KR20220076869A - Forecasting system and method for power consumption - Google Patents

Abstract

The present invention relates to a system and method capable of predicting future power usage by converting load data on the past power usage to load variation data and simplifying it through coding.
The present invention includes a load collecting unit for obtaining load data for each time period; a code conversion unit converting load data into load variation data and obtaining an integer corresponding to the load variation data; a pattern classification unit for collecting load patterns having the same coded integer pattern and extracting a representative pattern from the same integer pattern; A power usage prediction system including a prediction unit for predicting future power usage from the representative pattern coded through the pattern classification unit and the actual load pattern may be provided.
In addition, the present invention includes the steps of obtaining load data, which is basic data, and converting it into load variation data; A method of predicting power usage may be provided, including the step of simplifying the load variation data by coding it into an integer.

Description

Power usage prediction system and method

The present invention relates to a system and method capable of predicting future power usage by converting load data about the past power usage into load variation data and simplifying it through coding.

Recently, as issues such as de-nuclear power generation, fine dust, and abnormal climate have emerged in the power industry, DR (Demand Response), renewable energy, ESS (Energy Storage System), EMS (Energy Management System; energy management) The importance of clean energy and energy efficiency such as equipment) is increasing day by day.

Energy solution-related research such as DR capacity calculation, ESS optimal operation plan, and renewable energy generation prediction is actively being conducted. Among them, load prediction is an important part of energy efficiency.

In particular, in the case of an ESS for a factory load, since the charging/discharging time varies according to the prediction result of the load pattern, the accuracy of the prediction is directly related to the optimization of the optimal operation algorithm for the energy storage device.

The present invention converts the load change data from the load data on the past power usage to the load change data, simplifies it into a code, extracts a representative pattern, and learns this past power usage and each code corresponding to it in machine learning, and the machine learned in this way The goal is to provide a system and method that can input a real-time load pattern and code of the load pattern into a learning model, and predict future power usage based on the code and power usage up to the present time.

The solution means of the present invention is a load collecting unit for obtaining load data for each time period; a code conversion unit converting load data into load variation data and obtaining an integer corresponding to the load variation data; a pattern classification unit for collecting load patterns having the same coded integer pattern and extracting a representative pattern from the same integer pattern; A power usage prediction system including a prediction unit for predicting future power usage from the representative pattern coded through the pattern classification unit and the actual load pattern may be provided.

The present invention includes the steps of: obtaining load data, which is basic data, and converting it into load variation data; A method of predicting power usage may be provided, including the step of simplifying the load variation data by coding it into an integer.

As described above, the present invention requires a process of classifying the load pattern before learning in order to improve the prediction reliability and accuracy of the load pattern through machine learning learning. In particular, in the case of a complex and irregular load pattern, the result value can be improved when learning is performed by classifying it according to a certain standard. In other words, the pattern classification before learning has symptoms of overfitting, which is poor in generalization performance, and underfitting, in which the artificial intelligence (AI) fits too much to the training data, and the actual prediction performance is poor because it fails to learn at an appropriate level. It is possible to improve the accuracy of prediction by improving the error rate for the load pattern prediction result.

In the present invention, even if the number of clusters of the initial load pattern is set, the error calculation unit and the determination unit determine the appropriate number of clusters to perform pattern reclassification, so that the user's experience dependence on initial value selection can be reduced.

In order to use the machine learning load prediction for the purpose of calculating the operation plan of the ESS, the determination of the charging/discharging timing of the ESS is the most important.

The present invention converts the load data of the factory into the amount of change in load by time period, codes the load change data to be simplified with integers, extracts a representative pattern by the simplified code, and a machine learning model through machine learning learning of the representative pattern and predict future power usage through these machine learning models.

The present invention may provide an improved pattern classification step suitable for calculating the ESS charge/discharge plan in order to learn the load prediction artificial intelligence. Since the pattern classification step is classified based on the amount of change in the load, it can be suitable as learning data for artificial intelligence that determines the charging/discharging timing of the ESS.

