG. 8 is intended for a detailed explanation of steps S140 and S150. As previously described, the model generation unit (200) of the present invention, assuming, for example, that there are two consumer groups, prepares water usage data of a plurality of consumers (30) belonging to a first consumer group as first training data, and prepares water usage data of a plurality of consumers (30) belonging to a second consumer group as second training data. Then, it generates a first prediction model using the first training data corresponding to the first consumer group, and generates a second prediction model using the second training data corresponding to the second consumer group. That is, a prediction model is generated for each consumer group. In the case of FIG. 8, a method for generating a prediction model corresponding to any one consumer group is explained.

[0073] Referring to FIG. 8, the model generation unit (200) prepares training data corresponding to a consumer group in step S410. The training data includes water usage data for a first period, which is the training input data, and water usage data for a second period corresponding to the first period, which is the label (target value) corresponding to the training input data. In an embodiment of the present invention, the first period refers to a period from a reference point to a predetermined past point in time. The second period refers to a period from a reference point to a predetermined future point in time. According to an additional embodiment, the training input data may further include weather observation data, weather forecast data, numerical weather prediction model data, etc., for the first period received from a weather server (40).

[0074] Then, in step S420, the model generation unit (200) generates multiple prediction models having different structures through machine learning / deep learning using training data. These prediction models receive water usage data representing the water usage of the first period and are trained to produce prediction data representing the water usage of the second period through operations on the received water usage data of the first period. For example, the multiple prediction models may have structures such as a Recurrent Neural Network (RNN), a Transformer, or a Linear.

[0075] Next, in step S430, the model generation unit (200) selects the prediction model with the highest evaluation value based on the evaluation indicator among the multiple prediction models generated earlier (S420) as the prediction model corresponding to the consumer group. At this time, the evaluation indicator may be MSE (Mean Squared Error), RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), MAPE (Mean Absolute Percentage Error), etc.

[0076] Next, the method for predicting water usage in the aforementioned step S160 will be explained in more detail. FIG. 9 is a flowchart illustrating a method for predicting water usage of a consumer using a prediction model according to an embodiment of the present invention. FIG. 10 is an illustrative diagram illustrating a method for predicting water usage of a consumer using a prediction model according to an embodiment of the present invention. To emphasize, FIG. 9 and FIG. 10 are for a detailed explanation of step S160.

[0077] Referring to FIG. 9, the data processing unit (100) collects water usage data for a first period of a consumer in step S510. At this time, the data processing unit (100) can collect water usage data for a plurality of consumers (30) of the first period through a plurality of smart meters (20). Here, the first period refers to a period from a reference point in time to a predetermined past point in time. Here, the reference point in time may be the point in time (present) when a process for predicting water demand using a prediction model is performed. Basically, water usage data is time-series data and is based on values ​​of water usage measured over a certain period. Water usage data may be data obtained by extracting statistical features from values ​​of water usage measured over a certain period and normalizing them. According to one embodiment, water usage data may be water usage statistics (average value, median value, etc.) for a predetermined unit period. According to an additional embodiment, the data processing unit (100) can collect at least one of weather observation data, weather forecast data, and numerical weather prediction model data for the first period from the weather server (40) in response to water usage data for the first period.

[0078] In step S510, the data processing unit (100) inputs the water usage data (IN) of the first period of the consumer as input data to a plurality of prediction models corresponding to a plurality of consumer groups. For example, referring to FIG. 10, it is assumed that there are two plurality of prediction models. That is, it is assumed that the plurality of prediction models are a first prediction model and a second prediction model. The first prediction model is a prediction model corresponding to the first consumer group, and the second prediction model is a prediction model corresponding to the second consumer group. This means that the first prediction model is a prediction model learned using the water usage data of a plurality of consumers belonging to the first consumer group, and the second prediction model is a prediction model learned using the water usage data of a plurality of consumers belonging to the second consumer group. That is, the data processing unit (100) inputs the water usage data (IN) of the first period of the consumer to the first prediction model and the second prediction model. According to an additional embodiment, the input data may further include at least one of the weather observation data, weather forecast data, and numerical weather forecast model data of the first period received from the weather server (40).

[0079] When water usage data (IN) for the first period is input, multiple prediction models produce prediction data that predicts the water usage of a consumer in the second period through calculation (weighting calculation) on the consumer's water usage data for the first period in step S520. The second period refers to the period from a reference point in time to a predetermined future point in time. Here, the reference point in time may be the point in time (present) when a process for predicting water demand using the prediction models is performed. For example, referring to FIG. 10, the first prediction model derives first prediction data (Z1), and the second prediction model derives second prediction data (Z2).

[0080] Next, the demand forecasting unit (300) applies the cluster similarity weight corresponding to the consumer group of the consumer to the forecast data calculated by the forecast model corresponding to the consumer group in step S530, thereby deriving a weighted sum of the forecast data of multiple forecast models. For example, referring to FIG. 10, the demand forecasting unit (300) applies the first cluster similarity weight (W1) to the first forecast data (Z1) derived by the first forecast model, and applies the second cluster similarity weight (W2) to the second forecast data (Z2) derived by the second forecast model to derive a weighted sum (Z = W1 × Z1 + W2 × Z2). Here, the first cluster similarity weight (W1) is the cluster similarity weight for the first consumer group of the consumer, and the second cluster similarity weight (W2) is the cluster similarity weight for the second consumer group of the consumer. These cluster similarity weights are calculated through Equation 2 as described above (see S390).

