JP5300761B2 - Energy distribution calculation device - Google Patents

Description

  The present invention relates to an energy allocation calculation apparatus and method for allocating an amount of energy consumed by a plurality of devices to each device.

  From the viewpoint of CO2 reduction, there is an increasing demand for monitoring the amount of energy consumed in commercial buildings. When it becomes necessary to manage the amount of energy consumed in units of companies under the revised Energy Saving Law, it is necessary to grasp the amount of energy consumed in units of tenants online in commercial buildings used by multiple tenants.

  Generally, in a commercial building, a measuring device is often installed for each system in the building, and it is difficult to permanently add a measuring device to existing facilities for each tenant. Therefore, instead of directly measuring the energy consumption amount of each tenant, it is necessary to perform a distribution process that measures the total energy consumption amount of tenants in the same system and distributes the total energy consumption amount to each tenant.

  Traditionally, the purpose of apportionment was to determine charges for electricity, gas, etc. on a tenant basis, so each month based on static information such as the tenant's occupation area or the installed device's rated value. Tenant ratios were determined, and the cumulative energy consumption of the entire building was apportioned to each tenant according to the ratio.

  However, especially when monitoring the amount of energy consumed by each tenant in real time, applying a fixed apportioning ratio will cause the amount of energy consumed for each tenant to be fixed to the amount of energy consumed by the entire building. Therefore, a reasonable apportioning result cannot be obtained.

JP 2007-219912 A JP 2007-148726

  The present invention provides an energy distribution calculation device that can appropriately estimate the amount of energy consumed by each device.

An energy allocation calculation apparatus as one aspect of the present invention is as follows.
A first database that stores a plurality of individual consumption patterns representing a temporal change pattern of energy consumption of each of the plurality of devices from a start time to an end time for each of the plurality of devices;
Measuring means for acquiring the entire consumption time series data by measuring the time change of the energy consumption amount by the whole of the plurality of devices at a predetermined time from the start time to the end time;
A time series database for storing the total consumption time series data acquired by the measuring means;
For the target period from the start time to the first time that is the time before the end time, the distance between the sum in the target period and the partial data of the target period in the total consumption time series data is a minimum or less than a threshold value A pattern selection unit that selects at least one of the individual consumption patterns for each device,
Based on the individual consumption pattern selected for each device and the partial data of the target period in the total consumption time series data, the individual indicating the time change of the energy consumption actually consumed by the device in the target period An allocation calculation unit for estimating consumption time series data,
The distribution calculating unit multiplies the partial data by the ratio of each value of each selected individual consumption pattern to the sum of each selected individual consumption pattern, thereby calculating the individual consumption time series data for each device. It is characterized by estimation.

  According to the present invention, it is possible to appropriately estimate the amount of energy consumed actually by each device.

1 is a block diagram showing a configuration of an energy allocation calculation device according to a first embodiment of the present invention. FIG. 2 is a supplementary explanatory diagram for explaining the operation of the apparatus of FIG. It is a figure which shows an example of the data of the integrated power consumption measured by the measurement means, and the time series data derived | led-out from the said data. It is a figure which shows an example of a 1st database. FIG. 3 is a diagram schematically showing a graph of the contents of a first database. It is a figure which shows an example of a complement pattern typically. It is a figure which shows an example of the apportionment process. FIG. 2 is a flowchart showing a flow of operations of the energy allocation calculation apparatus of FIG. FIG. 6 is a diagram for explaining a modification of the first embodiment. FIG. 6 is a block diagram showing a configuration of an energy allocation calculation device according to a second embodiment of the present invention. FIG. 11 is a supplementary explanatory diagram for explaining the operation of the apparatus of FIG. It is a figure which shows typically the example of the pattern obtained from the time series data for pattern generation, and the said data. It is a detailed block diagram of a time series clustering unit. It is a figure explaining determination of candidate time (candidate of representative time), curve approximation, and calculation of approximation error. 5 is a flowchart showing a flow of operations (first example) of a time-series clustering unit. 6 is a flowchart showing an operation flow (second example) of a time-series clustering unit. 6 is a flowchart showing a basic operation flow according to the second embodiment. FIG. 6 is a block diagram showing a configuration of an energy allocation calculation device as a third embodiment of the present invention. It is a figure which shows an example of a 4th database.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of an energy allocation calculation apparatus according to the first embodiment of the present invention. FIG. 2 is a supplementary explanatory diagram for explaining the operation of the apparatus of FIG.

  In the present embodiment, when there is a measuring unit 11 (watt-hour meter A) that sequentially measures the power consumption of a plurality of devices (here, device 1, device 2, and device 3), the power consumption of the entire device Are to be apportioned appropriately to device 1 to device 3, respectively.