Also, when performing ESS peak reduction operation, the code indicates the increase or decrease of the load (load change amount). The results of the machine learning model can be easily utilized when determining the ESS operation schedule because it can be used at the time of the ESS charging decision, and the amount of charge and discharge at the time can be determined from the predicted power usage data.

1 is a block diagram of a system for predicting power usage according to the present invention.
2 is a flowchart illustrating a method of predicting power usage according to an embodiment of the present invention.
3 is a block diagram showing the process from the machine learning learning of the present invention to the future power amount prediction step.
Figure 4 (a) is a 24 hour load pattern graph for 365 days of the factory, (b) is a graph showing the load variation pattern.
5 is a graph illustrating loads classified by day and simplified patterns 1 and 2;
6 is a graph showing the daily load of the factory.
7 is a graph showing the daily load variation of the factory.
8 is a graph comparing the amount of change in load having the simplified reference value Step ₁ .
9 is a graph comparing the amount of change in load having the simplified reference value Step ₂ .
10 is a graph comparing the amount of change in load having the simplified reference value Step ₃ .
11 is a graph of pattern 1 classified by the present invention.
12 is a graph of pattern 2 classified by the present invention.
13 is a graph of pattern 3 classified by the present invention.
14 is a graph of pattern 4 classified by the present invention.
15 is a graph showing the load change of patterns 1 to 4 classified according to the present invention.

Hereinafter, specific contents for carrying out the present invention will be described in detail based on the accompanying exemplary drawings.

In describing the specific contents of the present invention, all of the patterns referred to as load patterns are omitted for convenience.

1 is a block diagram of a system for predicting power usage according to the present invention.

Referring to FIG. 1 , the power usage prediction system 10 of the present invention includes a load collection unit 100 , a code conversion unit 200 , a pattern classification unit 300 , an error calculation unit 400 , a determination unit 500 , It may include a machine learning learning unit 600 and a prediction unit 700 .

The load collection unit 100 may obtain load data (load amount), which is basic data.

When the code conversion unit 200 converts the load data into the load change data, first, the load data is divided and classified by time period, and the load change amount for each time period may be calculated.

In addition, the code conversion unit 200 may simplify the preparation of an integer pattern by encoding the load variation data into integers to distinguish the daily load data for each pattern. Here, the integer may consist of, for example, -2α to 2α.

The code conversion unit 200 converts the raw data into load change amount (ΔLoad _t,n ) data for the increase/decrease in order to simplify the 24-hour load into an hourly integer, and then converts the raw data into an integer value according to the change amount (Code _t,n ) can be replaced with

In order to calculate the integer value (Code _t,n ), the amount of change must be divided into 2α equal parts, and at this time, the simplified reference value (Step _n ) can be obtained by Equation 1 described in detail below, and divided into 2α equal parts. It can be the size of the equal value when

Here, α may be a variable that determines the number of equal parts for the simplification of the load pattern.

In this case, the simplified load pattern coded as an integer may be newly calculated when the load variation (ΔLoad _t,n ) exceeds the simplification reference value (Step _n ) as shown in Equation (2).

The pattern classification unit 300 obtains an integer value (Code _t,n ) for each time period for all dates n by the code conversion unit 200, and then a pattern having the same integer value (Code _t,n ) for 24 hours can be gathered And, it is possible to extract a plurality of representative patterns by the same collected pattern.

The error calculating unit 400 calculates the error rate of the representative pattern and individual load patterns calculated as the average value of the clustered load patterns in order to determine the similarity of how similar the load patterns classified through the pattern classification unit 300 are. can

The error rate can be calculated using the average absolute ratio error rate (MAPE _error ) calculation formula based on Equations 4 and 5.

The determination unit 500 calculates the error rate of the load pattern classified through the error calculating unit 400, and then determines whether the extracted representative pattern and the individual load pattern satisfy a predetermined error range, and if not satisfied, a simplified reference value (Step _α ) can be calculated again.

For example, assuming that the verification reference value (predetermined error rate) is 50%, if the average absolute percentage error rate (MAPE _error ) is greater than 50%, it is determined that the clusters are not clustered in the same pattern, and the standard value for simplifying the data classification process (Step _α ) By increasing α by 1, the pattern classification process may be repeated through the pattern classification unit 300 .