[0081] Finally, the demand forecasting unit (300) predicts the water usage of the consumers in the second period according to the weighted sum in step S540.

[0083] FIG. 11 is a drawing showing a computing device according to an embodiment of the present invention. The computing device (TN100) of FIG. 11 may be a device described in the present specification, such as a prediction device (10).

[0084] In the embodiment of FIG. 11, the computing device (TN100) may include at least one processor (TN110), a transceiver (TN120), and a memory (TN130). Additionally, the computing device (TN100) may further include a storage device (TN140), an input interface device (TN150), an output interface device (TN160), etc. The components included in the computing device (TN100) may be connected by a bus (TN170) to communicate with each other.

[0085] The processor (TN110) can execute a program command stored in at least one of the memory (TN130) and the storage device (TN140). The processor (TN110) may refer to a central processing unit (CPU), a graphics processing unit (GPU), or a dedicated processor on which methods according to embodiments of the present invention are performed. The processor (TN110) may be configured to implement the procedures, functions, and methods described in relation to embodiments of the present invention. The processor (TN110) can control each component of the computing device (TN100).

[0086] Each of the memory (TN130) and the storage device (TN140) can store various information related to the operation of the processor (TN110). Each of the memory (TN130) and the storage device (TN140) may be composed of at least one of a volatile storage medium and a non-volatile storage medium. For example, the memory (TN130) may be composed of at least one of read-only memory (ROM) and random access memory (RAM).

[0087] The transmitting and receiving device (TN120) can transmit or receive wired or wireless signals. The transmitting and receiving device (TN120) can be connected to a network to perform communication.

[0089] Meanwhile, the method according to the embodiment of the present invention described above may be implemented in the form of a program readable through various computer means and recorded on a computer-readable recording medium. Here, the recording medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the recording medium may be those specifically designed and configured for the present invention, or those known and available to those skilled in the art of computer software. For example, the recording medium includes magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions may include machine language wires, such as those generated by a compiler, as well as high-level language wires that can be executed by a computer using an interpreter, etc. Such hardware devices may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

[0090] Although the present invention has been described above using several preferred embodiments, these embodiments are illustrative and not limiting. As such, those skilled in the art will understand that various changes and modifications can be made in accordance with the doctrine of equivalents without departing from the spirit of the invention and the scope of rights set forth in the appended claims. Explanation of the symbols

[0091] 10: Prediction device 20: Smart meter 30: Consumer 40: Weather Server 100: Data processing unit 200: Model Creation Section 300: Demand Forecasting Department

