JP2004266927A - Apparatus and method for predicting time series waveform, demand and supply regulating system, power demand and supply regulating system, computer program and computer-readable recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To make accurately predictable the future trend of time series waveform changing unsteadily. <P>SOLUTION: A method of predicting a time series waveform includes steps of extracting a cluster 52f one prior to a prediction start time t<SB>c</SB>from a plurality of clusters 52, generated based on time series information 51 of power consumption at a predetermined time (t<SB>0</SB>to t<SB>p</SB>) one prior to from the prediction starting time t<SB>c</SB>; identifying the cluster 53 appearing next of the extracted cluster 52f based on a past result; accurately identifying the cluster at the prediction starting time t<SB>c</SB>, based on a probabilistic statistical basis at the prediction starting time t<SB>c</SB>, by setting the identified cluster to the cluster at the prediction starting time t<SB>c</SB>; and thereby making accurately predictable the change in the cluster subsequent to the prediction start time t<SB>c</SB>. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

【０１４８】
上記（７式）において、ＤＳは、要求電力に対して不足している電力量であり、ＳＳは、上記不足電力量ＤＳ分をリカバリするための電力量である。区間Ｔは上記不足分をリカバリする区間（時間）である。区間ｔはリカバリ区間Ｔのうち、一定の電力（上記不足分をリカバリするための電力値ｘ）で発電を行う区間（時間）である。なお、以下の説明では、現在時刻ｔ_Ｃにおける電力値Ｐ_Ｃを不足電力と表し、不足分をリカバリするための電力値Ｐ_Ｒをリカバリ電力と表す。
【０１４９】
そして、図１４に示すように三角形の比の定理を使って上記一定区間電力発電区間ｔを以下の（８式）のように既知の値で表現する。
Ｔ：ｔ＝（Ｔ×ｂ／２）：（Ｔ×ｂ／２−ｘ）
ｔ＝Ｔ−２×（ｘ／ｂ）・・・（８式）
【０１５０】
上記（８式）において、ｂは、不足電力Ｐ_ｃからリカバリ電力Ｐ_Ｒまで上昇させる際の速度であり、そのときの勾配をｄ［度］とすると、以下の（９式）で表される。
ｂ＝ｔａｎｄ・・・（９式）
【０１５１】
そして、上記（８式）を上記（７式）に代入すると、以下の（１０式）が成立し、要求電力に対する不足分をリカバリするために、どの位電力を増加させればよいのかを求めることができる。
ｘ＝［Ｔｂ±｛（Ｔ×ｂ）^２−４ＤＳｂ｝^１／２］／２・・・（１０式）
【０１５２】
次に、図１５のフローチャートを参照しながら、電力調整装置９における動作を説明する。
【０１５３】
まず、最初のステップＳ１１において、データ処理部１０１に、第１の需要家６ａからの通告電力に関する情報と、第１〜第３の発電所１０ａ〜１０ｃにおける現在の発電量に関する情報とを、インターネット通信網４を介して入力する。また、第２〜第４の需要家６ｂ〜６ｄにおける今後の電力需要に関する情報を入力する。そして、上記入力した各種の情報を制御条件格納領域１０２ａに格納する。
【０１５４】
次に、ステップＳ１２において、ポリゴン分割部１０３は、制御条件格納領域１０２ａに格納されている通告電力に関する情報に基づいて、次区間において要求電力に変更があるか否かを判定する。この判定の結果、要求電力に変更がある場合にはステップＳ１３に進む。一方、要求電力に変更がない場合には、ステップＳ１３〜Ｓ１５の動作を省略して後述するステップＳ１６に進む。
【０１５５】
そして、ステップＳ１３において、ポリゴン分割部１０３は、制御条件格納領域１０２ａに格納されている所定の条件に従って最適性を無視した現区間における電力制御特性線を生成し、上記生成した電力制御特性線の変曲点を境界とした複数のポリゴン構造体Ｐ１、Ｐ１１〜Ｐ１５、Ｐ２１〜Ｐ２５を形成する。
【０１５６】
次に、ステップＳ１４において、ポリゴン定義部１０４は、ステップＳ１３で形成されたポリゴン構造体Ｐ１、Ｐ１１〜Ｐ１５、Ｐ２１〜Ｐ２５を解析し、個々のポリゴン構造体Ｐ１、Ｐ１１〜Ｐ１５、Ｐ２１〜Ｐ２５に対する属性情報を生成して記憶する。
【０１５７】
次に、ステップＳ１５において、変動ポリゴン選択部１０５は、制御条件格納領域１０２ａからポリゴン構造体の選択に関わる条件を抽出する。そして、上記抽出した条件と、ステップＳ１４で生成された個々のポリゴン構造体Ｐ１、Ｐ１１〜Ｐ１５、Ｐ２１〜Ｐ２５に対する属性情報とに基づいて、差分計算を行う２つのポリゴン構造体を選択し、上記選択した２つのポリゴン構造体に関するポリゴン情報をポリゴン面積差分計算部１０６に出力する。
【０１５８】
次に、ステップＳ１６において、ポリゴン面積差分計算部１０６は、ステップＳ１５で選択されたポリゴン構造体に関する情報に基づいて、区間全体（ｔ_ｆ−ｔ_ｓ）の面積極性がプラスマイナス０となるような電力制御特性（条件）を算出する。
【０１５９】
なお、ステップＳ１２において要求電力の変更がないと判定された場合には、このステップＳ１６において、ポリゴン面積差分計算部１０６は、上記（９式）を満足するような電力制御特性（条件）を算出する。
【０１６０】
次に、ステップＳ１７において、ポリゴン面積差分計算部１０６は、ステップＳ１６で設定した電力制御特性（条件）が所定の適正範囲内であるか否かを判定する。この判定の結果、所定の適正範囲内にない場合には、ステップＳ１５の動作に戻り、設定した条件が所定の適正範囲内となるまでステップＳ１５〜Ｓ１７の動作を繰り返し行う。なお、この際、ステップＳ１５において、変動ポリゴン選択部１０５は、前回選択したポリゴン構造体のうちの少なくとも１つが異なるように、ポリゴン構造体を選択し直すようにする。
【０１６１】
次に、ステップＳ１８において、ステップＳ１６で算出された電力制御特性を最適性が確保された電力制御特性として最適制御値格納領域１０２ｂに格納する。
【０１６２】
以上のように、本実施の形態では、前区間における要求電力に対し、次区間における要求電力の変更があった場合には、最適性を無視した電力制御特性線１１１を作成し、上記作成した電力制御特性線１１１の変曲点を境界とした複数のポリゴン構造体Ｐ１、Ｐ１１〜Ｐ１５、Ｐ２１〜Ｐ２５を形成し、上記形成したポリゴン構造体それぞれに対する属性情報と、上記ポリゴン構造体の選択に関わる所定の制御条件とに基づいて、上記複数のポリゴン構造体の中から２つのポリゴン構造体を選択し、上記複数のポリゴン構造体の面積の差分がプラスマイナス０となるように区間Ｘを決定してポリゴン構造体Ｐ１１、Ｐ２４を変更し、その変更結果に従って電力制御特性線１１１を変更するようにしたので、非定常的に変化する複雑な電力制御特性の作成と、上記電力制御特性の評価とを同時に行うことができ、初期制御特性線（最適性を無視して作成した制御特性線）を試行錯誤により何度も作成しなおす必要がなくなる。したがって、第１の需要家６ａである電力会社からの通告電力と、第２〜第４の需要家６ａ〜６ｄであるプラントで今後される使用されると予測される電力との両方を満足する最適な電力制御特性を高速に且つ確実に演算することができ、過不足のない電力を需要者群２に安定して供給することができる。
【０１６３】
なお、本願発明者は、本実施の形態の電力需給調整システム１をＩＰＰ（電力卸供給）システムに適用した場合に、上述した電力供給のための全ての処理を、人的介入を行うことなく１分以内で完了することができることを確認した。
【０１６４】
また、上述した図６のステップＳ１〜Ｓ９と、図１５のステップＳ１１〜１８の処理を繰り返し行うようにしたので、需要家群２における区間の初期の予測精度が高いのは勿論であるが、需要家群２における区間の終了に近づくほど、第２〜第４の需要家６ｂ〜６ｄで消費される電力をより正確に予想することができるようになる。したがって、需要家群２における電力需要と、電気事業者１における電力供給とのバランスが区間の途中で崩れたとしても、区間が終了するまでには、これらのバランスを確実に回復させることができる。