  The measuring means 11 (watt-hour meter A) measures the total power consumption of the devices 1, 2 and 3 at a predetermined time (sometimes referred to as a representative time for convenience of explanation) t. . The measuring means 11 performs the measurement even before δt time before the predetermined time t, and divides the difference between the integrated value at time t and the integrated value at time t−δt by δt, so that the power consumption at time t An amount (power consumption per unit time from time t−δt to time t) is obtained. Hereinafter, the power consumption at time t may be referred to as an observed value. The measuring means 11 records the obtained power consumption (observed value) in the time series database P in association with the time t. Therefore, the power consumption amount at a predetermined observation time from the start time to the end time is recorded in the time series database P (S11 in FIG. 2). The time series data P recorded in the time series database P corresponds to the entire consumption time series data of the present invention. The time-series data is data indicating a temporal change in power consumption (energy consumption) by the entire device.

  FIG. 3 shows an example of accumulated power consumption data that is measured momentarily by the measuring means 11 and time-series data derived from the data.

  As shown in FIG. 3 (A), the integrated power consumption is measured at the observation time designated in advance from the start time t0 and before the δt time. For example, cumulative power consumption q0 (= 0) at time t0, cumulative power consumption q1 'at time t1-δt, cumulative power consumption q1 at time t1, cumulative power consumption q2' at time t2-δt, at time t2 The accumulated power consumption q2... Is measured. As shown in FIG. 3B, the power consumption q (t1) at time t1 is (q1-q1 ') / δt, and the power consumption q (t2) at time t2 is (q2-q2') / δt,... are recorded as time series data in the time series database P. When δt = 1, division by δt may not be performed. For example, the power consumption q (t1) may be obtained as q1-q1 '. Note that the method of acquiring the power consumption shown here is an example, and acquisition can be performed using any method according to the calculation function provided in the measurement unit 11.

  The database 1 (first database) in FIG. 1 includes the power consumption pattern (consumption pattern of the watt hour meter A) by the entire devices 1 to 3 from the start time to the end time, and typical power consumption patterns for each device. And are registered.

  FIG. 4 shows an example of the database 1. Fig. 4 (A) shows the consumption pattern of the entire device (total consumption pattern), Fig. 4 (B) shows the consumption pattern of device 1 (individual consumption pattern), and Fig. 4 (C) shows the consumption pattern of device 2 (individual consumption pattern). FIG. 4D shows the consumption pattern (individual consumption pattern) of the device 3.

  As shown in FIG. 4A to FIG. 4D, the representative time and the power consumption at that time are recorded for each pattern. Specifically, the power consumption amount at the representative time means the amount of power consumed per unit time from the predetermined time interval Δt before the representative time to the representative time. A pattern creation method will be described in the second embodiment. For easy understanding, the contents of the database 1 displayed in a graph are schematically shown in FIG. 5 and the reference numeral 1002 in FIG. However, for ease of viewing, each point (representative time and value) is connected with a smooth curve.

  In the example of FIG. 4, the number of patterns is p for the entire device and each device, and the number of patterns is the same, but the number of patterns may be different for each device.

  In addition, the representative time is the same for each device or for each pattern (both τ1,..., Τn), but may be different. The number of representative times (points) is the same for each device or pattern, but may be different. In this case, as will be described later, the representative points can be aligned with the overall consumption time series data from the approximate function of the pattern.

  The time series complementing unit 12 determines the pattern of the entire device that is partially similar to the time series data stored in the time series database P (the distance is the smallest) or the distance falls within a threshold value in the database 1. Select from the entire device pattern (total consumption pattern) (see FIG. 4A) (S12 in FIG. 2).

  Partially similar means that data is similar in a recording period in which time-series data is recorded. For example, as shown in FIG. 2, a pattern similar to the partial data 1001 of the target period that is a period from the start time to the current time (first time) is selected from the patterns of the entire device in the database 1.

  Various similar methods can be used as the similar calculation method. For example, the sum or square sum of the difference in values for each representative time included in the recording period is calculated between the time-series data and the pattern of the entire device. The pattern is calculated as the distance, and the pattern having the lowest distance or an error equal to or smaller than the threshold is selected. Alternatively, if the representative time is different between the time series data and the entire device pattern, the entire device pattern is approximated by a curve (such as a spline curve), and the time You may calculate the sum of the difference of the value for every observation time (representative time) contained in series data, or the sum of squares as an error.

  The time series complementing unit 12 extracts a portion from the current time to the end time in the selected pattern (total consumption pattern), and adds (complements) it to the end of the time series data (S13 in FIG. 2). What added the said part to the end of time series data is called a complement pattern. FIG. 6 schematically shows an example of the complement pattern. A broken line portion is a portion added to the end of the time series data. In FIG. 6, for easy understanding, it is displayed as a curve connecting points.

  If multiple patterns are considered to be similar, for example because the time from the start time to the current time is short, create an average pattern that takes the average value of all similar patterns, and The portion after the current time may be supplemented to the end of the time series data.

  The pattern selection unit 13 selects a pattern whose sum is most similar to the complementary pattern (the distance is the smallest) or whose distance is equal to or less than the threshold from the pattern group of each device (S14 in FIG. 2).