Claims

Claim 1 A method for predicting water demand comprises: a step in which a data processing unit classifies a plurality of consumers into a plurality of consumer groups based on the similarity of water usage data of a plurality of consumers; a step in which a model generation unit prepares water usage data of a plurality of consumers belonging to a consumer group as training data for each of the plurality of consumer groups; a step in which the model generation unit generates a plurality of prediction models corresponding to the plurality of consumer groups using the training data corresponding to the plurality of consumer groups; and a step in which a demand forecasting unit predicts the water usage of consumers using the plurality of prediction models; wherein the step of classifying into a plurality of consumer groups further comprises a step in which the data processing unit derives the cluster center of the plurality of consumer groups and the dynamic time warping similarity of each consumer as a cluster similarity weight corresponding to the plurality of consumer groups; and wherein the step of predicting the water usage of consumers comprises a step in which a plurality of prediction models produce prediction data that predicts the water usage of consumers in a second period, which is a period from a reference point to a predetermined future point, through calculations on the water usage data of consumers in a first period, which is a period from a reference point to a predetermined past point. A method for predicting water demand, characterized by comprising: a step in which the demand forecasting unit applies cluster similarity weights corresponding to the consumer group of the consumer to the prediction data calculated by the prediction model corresponding to the consumer group to derive a weighted sum of the prediction data of a plurality of prediction models; and a step in which the demand forecasting unit predicts the water usage of the consumer during the second period according to the weighted sum. Claim 2 delete Claim 3 delete Claim 4 In claim 1, the step of deriving the cluster similarity weight is performed by the data processing unit using a mathematical formula A method for predicting water demand, characterized by calculating cluster similarity weights according to, wherein W is a cluster similarity weight, X is a consumer, Y is a cluster center of a consumer group, DTW is a dynamic time warping similarity, f(X) is a consumer group to which consumer X belongs, and y is an index of a consumer group. Claim 5 A method for predicting water demand according to claim 1, wherein the step of classifying the plurality of consumers into a plurality of consumer groups comprises: a step in which the data processing unit derives a dynamic time warping similarity representing the similarity of water usage data between the plurality of consumers based on a warping path cost for water usage data between the plurality of consumers; and a step in which the data processing unit classifies the plurality of consumers into a plurality of consumer groups according to the derived dynamic time warping similarity. Claim 6 A method for predicting water demand according to claim 5, wherein the step of deriving dynamic time warping similarity comprises: a step in which the data processing unit prepares water usage data of two different consumers; a step in which the data processing unit arranges the water usage data in rows and columns, respectively, to form a two-dimensional matrix in which the elements of the matrix are warping path costs; a step in which the data processing unit derives a path in which the sum of accumulated warping path costs on the matrix is ​​minimized; and a step in which the data processing unit derives dynamic time warping similarity by taking the square root of the warping path cost of the last element of the path in which the sum of warping path costs is minimized. Claim 7 In paragraph 6, the step of forming the matrix is ​​the data processing unit using a mathematical formula A method for predicting water demand characterized by calculating a warping path cost according to, wherein D is a warping path cost, x and y are water usage data of two different consumers, and i and j are time series indices of water usage data. Claim 8 A method for predicting water demand according to claim 5, wherein the step of classifying into a plurality of consumer groups comprises: an initialization step in which the data processing unit randomly sets a predetermined number of consumers among the plurality of consumers as the centers of an initial cluster; a classification step in which the data processing unit classifies the plurality of consumers into consumer groups having cluster centers where the dynamic time warping similarity between the consumers and the cluster centers is minimized for the plurality of consumers; a reselection step in which the data processing unit reselects the cluster centers for each consumer group to the consumers within the consumer group where the sum of squares of dynamic time warping similarity is minimized; a repetition step in which the data processing unit classifies the plurality of consumers into a plurality of consumer groups by repeating the classification step and the reselection step until the cluster centers are not changed according to the reselection; and a total repetition step in which the data processing unit repeats the initialization step, the classification step, the resetting step, and the repetition step while resetting different predetermined numbers of consumers as the centers of the initial cluster so that the total sum of the sums of squares of dynamic time warping similarity between consumers belonging to the plurality of consumer groups is minimized. Claim 9 A device for predicting water demand comprises: a data processing unit that classifies multiple consumers into multiple consumer groups based on the similarity of water usage data of multiple consumers; a model generation unit that prepares water usage data of multiple consumers belonging to each consumer group as training data for each consumer group and generates multiple prediction models corresponding to the consumer groups using the training data corresponding to the consumer groups; and a demand forecasting unit that predicts the water usage of consumers using the multiple prediction models. An apparatus for predicting water demand, comprising: a data processing unit deriving a cluster similarity weight corresponding to the cluster center of a plurality of consumer groups and the dynamic time warping similarity of each consumer; wherein when the plurality of prediction models produce prediction data predicting the water usage of consumers in a second period, which is a period from a reference point to a predetermined future point in time, through calculations on the water usage data of consumers in a first period, which is a period from a reference point to a predetermined past point in time, the demand prediction unit applies the cluster similarity weight corresponding to the consumer groups of the consumers to the prediction data produced by the prediction models corresponding to the consumer groups to derive a weighted sum of the prediction data of the plurality of prediction models, and predicts the water usage of consumers in the second period according to the weighted sum. Claim 10 delete Claim 11 delete Claim 12 In Clause 9, the above data processing unit is a mathematical formula An apparatus for predicting water demand, characterized by calculating cluster similarity weights according to, wherein W is a cluster similarity weight, X is a consumer, Y is a cluster center of a consumer group, DTW is a dynamic time warping similarity, f(X) is a consumer group to which consumer X belongs, and y is an index of a consumer group. Claim 13 An apparatus for predicting water demand according to claim 9, wherein the data processing unit prepares water usage data of two different consumers, arranges the water usage data in rows and columns respectively to form a two-dimensional matrix in which the elements of the matrix consist of warping path costs, derives a path in which the sum of accumulated warping path costs on the matrix is ​​minimized, and derives dynamic time warping similarity by taking the square root of the warping path cost of the last element of the path in which the sum of warping path costs is minimized. Claim 14 In Clause 13, the above data processing unit is a mathematical formula An apparatus for predicting water demand, characterized by calculating a warping path cost according to, wherein D is a warping path cost, x and y are water usage data of two different consumers, and i and j are time series indices of water usage data. Claim 15 An apparatus for predicting water demand according to claim 9, wherein the data processing unit performs: an initialization process in which a predetermined number of consumers are randomly set as the centers of an initial cluster among a plurality of consumers; a classification process in which the plurality of consumers are classified into consumer groups having cluster centers where the dynamic time warping similarity between the consumers and the cluster centers is minimized; a reselection process in which the data processing unit reselects the cluster centers for each consumer group to the consumers with the minimum sum of squares of dynamic time warping similarity within the consumer group; a repetition process in which the plurality of consumers are classified into multiple consumer groups by repeating the classification step and the reselection step until the cluster centers are not changed according to the reselection; and an overall repetition process in which the initialization step, the classification step, the resetting step, and the repetition step are repeated while resetting different predetermined numbers of consumers as the centers of the initial cluster so that the total sum of squares of dynamic time warping similarity between consumers belonging to the multiple consumer groups is minimized.

Images