When the average absolute ratio error rate (MAPE _error ) is less than 50%, it is determined that the pattern classification is successful, and the average value for each classification can be output as a representative pattern.

In addition, since the simplified reference value Step _α is determined according to the variable α, the number of clusters may change.

The error calculator 400 is not limited to the average absolute ratio error rate (MAPE _error ), and may be calculated using various other error formulas.

Referring to FIG. 3 , the code conversion unit 200 converts the load variation data based on the load data, simplifies it through coding, and makes an integer pattern, and the integer pattern coded through the pattern classifying unit 300 is the same load. After collecting and extracting a representative pattern, if a predetermined error rate reference value is satisfied through error rate verification, it is set as a representative pattern, and the machine learning learning unit 600 can learn by receiving information about the representative pattern.

The prediction unit 700 can predict the future power usage from the representative pattern and the actual load pattern coded through the pattern classification unit 300, and at this time, the pattern through the real-time load pattern by the learning of the machine learning learning unit 600 You can simplify, set up a machine learning model, and then predict future wattage.

2 is a flowchart illustrating a method of predicting power usage according to an embodiment of the present invention.

Referring to FIG. 2 , the present invention may be a method of simplifying an integer code based on load variation data to classify a load pattern.

The prediction method of the present invention may consist of a data simplification and classification process, a process of verifying the classification, and a process of learning and predicting.

The data simplification and classification process ( S10 ) may include converting load data, which is basic data, into load variation data ( S12 ) and simplifying the pattern of the converted load variation data ( S13 ).

The classification verification process (S20) includes the steps of classifying the simplified pattern by date by the same pattern (S21), calculating the error rate by calculating the deviation between the simplified and classified pattern by date and the actual load pattern of the cluster (S22) , when the calculated error rate does not satisfy the target value, re-calculating the pattern simplification reference value and performing again from the simplification of the data classification process (S23).

In the learning and prediction process (S30), when the error rate of the simplified load pattern satisfies the target value, the arithmetic average value of the corresponding load pattern is determined as the representative pattern, and the machine learning learning step (S31) can be included and the future power amount can be predicted.

In more detail, a method of predicting power usage will be described as follows.

[Data simplification and classification process] (S10)

For example, the operation of performing the step (S11) of dividing and classifying the load data of the workplace by 24 hours per day is preceded, and the step (S12) of additionally calculating the amount of change in the load for each time period may be performed.

Then, in the pattern simplification step S13 , the daily load data may be simplified to an integer of -2α to 2α.

Here, α may be a variable that determines the number of equal divisions for pattern simplification.

In order to simplify the 24-hour load into an hourly integer, the raw data is converted into load change amount (ΔLoad _t,n ) data for the increase/decrease as shown in FIG. 4 , and then is replaced with an integer value (Code _t,n ) according to the load change amount can do.

Accordingly, n hourly integer sets can be obtained from n days of load data.

4 is a graph showing the load pattern of one day (24 hours) for one year (365 days) of electricity usage at the workplace, the horizontal axis is divided into 24 hours by hour, and the vertical axis is the load amount ( Load) shows the change, (b) shows the load change with respect to the load change of (a), and each of the plurality of lines shown in the graph of FIG. 4 shows the power consumption load of the workplace for one day.

In the load variation (ΔLoad _t,n ), t is time and n is date.

In order to calculate a simplified integer value (Code _{t,n )} from the load change amount (ΔLoad t,n ), the load change amount can be divided into 2α equal parts. At this time, the simplification reference value Step _α can be obtained through the step S14 according to Equation 1 below.

The simplified reference value Step _α may mean the size of the equalization value when dividing 2α into equal parts.

The integer value (Code _t,n ) of the simplified load pattern coded as an integer is the simplified reference value (Step α) based on the following Equation 2 when the load change amount (ΔLoad _t,n ) exceeds the simplified reference value (Step _{α )} ) can be recalculated. That is, the integer 0 to the integer 1, the integer 1 to the integer 2, etc. may be set.