More specifically, assume that the complement pattern is expressed as y (t1), y (t2),..., Y (tn). Up to the current time, the value of the representative time (predesignated time) in the entire consumption time series data is the value of the representative time of the complemented pattern portion after the current time. In this case, the pattern selector 13 searches the j1, j2, j3 that the distance E ₁ defined by the following equation (1) below the minimum or threshold. E ₁ indicates the sum of squares of the difference between the complementary pattern and the sum of the patterns of each device. j1 is the pattern number of device 1, j2 is the pattern number of device 2, and j3 is the pattern number of device 3. The searched j1, j2, and j3 that minimizes E1 are represented as J1, J2, and J3. Further, regarding the representative time after the current time, when the representative time does not match between the complementary pattern and the pattern of each device, the representative time value of the complementary pattern is acquired using the spline curve.

  Based on the pattern selected for each device, the distribution calculation unit 14 apportions the time-series power consumption indicated by the complement pattern to each device (S14 in FIG. 2).

Specifically, if the power consumption of devices 1 to 3 after apportionment calculation at time t is x1 (t), x2 (t), and x3 (t), respectively, 7). That is, the power consumption amount for each device is obtained by multiplying the value of the complementary pattern by the ratio of the value of each pattern to the sum of the patterns.

  Thus, the power consumption consumed by each device is acquired in time series for each time from the start time to the end time. The value for each device obtained by equation (2) is a function that is approximated by a curve. However, the power consumption at time t for each device is an estimate of the power consumption actually consumed by the entire device from the start time to the current time, and the entire device from the current time to the end time. This is a predicted value of power consumption consumed in the future. Data indicating these power consumption amounts from the start time to the end time acquired for each device is referred to as individual consumption time series data.

  Here, in the above equation (2), f (t | v) represents an approximate curve obtained using the representative point v (value at the representative time). f (t | y) represents an approximate curve obtained by using y (observation points (representative points) of the entire consumption time series data until the current time, and representative points of the pattern after the current time).

  Stores information related to the facility information database 18, watt-hour meter and equipment. For example, the rating information of each device and the information on which watt hour meter each device is connected to are stored.

  The distribution result determination unit 15 determines whether the power consumption amounts x1 to x3 for each device obtained by the distribution calculation unit 14 are consistent with the rating information for each device recorded in the facility information database 18. For example, when the power value of the consumed energy amount x1 exceeds the rated power (threshold value) of the device 1, the pattern selection unit 13 is instructed to redo the pattern selection. As the re-selection of the pattern, for example, a combination of patterns with the same total sum (the next smallest distance) or another combination with a distance equal to or less than a threshold value is selected.

  When the distribution result determination unit 15 determines that there are no contradictory devices, the output unit 16 outputs the distribution result (the function obtained by the equation (2)) by the distribution calculation unit 14. By looking at the distribution result, the administrator can change the amount of power that each device would have consumed until the current time, and the amount of power consumption that is expected to be consumed by each device from the current time to the end time. You can see changes.

  FIG. 8 is a flowchart showing a flow of operations of the energy allocation calculation apparatus of FIG.

  In step S101, the time series complementing unit 12 reads from the time series database P time series data in which observation values are recorded at a predetermined time (representative time) from a given start time to the current time.

  In step S102, the time series complementing unit 12 selects a pattern similar to the time series data up to the current time from the pattern of the entire device recorded in the database 1.

  In step S103, the time series complementing unit 12 estimates an observation value for each representative time from the current time to a predetermined end time using the selected pattern. Specifically, a portion after the current time of the pattern is extracted. Then, the extracted part is complemented to the end of the time series data up to the current time to obtain a time series pattern (complement pattern).

  In step S104, the pattern selection unit 13 selects the patterns (f1 (t), f2 (t), f3 (t)) of the devices 1 to 3 so that the sum is most similar to the complementary pattern, for example. Details are as described above.

  In step S105, the distribution calculation unit 14 distributes the power consumption indicated by the complementary pattern based on the patterns f1 (t), f2 (t), and f3 (t) selected for each device. Thereby, the individual consumption time series data (refer to Formula (2)) indicating the power consumption from the start time to the end time in time series is acquired.

  In step S106, the distribution result determination unit 15 determines whether the distribution result contradicts the rating information for each device. When there is no contradiction, the process of this flow is terminated, and when there is a device inconsistent with the rating information, the process returns to step S104, and pattern selection for each device is performed again. In this case, for example, the pattern for each device is selected so that the sum is similar to the sum of the patterns for each device selected immediately before. That is, the most similar pattern combination is searched for except for the combination of selected patterns.

  Here, the modification of this embodiment is shown. First, if multiple patterns (total consumption patterns) are similar to the time series data (total consumption time series data), create an average pattern that takes the average of all the similar patterns, and He said that the average pattern may be used to complement the data. At this time, if variations (standard deviation or variance) of these similar patterns are recorded, a complementary pattern can be obtained with a confidence interval as a variation range.

  FIG. 9 is a diagram for explaining a modification of the present embodiment.