【０１６５】
なお、本実施の形態のようにして電力制御特性を演算するようにすれば、最適な電力制御特性を高速に且つ確実に得ることができ好ましいが、電力制御特性を演算する方法としては、上述した方法に限定されない。例えば、区間求積法、シンプソンの積分公式、またはガウス型数値積分法などの定積分近似解法を用いて行うことも可能である。
【０１６６】
また、本実施の形態では、前区間における要求電力に対し、次区間における要求電力が増加する場合について説明したが、前区間における要求電力に対し、次区間における要求電力が減少する場合についても上述したのと同様にしてポリゴン構造体を定義して、最適な電力制御特性を得ることができるということは言うまでもない。
【０１６７】
（本発明の他の実施形態）
上述した実施形態の機能を実現するべく各種のデバイスを動作させるように、上記各種デバイスと接続された装置あるいはシステム内のコンピュータに対し、上記実施形態の機能を実現するためのソフトウェアのプログラムコードを供給し、そのシステムあるいは装置のコンピュータ（ＣＰＵあるいはＭＰＵ）に格納されたプログラムに従って上記各種デバイスを動作させることによって実施したものも、本発明の範疇に含まれる。
【０１６８】
また、この場合、上記ソフトウェアのプログラムコード自体が上述した実施形態の機能を実現することになり、そのプログラムコード自体、およびそのプログラムコードをコンピュータに供給するための手段、例えば、かかるプログラムコードを格納した記録媒体は本発明を構成する。かかるプログラムコードを記録する記録媒体としては、例えばフレキシブルディスク、ハードディスク、光ディスク、光磁気ディスク、ＣＤ−ＲＯＭ、磁気テープ、不揮発性のメモリカード、ＲＯＭ等を用いることができる。
【０１６９】
また、コンピュータが供給されたプログラムコードを実行することにより、上述の実施形態の機能が実現されるだけでなく、そのプログラムコードがコンピュータにおいて稼働しているＯＳ（オペレーティングシステム）あるいは他のアプリケーションソフト等と共同して上述の実施形態の機能が実現される場合にもかかるプログラムコードは本発明の実施形態に含まれる。
【０１７０】
さらに、供給されたプログラムコードがコンピュータの機能拡張ボードやコンピュータに接続された機能拡張ユニットに備わるメモリに格納された後、そのプログラムコードの指示に基づいてその機能拡張ボードや機能拡張ユニットに備わるＣＰＵ等が実際の処理の一部または全部を行い、その処理によって上述した実施形態の機能が実現される場合にも本発明に含まれる。
【０１７１】
【発明の効果】
上述したように本発明によれば、時系列波形の予測を開始するときまでの所定時間における時系列波形の実績情報に基づいて生成した複数のクラスターの中から、上記予測を開始した時刻を含むクラスターの１つ前のクラスターを抽出し、上記抽出したクラスターの次に出現するクラスターを、過去の学習結果に基づいて同定し、上記同定したクラスター以降に出現するクラスターを、上記過去の学習結果に基づいて順次選択するようにしたので、上記予測を開始した時刻におけるクラスターを確率統計学的根拠に基づいて高精度に同定することができる。これにより、上記予測を開始した時刻以降におけるクラスターの変遷を高精度に予測することができる。したがって、非定常的に変化する時系列波形の今後の動向を高精度に予測することができる。
【図面の簡単な説明】
【図１】本発明の実施の形態を示し、本発明の実施の形態を示し、電力需給調整システムの構成の一例を示した図である。
【図２】本発明の実施の形態を示し、電力需要予測装置の構成の一例を示したブロック図である。
【図３】本発明の実施の形態を示し、複数のクラスターを生成する際の処理を説明するための図である。
【図４】本発明の実施の形態を示し、複数のクラスター相互間の関係を学習する際の処理を説明するための図である。
【図５】本発明の実施の形態を示し、電力需要を予測する際の処理を説明するための図である。
【図６】本発明の実施の形態を示し、電力需要予測装置の動作を説明するフローチャートである。
【図７】本発明の実施の形態を示し、本実施の形態における予測方法で電力需要を予測した場合と、他の方法で電力需要を予測した場合とにおける予測精度を調査した結果を示した図である。
【図８】本発明の実施の形態を示し、予測開始時刻におけるクラスターを人的介入により任意に同定して電力需要を予測する方法を説明する図である。
【図９】本発明の実施の形態を示し、２０時間先までの電力需要を１時間毎に予測したときの結果と、実績データとの誤差を表形式で示した事例の図である。
【図１０】本発明の実施の形態を示し、電力調整装置の構成の一例を示したブロック図である。
【図１１】本発明の実施の形態を示し、要求電力に変更があった場合の電力制御特性の一例を示した図である。
【図１２】本発明の実施の形態を示し、ポリゴン構造体の属性情報を説明するための図である。
【図１３】本発明の実施の形態を示し、ポリゴン構造体における面積の極性の関係と、三角形のポリゴン構造体における斜辺の極性の関係とを示した図である。
【図１４】本発明の実施の形態を示し、要求電力に変化がない場合の電力制御特性の一例を示した図である。
【図１５】本発明の実施の形態を示し、電力調整装置における動作を説明するフローチャートである。
【符号の説明】
１ 電力需給調整システム
２ 需要家群
３ 電力系統ネットワーク
４ インターネット通信網
５ 電気事業者
６ 需要家
７ 需要電力計測装置
８ 電力需要予測装置
９ 電力調整装置
１０ 発電所
１１ 供給電力計測装置
２１、１０１ データ処理部
２２、１０２ 情報記憶部
２２ａ 学習対象時系列情報格納領域
２２ｂ 予測開始時点前時系列情報格納領域
２２ｃ 学習情報格納領域
２２ｄ 予測結果時系列情報格納領域
２３ クラスター学習器
２３ａ クラスター分割部
２３ｂ クラスター分類部
２４ａ 初期クラスター同定部
２４ｂ クラスター選択部
２４ クラスター予測器[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a time-series waveform prediction device, a supply and demand adjustment system, a power supply and demand adjustment system, a time-series waveform prediction method, a computer program, and a computer-readable recording medium, and in particular, a non-stationarily changing time-series waveform. It is suitable for use in predicting future trends in information.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, a future trend of a time-series waveform is predicted (for example, see Patent Literatures 1 to 3).
In particular, there is a technique for averaging actual values of the time-series waveform information within a predetermined time, and employing the averaged value as a predicted value of the time-series waveform information.
[0003]
[Patent Document 1]
JP-A-8-308108 [Patent Document 2]
JP 2002-169613 A [Patent Document 3]
JP, 2002-41714, A
[Problems to be solved by the invention]
However, the conventional technique described above has a problem in that if the time-series waveform changes unsteadily due to abrupt increase or decrease in the time-series waveform, for example, the accuracy of prediction is reduced.
[0005]
That is, in the above-described conventional technique, a future trend can be predicted with high accuracy for a time-series waveform that constantly changes or a time-series waveform with little change, but a time-series waveform that changes unsteadily. However, it was difficult to predict future trends with high accuracy.