C _t,n may be an integer value rounded down by dividing the load change amount and the simplified reference value Step _α as shown in Equation 3 below.

4, the first line (L1) and the second line (L2) are set for the maximum and minimum values of the load change amount (ΔLoad _t,n ) of the load pattern with respect to the power consumption of the factory (load), respectively, and , it is possible to form a plurality of dividing lines L3, L4, and L5 by setting the simplified reference value Step _α using an integer within the section between the first line L1 and the second line L2.

Accordingly, FIG. 4 shows an example in which four simplified reference value (Step _α ) sections are set within the maximum and minimum values of the load pattern data for the load change amount (ΔLoad _t,n ).

Accordingly, in FIG. 4(b) , the load pattern for the load change amount located on the 0 line of the load change amount axis (vertical axis) is an integer of 0, 1 If it is located above the dividing line L4, it is 1 or 2 In the case of being located above the dividing line (L4) (L3), it can be coded as 2, etc.

For example, the load pattern data line positioned in the section between the dividing line L3 and the dividing line L4 forming the simplified reference value Step _α may be set to the same integer. That is, the load pattern data lines existing at positions exceeding the dividing line L4 and not exceeding the dividing line L3 in the integer 0 may be equally coded as 1 and set.

Conversely, when the load pattern line is less than 0, it can be set to minus (-), and when it exceeds one equalization line L5, it can be set to -1. Similarly, the load data line existing between the dividing line L5 and the dividing line L2 may be coded with the same integer.

In other words, the section is defined by the dividing line, the code value of the section is determined, and +1 to the previous code value when it exceeds the dividing line upward, -1 to the previous code value when it exceeds downward, and in this way, 24 hours A code value can be created for In this case, the initial value may be 0.

If it exceeds two equalization lines upward at the next time, it may be +2, and if it exceeds three equalization lines upward, it may be +3. In the opposite case, it can be -2 and -3 in the same way.

Accordingly, FIG. 5 shows patterns 1 and 2 by simply patterning by collecting the same integer codes based on the graph shown in FIG. 4(b). However, the present invention is not limited thereto, and there may be load patterns having a greater number of identical integer codes.

In other words, after obtaining integer values (Code _t,n ) for all dates n, if patterns having integer values (Code _t,n ) for the same 24 hours are collected, they can be classified as shown in FIG. 4 .

[Classification verification process] (S20)

Meanwhile, in the classification verification process, a representative pattern may be calculated by verifying the simplified classified load pattern.

In order to determine the degree of similarity of how similar the classified load patterns are clustered, the average value of the clustered patterns and the error rate of individual patterns may be calculated.

The error rate may be calculated using a mean absolute percentage error (MAPE) calculation method as shown in Equation (4).

In Equation 4, △Load _t,n may mean a load change amount, and Xt may mean an average value of △Load _t,n for each time period.

In Equation 5, ΔLoad _t, n may mean a load change amount, Xt may mean an average value for each time period of ΔLoad _t,n , and N may mean the number of data (number of days).

Assume that the verification reference value (constant error) is 50%, and if the average absolute ratio error (MAPE _error ) is greater than 50%, it is determined that the clusters are not clustered with the same load pattern, and the variable α of the reference value (Stepα) for simplifying the data classification process By increasing by 1, the classification process of the load pattern can be repeated again.

In the opposite case, that is, when the average absolute ratio error (MAPE _error ) is less than 50%, the error rate is small, so it is determined that the classification of the load pattern is good, and the average value of the load pattern for each classification can be output as a representative pattern.

As raw data, the load data of company A's gas plant is used for 4 months, and the load data of the plant is shown in Table 1.

division Load (kWh) Maximum load per hour 36,404.98 Minimum load per hour 17,612.32 average daily load 734,391.94 Maximum load increment per hour 4,185.70 Maximum load reduction per hour 4,064.19

7 is a graph in which load data of a gas plant is converted into load variation data in order to classify patterns according to the present invention. To determine the increase or decrease of the load from the load change data, and to calculate the integer value (Code _t,n ), the simplified reference value (Step _α ) must be calculated.