  Until the current time, time series data (total consumption time series data) of the entire device has been observed as shown in section 1010 in FIG. 9 (A), and the patterns (1) to (4) in database 1 ( It is assumed that the “total consumption pattern” is registered.

  For the period from the start time to the current time, patterns (1) to (3) having a small distance are selected by the time series complementing unit 12. Pattern (4) is out of selection due to its large distance.

  A distance is calculated for each of the patterns (1) to (3) up to the current time. At this time, the smaller the distance, the greater the weight. Weight 1 is assigned to pattern (1), weight 0.5 is assigned to pattern (2), and weight 0.5 is assigned to pattern (3).

  The thick line L1 in the section 1020 in FIG. 9 takes the weighted average by assigning the weights 1, 0.5, 0.5 to the patterns (1), (2), and (3), respectively, and shows the portion after the current time of the weighted average. This is the part that was taken out and added to the end of the time series data.

  The three dotted lines B1 to B3 correspond to portions after the current time of the patterns (1) to (3), respectively. The thin lines S1 and S2 define the range of variation due to the standard deviation.

  Similarly, averages and variations may be calculated for patterns selected for each device. Acquire all pattern combinations whose distance from the complementary pattern (which may be averaged) is less than or equal to the threshold, and calculate the average and variance of the acquired patterns for each device (considering all weights as 1) ). For each device, a distribution process is performed using an average pattern, and a distribution result obtained for each device is obtained with a range of variation (confidence interval) for each device.

  In the above embodiment, a partial pattern of the total consumption pattern is supplemented to the total consumption time series data, and by using the complementary pattern, the amount of power that each device would have individually consumed and the consumption is expected in the future. The time change of electric energy was shown. However, when it is not necessary to show the time change of the power consumption over the future, the power consumption shown in the total consumption time series data may be apportioned to each device. In this case, a pattern of each device whose sum is similar to the entire consumption time series data in the target period (from the start time to the current time (first time)) is selected, and apportioning processing may be performed in the same manner as described above. .

  As described above, according to the present embodiment, a pattern (individual consumption pattern) is selected for each device so that the sum is most similar to the total consumption time-series data or the complementary pattern, and power consumption by the entire device is based on the selected pattern. Since the amount is apportioned, it is possible to distribute the power consumption to each device so that it is closer to reality than the apportionment of the total power consumption at a fixed ratio as in the prior art.

<Second Embodiment>
FIG. 10 is a block diagram showing a configuration of an energy allocation calculation device according to the second embodiment of the present invention. FIG. 11 is a supplementary explanatory diagram for explaining the operation of the apparatus of FIG.

  A database 2 (second database), a time-series clustering unit 22, an operation information acquisition unit 23, a database 3 (third database), and a weather information acquisition unit 25 are added to the apparatus of FIG.

  The measuring means 11 is temporarily attached to each of the devices 1 to 3, measures the power consumption of each device from the start time to the end time, acquires the time series data for pattern generation for each device, and the database 2 To record. The time interval measured at this time is, for example, constant, and the number of times observed in the period from the start time to the end time is the number of representative times included in the pattern described in the first embodiment or predetermined. More than the number of times. A plurality of measuring means 11 may be prepared. Similarly, the power consumption of each device as a whole is measured to acquire time series data for pattern generation for the device as a whole and recorded in the database 2. The measurement of the devices 1 to 3 and the entire device may be performed, for example, in the same period (same time zone on the same day), or may be performed in different periods. The structure of the database 2 is schematically shown in the lower left of FIG.

  The operation information acquisition unit 23 determines the operating status of each device at the time of measurement by the measurement unit 11 of time series data stored in the time series database P and at the time of measurement by the measurement unit 11 of pattern generation time series data stored in the database 2. (Usage status) is collected, and the collected operating status is recorded in the database 3 in association with time information. The operation information indicates, for example, whether the corresponding device is operating.

  The pattern selection unit 13 can narrow down the pattern to be selected (individual consumption pattern) using the operation information collected when the time-series data P is acquired. For example, during the non-operation period, a pattern in which the amount of energy consumption is zero or less than or equal to a threshold value (when considering standby power or the like) may be specified, and selection may be performed for the specified pattern.

  The time-series clustering unit 22 generates a pattern for the device 1 from the time-series data for pattern generation of the device 1 recorded in the database 2 using clustering, and registers it in the database 1. Similarly, for the devices 2 and 3 and the entire device, patterns are generated from the respective pattern generation time-series data and registered in the database. In FIG. 12, as a simple example, there are three (1) to (3) pattern generation time series data for device 1, and one cluster 1 and time series data from time series data (1) and (2). Cluster 2 is obtained from (3), pattern 1 is generated from cluster 1, and pattern 2 is generated from cluster 2. The detailed configuration and operation of the time series clustering 22 will be described later.

  The meteorological information acquisition unit 25 acquires meteorological information of the area where each device is arranged. The weather information represents, for example, temperature or humidity. The weather information acquisition unit 25 temporarily stores the acquired weather information in an internal buffer. The weather information may be recorded at regular time intervals or may be acquired only at the start time. Weather information may be stored in association with the acquired time.