[0006]
The present invention has been made in view of the above-described problems, and has as its object to enable highly accurate prediction of a future trend of a non-stationarily changing time-series waveform.
[0007]
[Means for Solving the Problems]
The time-series waveform prediction device of the present invention is a time-series waveform prediction device that predicts a future trend of a time-series waveform given from a device to be predicted, and within a predetermined time until the start of the prediction. Cluster generating means for generating a plurality of clusters based on the performance information indicating the trend of the device to be predicted; and a cluster including a time at which the prediction is started among the plurality of clusters generated by the cluster generating means. Cluster extraction means for extracting the previous cluster and identifying the cluster appearing next to the extracted cluster based on the past learning result, and clusters appearing after the cluster identified by the cluster identification means Cluster selecting means for sequentially selecting based on the past learning results; and And cluster identified me, based on the cluster selected by the cluster selecting means is characterized by having a prediction means of the time-series waveform of predicting the future trend of the time-series waveform.
[0008]
The supply and demand adjustment system of the present invention adjusts the operation of the prediction device for a time-series waveform according to any one of the above, and the device to be predicted so that the time-series waveform predicted by the prediction device for a time-series waveform is obtained. And an adjusting device.
[0009]
An electric power supply and demand adjustment system according to the present invention includes: a time-series waveform prediction device described above; and a power generation device that supplies power to the prediction target device so as to obtain a time-series waveform predicted by the time-series waveform prediction device. A power adjusting device for adjusting the power generation operation of the facility.
[0010]
The method for predicting a time-series waveform according to the present invention is a method for predicting a future trend of a time-series waveform given from a device to be predicted, which is a method for predicting a future trend. A cluster generation process for generating a plurality of clusters based on actual information indicating a trend of the device to be predicted; and a cluster including a time at which the prediction is started, from a plurality of clusters generated by the cluster generation process. A cluster identification process for extracting the previous cluster and identifying a cluster that appears next to the extracted cluster based on the past learning result, and a cluster that appears after the cluster identified by the cluster identification process A cluster selection process for sequentially selecting based on the past learning results and a cluster identification process. And cluster identified me, based on the cluster selected by the cluster selection process is characterized by performing a prediction processing of the time-series waveform of predicting the future trend of the time-series waveform.
[0011]
The computer program of the present invention, when predicting the future trend of the time-series waveform given from the device to be predicted, the performance information representing the trend of the device to be predicted within a predetermined time until the start of the prediction A cluster generation process for generating a plurality of clusters based on the cluster generation process, and extracting, from the plurality of clusters generated by the cluster generation process, a cluster immediately preceding the cluster including the time at which the prediction was started, and A cluster identification process for identifying a cluster that appears next to the cluster based on the past learning result, and clusters that appear after the cluster identified by the cluster identification process are sequentially selected based on the past learning result. Cluster selection processing and the clusters identified by the cluster identification processing described above. Star and, based on the cluster selected by the cluster selection process is characterized in that to execute the prediction processing of the time-series waveform of predicting the future trend of the time-series waveform to the computer.
[0012]
A computer-readable recording medium according to the present invention has the above-described computer program recorded thereon.
[0013]
BEST MODE FOR CARRYING OUT THE INVENTION
Next, an embodiment of a time-series waveform prediction device, a time-series waveform prediction method, a supply and demand adjustment system, a power supply and demand adjustment system, a computer program, and a computer-readable recording medium according to the present invention will be described with reference to the drawings. explain.
[0014]
FIG. 1 shows an embodiment of the present invention and is a diagram showing an example of a configuration of a power supply and demand adjustment system. In the present embodiment, information about power consumed by a plurality of customers will be described as an example of a time-series waveform to be predicted.
[0015]
In FIG. 1, a power supply and demand adjustment system 1 includes a customer group 2, a power system network 3, an Internet communication network 4, and an electric utility 5.
The customer group 2 includes first to fourth customers 6a to 6d and first to fourth power demand measuring devices 7a to 7d.
[0016]
The first customer 6a is a power company, requests required power from the electric power company 5 via the first demand power measuring device 7a and the Internet communication network 4, and transmits the power via the power system network 3. The power supply 5 is provided with equipment capable of receiving the requested power supply from the electric utility 5.
[0017]
The first demand power measuring device 7a measures the power demanded by the power company as the first customer 6a, and transmits the information on the measured power to the electric power company 5. In the following description, the power required by the power company is referred to as notification power.
[0018]
The second to fourth consumers 6b to 6d are plants for manufacturing a predetermined product, and are configured to have facilities capable of receiving power supply from the electric power company 5 through the power system network 3. ing.
[0019]
The second to fourth demand power measuring devices 7b to 7d monitor the power consumption of the second to fourth consumers 6b to 6d, respectively, and transmit the information on the monitored power consumption via the Internet communication network 4. It is configured to transmit to the electric utility 5. The information on the power consumption includes a power value and a power amount consumed by the second to fourth consumers 6b to 6d.
[0020]
The electric utility 5 includes a power demand prediction device 8, a power adjustment device 9, first to third power plants 10a to 10c, and first to third supply power measurement devices 11a to 11c. I have.
[0021]
The power demand prediction device 8 inputs the information on the power consumption from the second to fourth power demand measurement devices 7b to 7d, and based on the input information on the power consumption, outputs the second to fourth customers 6b. 6d to predict future power demand. A specific method for estimating the power demand will be described later.
[0022]
The power adjustment device 9 includes information on the future power demand in the second to fourth customers 6b to 6d predicted by the power demand prediction device 8, information on the notification power from the first customer 6a, The power to be supplied is adjusted based on the information on the current power generation amount in the first to third power stations 10a to 10c.
[0023]
The power adjusting device 9 creates a power control characteristic for generating power according to a request of the first to fourth consumers 6a to 6d for each predetermined section based on the input various information. (Refer to FIG. 11 and FIG. 14) to adjust the power generation amounts of the first to third power plants 10a to 10c according to the power control characteristics created as described above. A specific method for creating the power control characteristics will be described later.
[0024]
The first to third power stations 10 a to 10 c generate power under the control of the power adjusting device 9 and supply the generated power to the customer group 2 via the power system network 3. Each of the first to third power plants 10a to 10c has a generator, a turbine, a boiler, a DCS (distributed control system), and the like.
[0025]
However, the first to third power plants 10a to 10c are not limited to such, and may be an aggregate of distributed power sources. For example, power generation equipment that can arbitrarily control the output of a small gas engine, micro gas turbine, or cogeneration power generation, or arbitrarily control the output such as power generation using natural energy such as solar power or wind power generation It may be a power generation facility that is unable to do so, or may be a combination of these.
[0026]
The first to third supply power measuring devices 11a to 11c measure the current power generation amount in the first to third power plants 10a to 10c, and transmit the information on the measured power generation amount via the Internet communication network 4. For transmission to the power adjustment device 9.
[0027]
As described above, in the present embodiment, the amounts of power generated in the first to third power stations 10a to 10c are adjusted by the power adjustment device 9, and the power requested by the first to fourth consumers 6a to 6d is adjusted. We are trying to generate electricity.
[0028]
FIG. 2 is a block diagram illustrating an example of the configuration of the power demand prediction device 8.
As shown in FIG. 2, the power demand prediction device 8 according to the present embodiment includes a data processing unit 21, an information storage unit 22, a cluster learning unit 23, and a cluster prediction unit 24.
[0029]
When the information regarding the power consumption in the second to fourth consumers 6b to 6d is input, the data processing unit 21 stores the information regarding the input power consumption in the information storage unit 22 as a time-series waveform. Things. In the following description, information on power consumption input to the power demand prediction device 8 is referred to as power consumption time-series information.
[0030]
The information storage unit 22 is configured by a storage medium for storing various information, and includes a learning target time-series information storage area 22a, a time-series information storage area before prediction start time 22b, a learning information storage area 22c, And a prediction result time-series information storage area 22d.