When a pattern is classified with a low simplification reference value (Step _α with a small α), the spacing between the dividing lines becomes large, so the accuracy of the pattern decreases. As it becomes smaller, the number of data classified by the same pattern decreases, so it may be necessary to apply an appropriate step.

The results of calculating the size of the dividing line value while increasing _α of the simplified reference value (Step _α ) to 1 to 3 are shown in Table 2, and FIGS. It is a graph.

division Judgment standard load change (kWh) Step ₁ 4,124.95 Step ₂ 2,062.47 Step ₃ 1,031.24

The Min/Max line of FIG. 8 is a line representing the maximum load increase and the maximum load decrease per hour of the factory load change data.

In the graph of FIG. 8, when the interval between the Min/Max lines is divided into 2 equal parts based on 0 on the vertical axis and the load change amount in the time period to be determined increases larger than the size of C = 1 in the graph, the C values in the corresponding time period are 1 and 2 On the other hand, if the interval between the two lines is greatly reduced, the C value of the corresponding time period may be determined between -1 and -2, except for this, the C value is 0.

The simplified reference value (Step ₂ ) may be divided into four equal parts between the Min/Max lines as shown in FIG. 9 and C may be determined between -4 and 4 according to the amount of change in the load.

10 shows a simplified reference value (Step ₃ ) by dividing the Min/Max line into 8 equal parts, so that C can be determined between -8 and 8 according to the amount of change in the load. That is, when passing 8 equal parts upward at once, it becomes +8, and when it passes downward at once, it becomes -8.

An integer value (Code value) is determined according to the classified value of C, and through this, daily patterns can be simplified and classified.

Assuming that the criterion for determining the increase or decrease of load in pattern classification according to the present invention is Step ₂ , the results of classifying the patterns are the same as the four load patterns as shown in FIGS. 11 to 14 .

Referring to FIG. 11 , in the present invention, all Code _{t and n} values of the entire period are expressed as 0 and classified as one.

The algorithm according to the present invention can be equipped with an automatic correction function (α update function of Step _α ), and by automatically adjusting the simplification reference value appropriately, all meaningless and volatile parts are all converted to 0 code values and applied to the classification process. can

From the viewpoint of performing ESS charging and discharging schedules as a result of load clustering, it can be seen that the classification result of the present invention is an appropriate classification.

12 to 14 , it can be seen that similar patterns are classified.

As shown in Table 3, the MAPE error rate for the load variation of patterns 1 to 4 according to Figs.

the present invention Error rate (%) pattern 1 32.2 pattern 2 35.0 pattern 3 33.5 pattern 4 44.2

Since the present invention classifies clusters based on the amount of change (increase/decrease) of the load, as shown in FIG. 10, even if the size of the load is different, it can be classified in the same pattern. can have great advantages.

The pattern of increasing and decreasing load can be a very important factor in calculating the optimal ESS schedule.

15 is a graph showing the amount of change in load for each time period classified by the prediction method of the present invention. In FIG. 15, [a] to [d] show the amount of change in load of patterns 1-4.

In FIG. 15 , 1 column on the x-axis means 1 hour. Therefore, since the prediction method of the present invention classifies the pattern based on the load change amount (load increase/decrease amount), the load change amount of the classified pattern may have a constant value for each pattern in the corresponding time period.

The prediction method of the present invention classifies the pattern into a pattern with a constant amount of change, and the classification error also has little deviation, so it can show more useful results in determining the charge/discharge plan of the ESS when predicting through machine learning. .

Machine learning is a technology for predicting the future using big data with various data generation amounts, cycles, and formats.

The ESS schedule can be calculated based on the load prediction. Therefore, accurate load prediction can be very important in the optimal operation of ESS. In addition, since the pattern of increase and decrease of load is important in estimating the schedule of the ESS, a pattern classification method that considers the relevant matters may be required.

In the above, before using the AI (artificial intelligence) learning-based load prediction technique, a method of clustering patterns based on the increase and decrease of the load is provided.

3 is a block diagram illustrating a method of predicting future power usage by machine learning learning according to the present invention.