  The pattern selection unit 13 can narrow down the pattern to be selected (individual consumption pattern) using the weather information obtained from the weather information acquisition unit 25. For example, when the temperature is equal to or higher than the threshold value, a pattern in which the average power consumption is a certain value may be specified, and the specified pattern may be selected as a target.

  FIG. 13 is a detailed block diagram of the time-series clustering unit 22 of FIG.

  The time series clustering unit 22 includes a pattern configuration determination unit 31, a representative point number determination unit 32, a representative point determination unit 33, a classification unit 34, and an approximate function creation unit 35.

  The pattern configuration determination unit 31 determines the configuration for the pattern generation time series data of the entire device from the information recorded in the facility information database 18 and the operation information of the device recorded in the database 3.

  For example, it can be seen from the contents of the facility information database 18 that the device 1, the device 2, and the device 3 are arranged under the watt-hour meter A. Originally, it is desirable to obtain observation values individually for device 1, device 2, and device 3, but it is assumed that only the observation value of watt-hour meter A can be collected for some reason. At this time, if it is known that only the device 1 is operating in a certain time zone and the devices 2 and 3 are stopped, the observation value of the watt-hour meter A can be regarded as the observation value of the device 1. Therefore, in this case, the time series data of the entire device is regarded as the time series data of the device 1.

  In addition, if it is known that only device 3 is stopped on a certain measurement date, the total amount of device 1 and device 2 is observed, and device 1 and device 2 are regarded as one unit of the component. Therefore, it is regarded as time-series data of “device 1 + device 2”. In this case, the apportionment between “device 1 + device 2” and “device 3” can be realized.

  In the following, for simplicity of explanation, the time series data of “Equipment 1 + Equipment 2 + Equipment 3” (the time series data of the entire equipment acquired when all the equipments 1 to 3 are operating) Assume that the time series data, the time series data of device 2, and the time series data of device 3 are recorded in the database 2.

  The time series clustering unit 22 controls the representative point number determination unit 32, the representative point determination unit 33, the classification unit 34, and the approximate function creation unit 35, so that each of the entire device, device 1, device 2, and device 3, Perform pattern generation. The time-series clustering unit 22 pays attention to each of the entire device, the device 1, the device 2, and the device 3 individually, and generates a pattern using the pattern generation time-series data for each target object.

  The representative point number determination unit 32 determines the representative point number (number of representative times).

  The representative point determination unit 33 has a function of randomly selecting a time (candidate time) for the designated number (n) of points from the observation time. Also, it has a function of moving the selected time individually within a predetermined range. Fig. 14 (A) shows how four times are selected and one of them is moved.

  The approximate function creation unit 35 has a function of creating an approximate function of a specified point (time and value) group by curve approximation. The approximate function creation unit 35 has a function of calculating an approximation error between the created approximate function and the specified time series data. Further, it has a function of calculating an error (first evaluation value) Z described later.

Here, curve approximation and approximation error calculation will be described. For example, assume that curve approximation is performed with a spline curve. In this case, assuming that the time series data for pattern generation is represented by N observation points (observation time and observation value), a spline curve passing through n points selected from N observation points Draw (see Figure 14 (B)). Since the spline curve is an approximate curve that passes through the specified points, the approximation error of the approximate curve is calculated from the difference e (see FIG. 14B) between the remaining Nn observation points and the approximate curve. The approximation error is the total value or the maximum value of Nn differences e. The calculation method of the difference e between the observed value and the spline curve is specifically shown below.

  The classifying unit 34 obtains an n-dimensional vector representing the observed value at the representative time of the n points finally selected by the representative point determining unit 33 from the time-series data of the target object, and the vector is clustered into a plurality of clusters. Has a function to classify. Clustering is performed by, for example, the K-means method. Further, the classification unit 34 calculates a representative vector for each cluster. As a representative vector calculation method, an average of n vectors belonging to a cluster may be used, or a vector closest to the average may be selected.

  FIG. 15 shows an operation flow when the time-series clustering unit 22 controls the representative point number determination unit 32, the representative point determination unit 33, the classification unit 34, and the approximate function creation unit 35 to generate a pattern (first example). It is a flowchart which shows.

  First, a predetermined number of representative points (assumed to be n) is determined as an initial value (S201).

  The number of repetitions L is initialized to 1 (S202).

としてランダムに選択する（S203）（図14（A）参照）。 From the observation time included in the time series data for pattern generation of the target object (whole device, device 1, device 2, or device 3), the time of the specified number (n) is selected as the candidate time

Are randomly selected (S203) (see FIG. 14A).