[0031]
The learning target time-series information storage area 22a is an area for storing power consumption time-series information input from the data processing unit 21.
The time series information storage area 22b before the prediction start time is an area for storing power consumption time series information for a predetermined time input from the data processing unit 21. More specifically, the time series information storage area 22b before the prediction start time stores power consumption time series information from 30 minutes before the current time to the current time.
[0032]
The learning information storage area 22c is an area for storing a learning result in the cluster learning device 23. The details of learning in the cluster learning device 23 will be described later.
[0033]
The prediction result time-series information storage area 22d is an area for storing the prediction result in the cluster predictor 24. The details of the prediction in the cluster predictor 24 will be described later.
[0034]
The cluster learning device 23 generates a plurality of clusters based on the power consumption time-series information in the second to fourth consumers 6b to 6d, classifies the generated clusters, and shows a relationship between the classified clusters. And has a cluster division unit 23a and a cluster classification unit 23b.
[0035]
Here, the processing performed by the cluster division unit 23a and the cluster classification unit 23b will be described with reference to FIGS.
FIG. 3 is a diagram for explaining processing when generating a plurality of clusters based on power consumption time-series information. FIG. 4 is a diagram for explaining a process of classifying the plurality of generated clusters and learning a relationship between the classified clusters.
[0036]
The cluster division unit 23a generates a plurality of rectangular clusters 32a to 32l based on the power consumption time series information 31 fetched from the learning target time series information storage area 22a, as shown in FIG.
[0037]
Here, the boundary condition of each of the clusters 32a to 32l is as follows (Equation 1).
_{(| Node k -f i |>} | node k -f i + 1 | AND | node k -f i |> ε) OR
_{(| Node k -f i |>} εAND | grad i -grad i + 1 |> η) ··· (1 type)
[0038]
In the above (Equation 1), k is a determined boundary point on the time axis. i is a boundary point after the boundary point k on the time axis. node _k is a value (height) of power consumption at the boundary point k. f _i is the power consumption values at the boundary point i (height).
[0039]
Further, grad _i is a value node _k of the power consumption at the boundary points k, is the slope of the power consumption values _{f i} at the boundary point i. grad _{i + 1} is a gradient between the power consumption value node _k at the boundary point _k and the power consumption value fi _{+ 1} at a point on the time axis obtained by adding 1 to the boundary point i.
[0040]
Further, epsilon is the power consumption value node _k, is acceptable for the absolute value of the difference between _{f i.} η is an allowable range for the absolute value of the difference between the gradients grad _i and grad _{i + 1} .
[0041]
Data that satisfies the above (Equation 1) expressed in this way is treated as components of the same cluster (rectangle). Further, the value of the power consumption in this same cluster (rectangle height), to the value node _k of the power consumption at the boundary point k, a mean value between the power consumption value f _i at the boundary point i.
[0042]
If the difference between the power consumption values (heights of rectangles) in adjacent clusters falls within a predetermined allowable range, the two adjacent clusters (rectangles) are combined into one cluster (rectangle). Thus, each cluster is generated.
[0043]
The cluster division unit 23a outputs the information CD1 on each of the clusters 32a to 32l created in this way to the cluster classification unit 23b. In the following description, information on the clusters 32a to 32l is referred to as cluster division information CD1.
[0044]
The cluster classification unit 23b classifies each of the clusters 32a to 32l based on the cluster division information CD1 output from the cluster division unit 23b.
[0045]
Specifically, the cluster classification unit 23b extracts each of the clusters 32a to 32l from the cluster classification information CD1. Then, the extracted clusters 32a to 32l are classified based on the following (Equation 2), and the same numbers are assigned to the clusters regarded as the same by the classification.
(J ^* dw <w (i) <(j + 1) ^* dw) AND (k ^* dh <h (i) <(k + 1) ^* dh) (2)
[0046]
In the above (Equation 2), dw is a parameter representing the time interval. dh is a parameter representing the step size of power consumption (height). h (i) is the value (height) of the power consumption of the clusters 32a to 32l. w (i) is the time width of the clusters 32a to 32l. j is a multiple that determines the range of the width. k is a multiple that determines the height range.
[0047]
In the case of the example shown in FIG. 3, the clusters 32a, 32g, and 32k are regarded as the same cluster and are numbered (1). Also, the clusters 32b, 32e, 32h, and 32j are regarded as the same cluster, and are numbered (2). Then, the clusters 32c and 32f are regarded as the same cluster and are numbered (3). Further, the clusters 32d, 32i, and 32l are regarded as the same cluster, and are numbered (4).
[0048]
Then, for each of the clusters 32a to 32l, the cluster classification unit 23b extracts what cluster will appear next and records it in the learning information storage area 22c.
[0049]
More specifically, in the case of the example shown in FIG. 3, the clusters appearing next to the clusters 32a and 32g with the number {1} are the clusters 32b and 32h with the number {2}, respectively. This is the cluster 32l to which the number {4} is given.
[0050]
Therefore, in the table 40 (see FIG. 4) created in the learning information storage area 22c, 2 is recorded in the square where the cluster number 1 and the next cluster number 2 intersect, and the cluster number 2 1 is recorded in the square where 1 intersects the next cluster number 4 交.
[0051]
That is, the clusters that may appear next to the clusters belonging to the number (1) are the clusters belonging to the number (2) and the clusters belonging to the number (4), and their appearance probabilities are 66% ] (= ２) and 33 [%] (= 1/3) can be determined from the table 40.
[0052]
Further, in FIG. 3, clusters 32b, 32e, 32h, and 32j that appear next to the clusters numbered (2) are clusters 32c and 32f that are respectively numbered (3) and numbered (4). The cluster 32i is assigned, and the cluster 32k is assigned a number (1).
[0053]
Therefore, in the table 40, 2 is recorded at the cell where the cluster number 2 and the next cluster number 3 intersect, and 1 is recorded at the cell where the cluster number 2 and the next cluster number 4 intersect. Is recorded, and 1 is recorded in the square where the cluster number (2) and the next cluster number (1) intersect.
[0054]
That is, the clusters that may appear next to the cluster belonging to the number (2) are the cluster belonging to the number (1), the cluster belonging to the number (3), and the cluster belonging to the number (4). Can be determined from the table 40 that the appearance probabilities are 25 [%] (= 1/4), 50 [%] (= 2/4), and 25 [%] (= 1/4), respectively. .
[0055]
In FIG. 3, the clusters appearing next to the clusters 32c and 32f with the number (3) are the cluster 32d with the number (4) and the cluster 32g with the number (1), respectively. is there.
[0056]
Accordingly, in the table 40, 1 is recorded at the cell where the cluster number 3 and the next cluster number 4 intersect, and at the cell where the cluster number 3 and the next cluster number 1 intersect. Record 1.
[0057]
That is, the clusters that may appear next to the clusters belonging to the number (3) are the clusters belonging to the number (4) and the clusters belonging to the number (1), and their appearance probabilities are 50 [%], respectively. Can be determined from the table 40.
[0058]
In FIG. 3, the clusters appearing next to the clusters 32d and 32i with the number (4) are the clusters 32e and 32j with the number (2), respectively. Therefore, in the table 40, 2 is recorded in the square where the cluster number 4 and the next cluster number 2 intersect.
[0059]
That is, it is possible to determine from the table 40 that the cluster that may appear next to the cluster belonging to the number (4) is the cluster belonging to the number (2). In the example shown in FIG. 3, since the cluster 32l to which the number (4) is assigned is the last cluster to appear, there is no cluster that appears next to this cluster 32l.
[0060]
The cluster predictor 24 is for predicting future power demand in the second to fourth consumers 6b to 6d, and as shown in FIG. 2, an initial cluster identification unit 24a and a cluster selection unit 24b. And
[0061]
Here, the processing performed by the initial cluster identification unit 24a and the cluster selection unit 24b will be described with reference to FIG.