Referring to FIG. 3 , in the present invention, if a predetermined error rate reference value is satisfied through classification of a load pattern and calculation of a representative pattern, and then through error rate verification, a representative pattern is selected, and information about these representative patterns is provided to the machine learning learning unit (700) can be provided to learn.

The present invention learns by inputting simplified data through a code for a real-time load pattern and a corresponding load pattern for power usage through the machine learning learning unit 600, and then sets the machine learning model to configure the prediction unit 700 It is possible to predict the amount of electricity for the future.

In other words, according to the present invention, the load pattern and other learning materials can be learned through the machine learning learning unit 600 at the same time.

That is, the machine learning learning method is a method of simultaneously learning all data such as the operation time, temperature, division of weekdays or weekends, and operation schedule of the corresponding load (factory), and may be the most appropriate learning method.

However, it may be difficult to actually obtain such general data.

Therefore, in the present invention, it may be a method of converting load data, which is power usage data, into load variation data, generating a code and simplifying, and then learning the load variation data for the past power usage and the corresponding code in machine learning.

If the representative patterns extracted from two clustered load patterns with different loads are similar, correction is required to aggregate them into one cluster.

In order to supplement this problem, the present invention does not use the representative pattern for each individual pattern cluster, but uses an integerized code using the increase or decrease of the load as a means for learning the load pattern, and when machine learning performs code learning, , for example, a more accurate prediction result can be obtained because the code is the same even though the overall load size (load amount) is different as in the classification result of FIG. 11 .

That is, in the case of the corresponding loads clustered in FIG. 11 , the ESS charging and discharging schedules may be the same because the load pattern based on the load variation data is the same even if the load amount is different. 12 to 14 each have the same load pattern, so the ESS charging and discharging schedules of the corresponding cluster may all be the same.

Therefore, the present invention determines the charging and discharging schedule using the future power usage when determining the ESS charging and discharging schedule from the future power usage and code obtained as a result of machine learning prediction, and using the code to correct the power usage prediction error. ESS charging and discharging schedules can be adjusted.

In the present invention, the coding process is performed by calculating the load variation data through the data simplification and classification process, and the simplification reference value of the load variation data can be set in the coding process. Here, since the classification of the pattern may be derived differently according to the corresponding reference value, more improved classification results can be obtained by considering and reflecting the appropriate reference value calculation.

n... number of days L1... first line
L2... 2nd line L3,L4,L5... Equal line
t... time
Code _t,n ... Integer value Step _α ... Simplified reference value
△Load _t,n ... load change amount
S10... Data singularization and classification process
S20... Classification Verification Process
S30... Learning and Prediction Process
Step S11,S12,S13,S14,S21,S22,S23,S31...
10... pattern classification device
100... Load collection unit 200... Code conversion unit
300... Pattern classification unit 400... Error calculation unit
500... Judgment Unit 600... Machine Learning Learning Unit
700... prediction

Claims (17)

a load collecting unit that obtains load data for each time period;
a code conversion unit converting load data into load variation data and obtaining an integer corresponding to the load variation data;
a pattern classification unit for collecting load patterns having the same coded integer pattern and extracting a representative pattern from the same integer pattern;
a prediction unit for predicting future power usage from the representative pattern coded through the pattern classification unit and the actual load pattern;
A power usage prediction system comprising a.
According to claim 1,
Power including an error calculating unit for calculating the error rate of the representative pattern and individual patterns calculated as the average value of the clustered load patterns to determine the degree of similarity of how similar the load patterns classified through the pattern classification unit are. Usage forecasting system.
According to claim 1,
An error calculating unit for calculating the error rate of the representative pattern and the individual pattern is provided,
and a determination unit configured to determine the accuracy of the representative pattern calculated by the pattern classification unit based on the error rate.
According to claim 1,
In order to simplify the corresponding load data into integers for each time period, the code conversion unit obtains load variation data related to the increase/decrease of the load data, and then replaces the load variation data with an integer value.
According to claim 1,
The code conversion unit power usage prediction system for deriving a simplified reference value for calculating an integer value by the following equation.
[Equation]