  One candidate time is selected, and the selected candidate time is moved within a predetermined range (for example, a range up to adjacent candidate times) until the error (first evaluation value) Z is minimized (S204, S205). The moving destination is another observation time included in the range. Here, as the error Z, a total value, maximum value, average value, etc. of approximation errors of individual time-series data can be used. The selected candidate time is fixed at a position where the error Z is the smallest among the movements within the range. In the same manner, the remaining n−1 candidate times are similarly moved in the same manner, and sequentially fixed at the position where the error Z is minimized.

  The error Z when the n-th candidate time is fixed and the candidate time at that time are stored in the temporary storage unit (YES in S206, S207) (since the number of repetitions L = 1, here unconditionally Remember).

  When the number of repetitions L has not reached the predetermined value (NO in S208), the number of repetitions is incremented (S209), and the process returns to step S203. Thereafter, similarly, S204 to S207 are repeated until the number of repetitions L reaches a predetermined value. At this time, in step S206 in the second iteration count, it is determined whether or not the error Z obtained in S205 is smaller than the error Z stored in the temporary storage unit. When it is smaller, the contents of the temporary storage unit are updated with the error Z obtained at S205 and the candidate time at that time, and when it is more than that, the process proceeds to step S208 without updating the temporary storage unit.

  When the number of repetitions reaches a predetermined value (YES in S208), it is checked whether or not the error Z stored in the temporary storage unit is equal to or less than a threshold value (S210).

  When the error Z is larger than the threshold value (NO in S210), the number of representative points n is incremented by one, for example, and the process returns to step S202. At this time, all the contents of the temporary storage unit are erased.

を代表時刻として決定する（S212）。そして、着目対象のパターン生成用時系列データから、当該代表時刻の観測値を表すベクトルを取得し、ベクトルのクラスタリングを行う（S212）。そしてクラスタ毎に、当該クラスタに含まれるベクトルから代表ベクトルを生成し、生成した代表ベクトルを、着目対象のパターンとして、データベース1に登録する（S212）。 When the error Z is less than or equal to the threshold (YES in S210), the candidate time stored in the temporary storage unit

Is determined as the representative time (S212). Then, a vector representing the observed value at the representative time is obtained from the pattern generation time-series data of interest, and vector clustering is performed (S212). Then, for each cluster, a representative vector is generated from the vectors included in the cluster, and the generated representative vector is registered in the database 1 as a target pattern (S212).

  FIG. 16 is a flowchart showing an operation flow (second example) when the time-series clustering unit 22 performs pattern generation.

  First, set H = ∞.

  The flag is reset (S302). For example, resetting a flag sets the flag value to zero, and setting a flag corresponds to setting the flag value to one.

  The number of representative points (assumed to be n) is set to a predetermined initial value (S303).

  The number of repetitions L is initialized to 1 (S304).

としてランダムに選択する（S305）。 From the observation times in the pattern generation time-series data of the target of interest (whole device, device 1, device 2, or device 3), the time for the specified number of points (n) is selected as the candidate time

As a random selection (S305).

  One candidate time is selected, and the selected candidate time is moved within a predetermined range (for example, a range up to adjacent candidate times) until the error Z is minimized (S306, S307). The candidate time is fixed at the position where the error Z is the smallest in the range of movement. In the same manner, the remaining n−1 candidate times are similarly moved in the same manner, and sequentially fixed at the position where the error is minimized. S306 and S307 are the same processes as S204 and S205 in FIG.

For each of the time series data for pattern generation, curve approximation is performed at the observation points of n candidate times fixed in S307 and S306, and based on Equation (3) and Equation (4) below, The value E is calculated (S308). Then, the maximum value, average value, or total value of the value E is obtained as E * (second evaluation value) (S308). E * is an index for reconciling the number of representative points and an error, and the lower the better, the better (in this flow, an attempt is made to determine a candidate time that minimizes E *).

  It is checked whether or not E * is smaller than H (S309) (E * is always smaller than H because H = ∞ in the first processing).

  When E * is smaller than H, n candidate times fixed in S307 and S306 at this time are substituted into parameter S, and E * is substituted into H. (S310). Then, a flag is set (S311).

  It is checked whether or not the number of repetitions L has reached a predetermined value (S312). If the number of repetitions L has not reached the predetermined value (NO in S312), the number of repetitions is incremented (S313) and the process returns to step S305. . Thereafter, similarly, S305 to S311 are repeated until the number of repetitions L reaches a predetermined value.

  When the number of repetitions L reaches a predetermined value (YES in S312), it is checked whether the flag is set (S314).

  If it is set, the number of representative points n is incremented (S315), the flag is reset (S316), and the process returns to step S304.

として決定する（S317）。以降は、図15のS212と同様にして、パターンを取得し、データベース1に登録する。 On the other hand, when the flag is not set (NO in S314), the candidate time indicated by the parameter S is set as the representative time.

Is determined as (S317). Thereafter, the pattern is acquired and registered in the database 1 in the same manner as S212 in FIG.

  FIG. 17 shows a basic operation flow according to the present embodiment. In this embodiment, before the flow (S402) of the first embodiment, step S401 for creating a pattern of each device or the entire device using clustering is added. An example of the flow of S401 corresponds to the flow of FIG. 15 or FIG.