FIG. 5 is a diagram for explaining a process when estimating power demand.
[0062]
Initial cluster identification unit 24a, the prediction starting point before the time-series information storage area 22b, a time t ₀ by reading the power time-series information 51 from the _(now 30 minutes before the time of the time t _C) to the current time t _C , To the cluster dividing unit 23a of the cluster learning unit 23. In the following description, it represents a predictive power consumption record information 51 and time-series information 51 of the power consumption from time t ₀ to the current time t _C.
[0063]
The cluster division unit 23a generates a cluster group 52 (clusters 52a to 52g) in the same manner as described above based on the input predicted power consumption actual information 51, and performs cluster division on the generated clusters 52a to 52g. The information CD2 is output to the cluster classification unit 23b.
[0064]
Cluster classification unit 23b extracts the previous clusters 52f of the cluster 52g including the current time t _C from the cluster dividing information CD2, to the cluster 52f described above extraction, numbered in the manner described above. In the following description, this number is referred to as an identification cluster number.
[0065]
Then, the number of the cluster having the highest possibility of appearing next to the cluster number for identification is searched for and extracted from the table 40 shown in FIG. In the following description, this number is referred to as an initial cluster number. Then, the identification cluster number and the initial cluster number are output to the initial cluster identification unit 24a of the cluster estimator 24.
[0066]
Initial cluster identification unit 24a, the initial cluster number to the value of the power consumption of the corresponding cluster (the height), the power consumption of the current time given by the prediction for the power consumption record information 51 (predicted start time) t _C It is determined whether or not the difference from the value is within a predetermined allowable range eta.
[0067]
The determination if the result is within the allowable range eta is the initial cluster number clusters at the identified current as a cluster 53 at time t _C, the cluster 53 to the granted by that number (the initial cluster number) CN The data is output to the cluster selection unit 24b. In the following description, representative of the initial clusters as necessary cluster at the current time t _C.
[0068]
On the other hand, as a result of the determination, when the difference between the values of the respective power consumptions is not within the allowable range eta, the cluster within the allowable range eta is selected from the clusters that may appear next to the identification cluster number. Then, the cluster having the highest appearance probability is extracted from the table 40 from the searched clusters. Then, the extracted clusters were identified currently as a cluster 53 at time t _C, and outputs the cluster 53 to the granted by that number (initial cluster number) CN cluster selection section 24b.
[0069]
The cluster selecting unit 24b searches the table 40 for a cluster having the highest appearance probability among clusters that may appear next to the cluster with the number CN.
[0070]
Then, the cluster having the highest appearance probability is searched from the table 40 among the clusters that may appear next to the searched cluster. Then, such a search process is repeatedly performed until the cluster at the time _tq is searched.
[0071]
In this way side by side retrieved cluster search order to generate a cluster group 54, calculates the power consumption time series information at the current time t _C or higher from the cluster 53 and cluster groups 54. Then, and stores the calculated power time-series information at the current time t _C after the prediction result time-series information storage area 22d.
[0072]
In the above description, when there are a plurality of clusters that appear next to the searched cluster and the appearance probabilities of the plurality of clusters are the same, there is a possibility that each of the plurality of clusters will appear next. A cluster is searched, and an appearance probability of the searched cluster is calculated.
[0073]
Then, the calculated occurrence probability is to be identified as a cluster a cluster 53 at the current time t _C was used when you search for a cluster having the largest.
[0074]
Next, the power demand prediction operation of the power demand prediction device 8 will be specifically described with reference to the flowchart of FIG.
[0075]
First, in the first step S1, the initial cluster identification unit 24a waits until a command for predicting power demand is input from an external device (not shown), and upon receiving the command, proceeds to step S2. At this time, the process may proceed to step S2 when a predetermined time comes without receiving an external command.
[0076]
Next, in step S2, the initial cluster identification unit 24a reads the predicted power consumption actual information 51 from the time-series information storage area 22b before the prediction start time and outputs it to the cluster division unit 23a.
[0077]
Next, in step S3, the cluster division unit 23a generates a cluster group 52 based on the predicted power consumption result information 51, and outputs the cluster division information CD2 to the cluster classification unit 23b.
[0078]
Next, in step S4, the cluster classification section 23b from the cluster division information CD2, extracting previous clusters 52f of the cluster 52g current including the time _{t C,} the extracted cluster 52f relative number (for identification Cluster number).
[0079]
Then, a cluster number (initial cluster number) having the highest possibility of appearing next to the identification cluster number is extracted from the table 40 of the learning information storage area 22c, and the identification cluster number and the initial cluster number are initialized. Output to the cluster identification unit 24a.
[0080]
Next, in step S5, the initial cluster identification unit 24a, the initial clusters and the value of power consumption of the corresponding cluster number, the power consumption values at the current time t _C being given by a prediction for the power consumption record information 51 Is within a predetermined allowable range eta.
[0081]
The result of this determination, the process proceeds to step S6 if it is within the allowable range eta, initial cluster identification unit 24a is to identify clusters corresponding to the initial cluster number as a cluster 53 at the current time t _C, to the cluster 53 The assigned number (the initial cluster number) CN is output to the cluster selector 24b, and the process proceeds to step S8.
[0082]
On the other hand, if it is not within the allowable range eta, the process proceeds to step S7, and the initial cluster identifying unit 24a is within the allowable range eta from among the clusters that may appear next to the identification cluster number. Search for a cluster. Then, the probability occurrence among the clusters and the search extracts the highest cluster from the table 40, the extracted number clusters corresponding to the identified current as a cluster 53 at time t _C. Then, the number (initial cluster number) CN assigned to the cluster 53 is output to the cluster selector 24b, and the process proceeds to step S8.
[0083]
Next, in step S8, the cluster selecting unit 24b extracts, from the table 40, the cluster having the highest appearance probability among the clusters that appear next to the cluster number CN, and based on the extracted clusters, Create information.
[0084]
Next, in step S9, the cluster selection unit 24b determines whether all the time-series waveforms of the power consumption up to the predetermined time _tq have been created. As a result of this determination, if not created, the operations of steps S8 to S9 are repeated until power consumption time-series information at a predetermined time _tq is created. In this case, it goes without saying that in step S8, the cluster having the highest appearance probability is extracted from the clusters that appear next to the previously searched cluster.
[0085]
FIG. 7 is a diagram showing the results of examining prediction accuracy when power demand is predicted by the above-described prediction method according to the present embodiment and when power demand is predicted by a conventional method.
[0086]
The reference numeral 71 shown in FIG. 7A indicates the actual value of power consumption in the second to fourth consumers 6b to 6d.
Reference numeral 72 illustrated in FIG. 7B indicates a result (power consumption time-series information) of predicting power demand by the prediction method according to the present embodiment.
[0087]
A reference numeral 73 shown in FIG. 7C indicates the power demand by a conventional method that averages the actual power consumption data from a predetermined time before the prediction start time to the prediction start time, and uses the averaged value as a prediction value. This is the result (power-consumption time-series information).
[0088]
FIG. 9 is a diagram illustrating an example in which a result obtained when power demand up to 20 hours ahead is predicted every hour using each of the above-described prediction methods and an error between actual data and the actual data are shown in a table format.
[0089]
In the characteristic diagram 90 of FIG. 9, the column of “the present embodiment” shows an error between the result of estimating the power demand by the method of the present embodiment and the actual data.
Further, the column of "conventional" shows an error between the result of averaging the actual data of the power consumption and estimating the power demand by the conventional method using the averaged value as a predicted value and the actual data.
[0090]
As is clear from FIGS. 7 and 9, when the power demand is predicted as in the present embodiment, the prediction accuracy is significantly improved as compared with the case where the power demand is predicted by the conventional method. I understand.