Here, Step _α is the simplified reference value that is the size of each equal segment, and ΔLoad _t,n is Load variation data, MAX(ΔLoad _t,n ) is the maximum load variation, MIN(ΔLoad _t,n ) is the minimum load variation, t is the time, and n is the date. According to claim 1,
The code conversion unit,
A power usage prediction system that +1 each time the load variation data rises through each dividing line based on 0, and -1 every time it descends through each dividing line, and codes the load variation data as integers.
According to claim 1,
The code conversion unit, the power usage prediction system for encoding the load variation data into integers by the following equation.
[Equation]

here. Step _α is the simplified reference value, Code _t,n is an integer value according to the load change amount, C _t,n is the load change amount (ΔLoad _t,n ) divided by the simplified reference value (Step _α ) and rounded down, t is the time, n is the date.
According to claim 1,
An error calculating unit for calculating the error rate of the representative pattern and the individual pattern is provided,
When the representative pattern deviates from the predetermined error rate or more and cannot be adopted as the representative pattern according to the error rate of the calculated representative pattern, the error calculating unit reclassifies the load to re-calculate the simplified reference value Step _α . Power usage prediction system .
According to claim 1,
An error calculating unit for calculating the error rate of the representative pattern and the individual pattern is provided,
The error rate is a power usage prediction system that is calculated by the average absolute ratio error of the following equation.
[Equation]

Here, MAPE _error is the average absolute percentage error rate, △Load _t,n is the load change amount, Xt is the average value of △Load _t,n for each time period, and N is the number of days. According to claim 1,
an error calculating unit for calculating an error rate of the representative pattern and individual patterns;
A determination unit is provided to determine to set the representative pattern as the representative pattern according to the error rate of the representative pattern and the individual load pattern,
If the error rate calculated by the error calculation unit is satisfied within the verification reference value, the power usage prediction system for setting the average value of the load pattern for each classification as a representative pattern by the determination unit.
According to claim 1,
An error calculating unit for calculating the error rate of the representative pattern and the individual pattern is provided,
If the error rate calculated by the error calculating unit satisfies the verification standard, the average value of the load patterns for each classification is set as a representative pattern,
A machine learning learning unit that learns by receiving information about at least any one or more of the load's past operation time, temperature, weekday or weekend classification, operation schedule, or the set representative pattern,
Setting a machine learning model by the machine learning learning unit,
A power usage prediction system that inputs a real-time load pattern and a code of a real-time load pattern to the machine learning model, and predicts future power usage by a prediction unit based on the code and power usage up to the current time to be predicted. obtaining load data, which is basic data, and converting it into load variation data;
Simplifying by coding the load variation data into integers;
A method of predicting power usage, including
13. The method of claim 12,
classifying the integer values obtained for each time period based on the load variation data by date, and then classifying the integer values by the same integer pattern having the same integer value; including,
A power usage prediction method for extracting a representative pattern after classifying the simplified integer pattern by the same pattern.
13. The method of claim 12,
calculating an error rate by calculating a deviation between the integer pattern simplified for each day and an actual load of a group having the same integer pattern;
determining an arithmetic mean value of the integer pattern cluster as a representative pattern when the error rate satisfies a verification criterion;
A method of predicting power usage, including
15. The method of claim 14,
When the error rate is not satisfied with the target value, the power usage prediction method comprising the step of re-calculating the simplification reference value and starting from the simplification of the load change amount data.
13. The method of claim 12,
In order to simplify the load variation data, a variable α for determining the number of equal parts of the load variation data is employed.
13. The method of claim 12,
calculating an error rate by calculating a deviation between the simplified integer pattern for each day and an actual load of a corresponding group; including,
If the error rate calculated by the error calculating unit satisfies the verification standard, the average value of the load patterns for each classification is set as a representative pattern,
Machine learning learning by providing information about at least any one or more of the past operation time, temperature, weekday or weekend classification, operation schedule, or the set representative pattern of the load,
Setting a machine learning model by the machine learning learning,
Power usage prediction to predict the power usage up to the time to be predicted after inputting the code for the load pattern and the real-time load pattern so that the set machine learning model reflects and predicts the real-time load pattern based on the real-time load pattern code Way.

Images