<Third embodiment>
FIG. 18 is a block diagram showing a configuration of an energy allocation calculation apparatus as the third embodiment of the present invention. A database 4 (fourth database) is added to the apparatus of the second embodiment shown in FIG.

  The database 4 stores a correspondence relationship between the pattern of the entire device and the pattern created for each of the devices 1 to 3.

  In the second embodiment, the time series data of the devices 1 to 3 are measured in the time zone on the same day as the time series data of the entire device, and the time series data for a certain period (for a plurality of days), the database 2 Is stored. Similar to the second embodiment, the time-series clustering unit 22 performs clustering to create a pattern.

  For example, clusters R1 and R2 are obtained from the time series data of the entire device, and patterns R1 and R2 are created from the clusters R1 and R2. Clusters A1, A2, and A3 are obtained from the time series data of device 1, and patterns A1, A2, and A3 are created from clusters A1, A2, and A3. Clusters B1 and B2 are obtained from the time series data of device 2, and patterns B1 and B2 are created from clusters B1 and B2. Clusters C1 and C2 are obtained from the time series data of device 3, and patterns C1 and C2 are created.

  The time-series clustering unit 22 determines to which cluster the entire device and the time-series data of the devices 1 to 3 acquired on the same day belong, and records the correspondence of the determined cluster pattern in the database 4. That is, the time series data acquired in the same period has a temporal correspondence relationship, and this relationship is recorded in the database 4 as a correspondence between patterns. For example, if the time series data of the entire device on a certain day belongs to cluster R2 and the time series data of devices 1 to 3 belong to clusters A1, B3, and C2, respectively, the pattern A2 of the entire device, pattern A1 of device 1, and device 2 The correspondence between the pattern B3 and the pattern C2 of the device 3 is recorded in the database 4. An example of the database 4 is shown in FIG.

  Based on the correspondence relationship stored in the database 4, the combination of patterns to be evaluated by the pattern selection unit 13 can be reduced. For example, when R1 is selected as a pattern for the entire device, all the rows in which the entire device is R1 are selected from the database 4, and the pattern for each device indicated in the selected row is selected. For example, if the entire device is R1, the only rows are ID2 and ID3, pattern A1 for device 1, pattern B2 and B1 for device 2, and pattern C1 for device 3, these candidates Select the target.

  As described above, according to the present embodiment, the processing time can be shortened and more appropriate distribution can be realized.

  Note that the energy distribution calculation devices shown in the first to third embodiments can be realized by using, for example, a general-purpose computer device as basic hardware. That is, each element in the device can be realized by causing a processor mounted on the computer device to execute a program. At this time, this apparatus may be realized by installing the above program in a computer device in advance, or may be stored in a storage medium such as a CD-ROM or distributed through the network. The program may be implemented by appropriately installing it in a computer device. In addition, each database in the apparatus appropriately uses a memory, a hard disk or a storage medium such as a CD-R, a CD-RW, a DVD-RAM, a DVD-R, or the like incorporated in or externally attached to the computer apparatus. Can be realized.

  Note that the present invention is not limited to the above-described embodiment as it is, and can be embodied by modifying the constituent elements without departing from the scope of the invention in the implementation stage. In addition, various inventions can be formed by appropriately combining a plurality of components disclosed in the embodiment. For example, some components may be deleted from all the components shown in the embodiment. Furthermore, constituent elements over different embodiments may be appropriately combined.

1: Database
2: Database
3: Database
4: Database
11: Measuring means (electricity meter)
12: Time series complement
13: Pattern selection section
14: Allocation calculation part
15: Distribution result judgment section
16: Output section
18: Equipment information database
22: Time-series clustering section
23: Operation information acquisition unit
25: Weather information acquisition department
31: Pattern configuration determination unit
32: Representative point determination unit
33: Representative point determination section
34: Classification section
35: Approximate function creation part P: Time series database

Claims (10)