[0091]
Further, as shown in FIG. 8, an initial cluster identification unit processes without at 24a, power demand and identified arbitrarily by human intervention to be made on the basis of the cluster 81 in the prediction start time t _C, etc. the experience of the operator Is also conceivable.
However, when such a method is used, the prediction accuracy is improved as compared with the conventional method, but the prediction accuracy is lower than the method according to the present embodiment.
[0092]
That is, in the method of identifying the cluster 81 at the prediction start time t _C by human intervention and predicting the power demand, the cluster at the prediction start time t _C is identified without considering the transition of the power consumption before the start of the prediction. , The initial cluster 81 must be identified with the prediction start time t _C as a boundary.
[0093]
Therefore, when the true cluster at the prediction start time crosses the prediction start time t _C as in the cluster 53 of FIG. 5, the selection of the cluster at the prediction start time t _C is wrong, and Since the clusters to be obtained are also uncertain, the prediction accuracy decreases. In addition, the method of identifying the cluster 81 in the prediction start time t _C arbitrarily by human intervention, there is also a disadvantage that it is difficult to perform predictive processing continuously.
[0094]
In contrast, in the present embodiment, based on the transition of power consumption before the prediction starting. Thus to identify clusters 53 in the prediction start time t _C, probability statistics clusters 53 in the prediction start time t _C Identification can be performed with high accuracy based on the rationale. Therefore, the prediction accuracy of the power demand is dramatically improved, and the prediction process can be performed continuously. Further, even when the prediction is started, it is possible to identify the initial cluster 53 having the same probability of probability statistically, and it is possible to calculate the predicted time-series information with high accuracy and stability in an autonomous circulation calculation. it can.
[0095]
That is, as in the present embodiment, if the initial cluster 53 is identified based on the actual power consumption information in the vicinity before the prediction start time t _C , only the initial cluster 53 can be simply determined as identification information. not instantaneous power record information only, it is possible to consider the continuity of the past power consumption record information, it is possible to improve the prediction accuracy, further the prediction accuracy due to the difference of the predicted start time t _C Variation can be reduced as much as possible.
[0096]
As described above, in this embodiment, than cluster 52g including the predicted start time t _C out of a plurality of clusters 52 generated based on the power consumption time series information at a predetermined time prediction start time t _C before extracting the previous cluster 52f, the cluster 53 appearing in the next cluster 52f described above extracted and identified on the basis of past learning result, the cluster 53 identified above to the clusters in the prediction start time t _C Therefore, the cluster at the prediction start time t _C can be identified with high accuracy based on the probability statistical basis.
[0097]
Thus, it is possible to predict the evolution of the clusters in the prediction start time t _C after high precision. Therefore, non-stationarily changing power consumption, which is difficult to predict using linear programming, moving average, and frequency analysis, which are commonly used as future prediction methods for time-series information that constantly changes Can be predicted with high accuracy.
[0098]
In the present embodiment, a case has been described in which the power prediction is performed for each predetermined (unit) time. However, for a customer such as an office building in which the power demand is greatly different between the daytime and the nighttime, the present embodiment is used. It is preferable that the power demand is predicted on a daily basis by using the method of the embodiment, and the power demand is predicted in a combined manner with the above-described power demand prediction result for each predetermined (unit) time.
[0099]
Further, in the present embodiment, the power consumed by the customer group 2 has been described as an example of the non-stationarily changing time-series waveform, but the non-stationarily changing time-series waveform is not limited to this. Needless to say, for example, economic indicators such as stock prices may be used.
[0100]
Further, in the present embodiment, the future power demands of the second to fourth customers 6b to 6d are predicted, but the power demand prediction device 8 is configured to receive the power demand from the power company that is the first customer 6a. May be input from the first demand power measurement device 7a, and the input notification power may be added to the predicted result to predict the future trend of the power demand in the entire customer group 2. Good.
[0101]
Next, the power adjusting device 9 will be described.
FIG. 10 is a block diagram illustrating an example of the configuration of the power adjustment device 9.
10, a power adjustment device 9 includes a data processing unit 101, an information storage unit 102, a polygon division unit 103, a polygon definition unit 104, a variable polygon selection unit 105, a polygon area difference calculation unit 106, And a facility control unit 107.
[0102]
The data processing unit 101 acquires information on the future power demand in the second to fourth customers 6b to 6d predicted by the power demand prediction device 8, and information on the notification power in the first customer 6a. , And information on the current power generation amounts of the first to third power plants 10 a to 10 c via the Internet communication network 4 and to store them in the information storage unit 102.
[0103]
The information storage unit 102 is configured by a storage medium for storing various information, and has a control condition storage area 102a and an optimum control value storage area 102b.
[0104]
The control condition storage area 102a stores information input from the data processing unit 101, that is, information on the future power demand in the second to fourth consumers 6b to 6d predicted by the power demand prediction device 8, and This is an area for storing information on the notification power in the consumer 6a and information on the current power generation amount in the first to third power stations 10a to 10c, as well as storing control condition information.
[0105]
The control condition information is information indicating conditions when creating power control characteristics for generating power according to the demands of the first to fourth consumers 6a to 6d. The following five conditions are assumed. In the following conditions, the required power is the sum of the power predicted to be consumed by the second to fourth consumers 6b to 6d in the future and the notification power from the first customer 6a.
[0106]
First condition: control is performed near the required power in a section (current section) in which the power control characteristic is generated.
Second condition: The starting point of the power in the current section is the required power in the previous section.
Third condition: The end point of the power in the current section is the required power in the next section.
Fourth condition: In the current section, there is a time zone during which power cannot be increased or decreased due to the capacity of the first to third power plants 10a to 10c and the like.
Fifth condition: The gradient when increasing or decreasing the power is determined by a predetermined external factor such as the capacity of the first to third power plants 10a to 10c.
[0107]
The optimum control value storage area 102b is an area for storing power control characteristics created as described later.
[0108]
The polygon division unit 103 reads the control condition information from the control condition storage area 102, generates a power control characteristic line in the current section ignoring the optimality according to the condition, and calculates an inflection point of the generated power control characteristic line. As a boundary, a region surrounded by the power control characteristic line and a control target line including a required power value in the current section is divided to form a plurality of polygon structures.
[0109]
For example, when the power control characteristic line 111 ignoring the optimality as shown in FIG. 11 is generated, eleven polygonal structures P1, P11 to P15, and P21 to P25 are formed.
[0110]
The polygon definition unit 104 analyzes each of the polygon structures P1, P11 to P15, and P21 to P25 formed by the polygon division unit 103, and outputs attribute information for each of the polygon structures P1, P11 to P15, and P21 to P25. It is for generating and storing.
[0111]
FIG. 12 is a diagram for explaining attribute information of the polygon structures P1, P11 to P15, and P21 to P25 stored in the polygon definition unit 104.
[0112]
12, a polygon definition unit 104 includes a control constraint information storage area 121, a polygon variation type storage area 122, a polygon type storage area 123, an area polarity storage area 124, an inclination polarity storage area 125, an X-axis section It has a value storage area 126 and a Y-axis section value storage area 127.
[0113]
The control constraint information storage area 121 is an area for storing information on a polygon structure belonging to a time zone during which power cannot be increased or decreased. In the example of FIG. 11, since the power cannot be increased or decreased (fluctuated) in the first time zone a1 on the X axis (time axis) and the second time zone a2, the information on the polygon structures P14 and P22 is controlled. It is stored in the constraint information storage area 121.
[0114]
Note that such a time zone occurs because, for example, when increasing the power generation amount from the required power in the previous section to the required power in the next section, it is necessary to start or change the fuel system in the power generation plant. Therefore, it is necessary to secure a certain fuel system start-up time and a stable time.
[0115]
The polygon variation type storage area 122 is an area for storing polygon variation type information indicating whether the length in the X-axis direction can be changed for each of the polygon structures P1, P11 to P15, and P21 to P25. It is.
[0116]
In the example of FIG. 11, the length of the X-axis direction can be changed for the polygon structures P11 to P13, P15, P21, and P23 to P25, and the length of the X-axis direction cannot be changed. The polygon structures P1, P14, and P22 are stored in the polygon variation type storage area 122.
[0117]
The polygon type storage area 123 is an area for storing polygon type information indicating the type of each of the polygon structures P1, P11 to P15, and P21 to P25. In the example of FIG. 11, the types of the polygon structures P12, P15, and P21 are triangles. The types of the polygon structures P11, P14, P22, P24, and P25 are quadrangular. The types of the polygon structures P13 and P23 are quadrangles located above or below a triangle. Further, the type of the polygon structure P1 is a point. Therefore, these contents are stored in the polygon type storage area 123.
[0118]
In this embodiment, so that because of the aforementioned first condition that control in the vicinity of the requested power at the current period t _s ~t _f, sets a point of the type of the polygon structure. It is defined that the position of the polygon structure P1 whose type is a point can be changed.
[0119]
Area polarity storage area 124 is an area for storing information indicating the polarity of each polygon structure P11～P15, the area of P21~P25 in the case where the origin the value of the requested power at the current period _t s ~t _f is there. In the example of FIG. 11, since the polarities of the areas of the polygon structures P11 to P15 are negative and the polarities of the areas of the polygon structures P21 to P25 are positive, these are stored in the area polarity storage area 124.
[0120]
The inclination polarity storage area 125 is an area for storing the inclination polarities of the oblique sides of the triangular polygon structures P12, P15, P21, and P23. In the example of FIG. 11, since the polarities of the oblique sides (tilts) of the polygon structures P12, P15, P21, and P23 are all positive, these are stored in the tilt polarity storage area 125.
[0121]
FIG. 13 shows the relationship between the polarity of the area in the polygon structure stored in the area polarity storage area 124 and the relation between the polarity of the hypotenuse in the triangular polygon structure stored in the inclination polarity storage area 125.
[0122]
The X-axis section value storage area 126 stores the X-axis section values of the respective polygon structures P11 to P15 and P21 to P25 (the upper and lower limit values of the time to which the respective polygon structures P11 to P15 and P21 to P25 belong). It is an area for doing.
[0123]
The Y-axis section value storage area 127 stores the Y-axis section values of the respective polygon structures P11 to P15 and P21 to P25 (the upper and lower limits of the power to which the respective polygon structures P11 to P15 and P21 to P25 belong). It is an area for doing.
[0124]
As described above, in the present embodiment, each of the polygon structures P1 and P11 is defined by the control constraint, the polygon variation type, the polygon type, the area polarity, the inclination polarity, the X-axis section value, and the Y-axis section value. To P15 and P21 to P25 are defined.
[0125]
The variable polygon selection unit 105 selects two polygon structures having different polarities of area from the polygon structures P1, P11 to P15, and P21 to P25 defined by the polygon definition unit 105.
[0126]
At this time, the variable polygon selection unit 105 extracts conditions relating to selection of a polygon structure from the control condition storage area 102a, and selects a polygon structure according to the extracted conditions.
[0127]
In the example shown in FIG. 11, the first condition is extracted from the control condition storage area 102a, and a polygon structure is selected according to the first condition. More specifically, the polygon structure P1 is selected under the first condition that control is performed near the required power in the current section.
[0128]
For another polygon structure, if the sum of the areas of the non-variable fixed polygon structures P14 and P22 has a negative polarity, the variable polygon structures P21 and P23 to P25 having a positive area polarity. Is selected, the polygon structure P24 having the largest Y-axis value is selected.
[0129]
On the other hand, if the sum of the areas of the non-fluctuating fixed polygon structures P14 and P22 has a positive polarity, the absolute value of the Y-axis value among the fluctuating polygon structures P11 to P13 and P15 having a negative area polarity is obtained. The polygon structure P11 having the largest value is selected.
Then, the polygon information regarding the two polygon structures selected as described above is output to the polygon area difference calculation unit 106.
[0130]
The polygon area difference calculation unit 106 calculates the difference between the areas of the selected polygon structures based on the polygon information output from the variable polygon selection unit 105.
[0131]
In the example shown in FIG. 11, since the sum of the areas of the fixed polygon structures P14 and P22 indicates a positive polarity, the polygon structures P1 and P11 are the polygon structures to be subjected to the area difference calculation. However, here, for simplicity, the operation of the polygon area difference calculation unit 106 when the polygon structures P11 and P24 are targeted for the area difference calculation will be described.
[0132]
Polygon area calculation unit 106, the area polarity of the entire interval _(t f -t _S) calculates the value of the section X in Figure 11 so that the plus or minus 0.
[0133]
That is, a calculation that satisfies the following (formula 3) is performed.
G11 + G1 + Gminus = G24 + G2 + Gplus (Equation 3)
[0134]
In the above (Equation 3), G1 is a total value of integral values (areas) of the polygon structures P12 to P15. G2 is a total value of integral values (areas) of the polygon structures P21 and P23 to P25. G11 is an integral value (area) of the polygon structure P11 to be calculated. G24 is an integral value (area) of the polygon structure P24 to be calculated. Gminus is the integral value (area) of the actual power that is lower than the required power in the current section. Gplus is the integral value (area) of the actual power that exceeds the required power in the current section.
[0135]
Here, assuming that the value obtained by subtracting the required power in the previous section from the required power in the current section is D1, and the value obtained by subtracting the required power in the current section from the required power in the next section is D2, the following (Equation 4) and (Equation 4) (Equation 5) holds.
[0136]
G11 = D1 × X (Equation 4)
G24 = D2 × (Tt−X) = D2 × (Tt) −D2 × X (Equation 5)
[0137]
Therefore, the following (Equation 6) holds from the above (Equation 3) to (Equation 5).
Gminus + G1 + D1 × X = Gplus + G2 + D2 × (Tt) −D2 × X
(D1 + D2) × X = (Gplus + G2) − (Gminus + G1) + D2 × (Tt)
X = ((Gplus + G2)-(Gminus + G1) + D2 × (Tt)) / (D1 + D2) (Equation 6)
[0138]
In the above (Equation 6), the value of T is a time in the current section, and is a known value predetermined as control section information. The value of t is a known value that can be calculated from the inclination information of the triangular polygon structures P12, P15, P21, and P23 and the X-axis values of the fixed quadrangular polygon structures P14 and P22.
[0139]
D1 and D2 are known values that can be calculated from the required power in the preceding and following sections and the required power in the current section (control target section). Further, G1 and G2 are known values obtained by calculating by applying the areas of the respective polygon structures constituting the respective areas to the area formula. Further, Gplus and Gminus are excess / deficiency values of the actual power with respect to the required power in the current section (control target section), and are known values. Therefore, in the above (Equation 6), all information on the right side is known, and this model equation is established.
[0140]
If the value of the section X obtained by the above (Equation 6) does not satisfy the control constraint condition or deviates from the predetermined upper / lower limit value, the polygon area difference calculation unit 106 determines the current state of the polygon structure (( The power control characteristic line created by the polygon structure state reflecting the value of the section X obtained by the above (Equation 6)) is output to the variable polygon selection unit 105 as an initial power control characteristic line.
[0141]
Then, the changing polygon selection unit 105 selects again the combination of the polygon structures to be subjected to the difference calculation in accordance with the above-described control condition with respect to the initial power control characteristic line input from the polygon area difference calculation unit 106. Find the value of section X.
[0142]
Such a process is repeated until the optimality of the value of the section X is ensured, and the power control characteristic in a state where the optimality of the value of the section X is secured is stored in the optimal control value storage area 102b.