A first database that stores a plurality of individual consumption patterns representing a temporal change pattern of energy consumption of each of the plurality of devices from a start time to an end time for each of the plurality of devices;
Measuring means for acquiring the entire consumption time series data by measuring the time change of the energy consumption amount by the whole of the plurality of devices at a predetermined time from the start time to the end time;
A time series database for storing the total consumption time series data acquired by the measuring means;
For the target period from the start time to the first time that is the time before the end time, the distance between the sum in the target period and the partial data of the target period in the total consumption time series data is a minimum or less than a threshold value A pattern selection unit that selects at least one of the individual consumption patterns for each device,
Based on the individual consumption pattern selected for each device and the partial data of the target period in the total consumption time series data, the individual indicating the time change of the energy consumption actually consumed by the device in the target period An allocation calculation unit for estimating consumption time series data,
The distribution calculating unit multiplies the partial data by the ratio of each value of each selected individual consumption pattern to the sum of each selected individual consumption pattern, thereby calculating the individual consumption time series data for each device. An energy allocation calculation device characterized by estimating. The pattern selection unit acquires a plurality of combinations of individual consumption patterns for each device such that a distance between the total and the partial data of the target period in the total consumption time series data is equal to or less than the threshold value,
Calculate the average and variance of each individual consumption pattern for each device based on each combination,
The said allocation calculation part estimates the said individual consumption time series data with the confidence interval by the said dispersion | variation with respect to the said average using each average pattern for every said apparatus. Energy allocation calculation device. The first database stores a plurality of overall consumption patterns representing a temporal change pattern of energy consumption by the whole of the plurality of devices from the start time to the end time,
The pattern selection unit
Select a total consumption pattern from the first database in which the distance between the partial pattern in the target period and the partial data in the target period in the total consumption time-series data is a minimum or a threshold or less,
At the end of the partial data of the target period in the total consumption time series data, obtain a complementary pattern that adds a portion from the first time to the end time of the selected total consumption pattern,
Select at least one of the individual consumption patterns for each of the devices so that the sum of all periods from the start time to the end time and the distance between the complement pattern is the minimum or less than a threshold,
The distribution calculation unit, based on the individual consumption pattern selected for each device and the complementary pattern, changes with time in the amount of energy actually consumed by the device in the target period, and after the first time. Estimating the individual consumption time-series data indicating the time change of the amount of energy consumption that the device will consume by the end time from
2. The energy allocation calculation apparatus according to claim 1, wherein The pattern selection unit acquires a plurality of combinations of individual consumption patterns for each device such that a distance between the sum and the complementary pattern is equal to or less than the threshold value,
Calculate the average and variation of each individual consumption pattern for each device based on each combination,
The said allocation calculation part estimates the said individual consumption time series data with the confidence interval by the said dispersion | variation with respect to the said average using each average pattern for every said apparatus. Energy allocation calculation device.   The pattern selection unit acquires all the total consumption patterns whose distance from the partial data in the target period in the total consumption time series data is equal to or less than the threshold, and uses the average of the acquired total consumption patterns to perform the complement The energy allocation calculation apparatus according to claim 4, wherein a pattern is acquired. A facility information database for storing rating information for each device;
A distribution result determination unit that determines whether the individual consumption time-series data is inconsistent with the rating information for each device;
The pattern selection unit, when it is determined by the distribution result determination unit that there is a device inconsistent with the rating information, except for the set of individual consumption patterns that have been selected so far, The energy allocation calculation device according to any one of claims 3 to 5, wherein another set of individual consumption patterns for each device that has a minimum distance or less than a threshold is selected. Stores the time series data for pattern generation representing the time change of the amount of energy consumption for each of the device and the entire device obtained by actually measuring the device and the entire device from the start time to the end time. A second database;
(A) selecting a plurality of times between the start time and the end time;
(B) Obtain a numerical value group indicating the values of the plurality of times and the amount of energy consumption at the times from the time series data for pattern generation,
(C) Approximate calculation of a graph that passes the values indicated in the numerical value group at the plurality of times for each of the devices and the entire device, and calculates an approximation error between the graph and the time series data for pattern generation ,
(D) For each of the devices and the entire device, calculate an evaluation value based on the approximation error,
(E) The time when the highest evaluation value is obtained by changing the value of the time to be selected or the number of times to be selected and repeating (A) to (D) is determined as the representative time,
(F) by clustering vectors representing the representative time values in the pattern generation time-series data for each of the devices and the entire device, thereby generating a plurality of clusters each including one or more vectors,
(G) For each cluster, by calculating an average of the vectors or by selecting a vector closest to the average of the vectors, a representative vector is obtained, and the representative vector acquired for each cluster is A time-series clustering unit that registers the individual consumption pattern and the overall consumption pattern in the first database;
The energy distribution calculation device according to claim 3, further comprising: An operation information acquisition unit that acquires operation information of the device at the time of measurement by the measurement unit;
A third database for storing the operation information; and
The pattern selection unit specifies an individual consumption pattern in which the amount of energy consumption is zero or less than a threshold value during the non-operating period of the device for each device, and performs selection from the specified individual consumption pattern. The energy allocation calculation device according to any one of claims 1 to 7. A weather information acquisition unit that collects weather information of an area where the device is arranged at the time of measurement by the measurement unit;
2. The pattern selection unit identifies an individual consumption pattern having a time change in energy consumption according to the weather information for each device, and selects from the identified individual consumption pattern. The energy allocation calculation apparatus as described in any one of thru | or 8. A fourth database,
The time series data for pattern generation of the device and the entire device is measured simultaneously,
The time series clustering unit identifies the individual consumption pattern and the overall consumption pattern respectively corresponding to the time series data for pattern generation for each device measured simultaneously,
The fourth database stores the correspondence between the specified individual consumption pattern and the overall consumption pattern,
The pattern selection unit specifies all individual consumption patterns corresponding to the selected overall consumption pattern for each of the devices based on the fourth database, and performs selection from the specified individual consumption patterns. The energy allocation calculation apparatus according to claim 7.

Images