[0143]
The power generation equipment control unit 107 controls the first to third power stations 10a to 10c using the power control characteristics stored in the optimum control value storage area 102b and in which the optimality is ensured.
[0144]
On the other hand, when the required power in the previous section is the same as the required power in the next section, the power control characteristic is created as follows.
[0145]
FIG. 14 is a diagram illustrating an example of power control characteristics in the current section when there is no change in required power.
14, power conditioner 9, the amount of power to be generated in the current period (time t _{S ~} time t _f) is the power control to be equal to the amount of power in the current period when that caused the power required power The characteristic line 141 is set.
[0146]
That is, in the example shown in FIG. 14, the end time of the current segment a shortfall, from the current time t _C to the request electric power of the electric energy that is generated during the period from the start time t _S of the current period to the current time t _C t Recovery is performed between _f .
[0147]
More specifically, the power control characteristic line 141 is set so that the following (Equation 7) is satisfied.

[0148]
In the above (Equation 7), DS is a power amount insufficient for the required power, and SS is a power amount for recovering the insufficient power amount DS. The section T is a section (time) for recovering the shortage. The section t is a section (time) of the recovery section T in which power generation is performed with constant power (the power value x for recovering the shortage). In the following description, it represents the power shortage power value P _C at the current time t _C, represents the power value P _R for recovering the shortage and recovery power.
[0149]
Then, as shown in FIG. 14, the constant section power generation section t is represented by a known value as in the following (Equation 8) using the triangle ratio theorem.
T: t = (T × b / 2): (T × b / 2−x)
t = T−2 × (x / b) (8)
[0150]
In the above (8 formula), b is the rate at which increasing the power shortage P _c until recovery power P _R, when the gradient at that time and d [degree], is represented by the following (9 type) .
b = tan ... (Equation 9)
[0151]
Then, when the above equation (8) is substituted into the above equation (7), the following equation (10) is satisfied, and it is determined how much power should be increased in order to recover the shortage with respect to the required power. be able to.
x = [Tb ± {(T × b) ² −4DSb} ^1/2 ] / 2 (Equation 10)
[0152]
Next, the operation of the power adjustment device 9 will be described with reference to the flowchart of FIG.
[0153]
First, in the first step S11, the data processing unit 101 transmits, to the data processing unit 101, information on the notification power from the first customer 6a and information on the current power generation amount in the first to third power plants 10a to 10c via the Internet. Input via the communication network 4. In addition, information relating to future power demand in the second to fourth consumers 6b to 6d is input. Then, the various kinds of input information are stored in the control condition storage area 102a.
[0154]
Next, in step S12, the polygon division unit 103 determines whether or not the required power has changed in the next section based on the information on the notification power stored in the control condition storage area 102a. If the result of this determination is that there is a change in the required power, the flow proceeds to step S13. On the other hand, when there is no change in the required power, the operation of steps S13 to S15 is omitted, and the process proceeds to step S16 described later.
[0155]
Then, in step S13, the polygon dividing unit 103 generates a power control characteristic line in the current section ignoring the optimality according to the predetermined condition stored in the control condition storage area 102a, and generates the power control characteristic line of the generated power control characteristic line. A plurality of polygonal structures P1, P11 to P15, P21 to P25 with the inflection point as a boundary are formed.
[0156]
Next, in step S14, the polygon definition unit 104 analyzes the polygon structures P1, P11 to P15, and P21 to P25 formed in step S13, and analyzes the individual polygon structures P1, P11 to P15, and P21 to P25. Generate and store attribute information.
[0157]
Next, in step S15, the variable polygon selection unit 105 extracts conditions relating to selection of a polygon structure from the control condition storage area 102a. Then, based on the extracted conditions and the attribute information for the individual polygon structures P1, P11 to P15, and P21 to P25 generated in step S14, two polygon structures for which a difference calculation is performed are selected. The polygon information about the selected two polygon structures is output to the polygon area difference calculation unit 106.
[0158]
Next, in step S16, the polygon area difference calculation unit 106, based on the information on the polygon structures selected in step S15, such as the area polarity of the entire interval (t f _{-t s)} is ± 0 Calculate power control characteristics (conditions).
[0159]
If it is determined in step S12 that there is no change in the required power, in step S16, the polygon area difference calculation unit 106 calculates a power control characteristic (condition) that satisfies the above (Equation 9). I do.
[0160]
Next, in step S17, the polygon area difference calculation unit 106 determines whether or not the power control characteristic (condition) set in step S16 is within a predetermined appropriate range. If the result of this determination is not within the predetermined proper range, the operation returns to step S15, and the operations of steps S15 to S17 are repeated until the set conditions fall within the predetermined proper range. In this case, in step S15, the variable polygon selection unit 105 reselects a polygon structure so that at least one of the previously selected polygon structures is different.
[0161]
Next, in step S18, the power control characteristic calculated in step S16 is stored in the optimum control value storage area 102b as the power control characteristic with the optimality secured.
[0162]
As described above, in the present embodiment, when the required power in the next section is changed with respect to the required power in the previous section, the power control characteristic line 111 ignoring the optimality is created and the created A plurality of polygonal structures P1, P11 to P15, P21 to P25 are formed with the inflection point of the power control characteristic line 111 as a boundary, and attribute information for each of the formed polygonal structures and selection of the polygonal structure are performed. Based on the predetermined control conditions, two polygon structures are selected from the plurality of polygon structures, and the section X is determined so that the area difference between the plurality of polygon structures is plus or minus zero. Then, the polygon structures P11 and P24 are changed, and the power control characteristic line 111 is changed according to the change result. Sexual and creation, it is possible to perform the evaluation of the power control characteristics at the same time, the initial control characteristic line (control characteristic curve prepared ignoring optimality) need to be re-created several times by trial and error eliminated. Therefore, both the notification power from the power company as the first customer 6a and the power expected to be used in the plants as the second to fourth customers 6a to 6d are satisfied. It is possible to calculate the optimum power control characteristics at high speed and reliably, and it is possible to stably supply the power to the customer group 2 without excess or deficiency.
[0163]
In addition, when the inventor of the present application applies the power supply and demand adjustment system 1 of the present embodiment to an IPP (power wholesale supply) system, the above-described all processes for power supply can be performed without human intervention. It was confirmed that it could be completed within one minute.
[0164]
In addition, since the processes of steps S1 to S9 in FIG. 6 and steps S11 to S18 in FIG. 15 are repeatedly performed, the initial prediction accuracy of the section in the customer group 2 is of course high, As the end of the section in the consumer group 2 approaches, the power consumed by the second to fourth consumers 6b to 6d can be more accurately predicted. Therefore, even if the balance between the electric power demand in the consumer group 2 and the electric power supply in the electric power company 1 is broken in the middle of the section, it is possible to surely restore the balance by the end of the section. .
[0165]
It is preferable that the power control characteristic is calculated as in the present embodiment, so that the optimum power control characteristic can be obtained quickly and reliably. However, the method for calculating the power control characteristic is as described above. The method is not limited to this. For example, it is also possible to use a definite integral approximation method such as an interval quadrature method, Simpson's integral formula, or a Gaussian numerical integration method.
[0166]
Further, in the present embodiment, the case where the required power in the next section increases with respect to the required power in the previous section, but the case where the required power in the next section decreases with respect to the required power in the previous section is also described above. It goes without saying that a polygon structure can be defined in the same manner as described above to obtain the optimum power control characteristics.
[0167]
(Another embodiment of the present invention)
In order to operate various devices in order to realize the functions of the above-described embodiments, software programs for realizing the functions of the above-described embodiments are provided to an apparatus or a computer in a system connected to the various devices. The present invention also includes those which are supplied and executed by operating the above-described various devices according to a program stored in a computer (CPU or MPU) of the system or the apparatus.
[0168]
In this case, the program code itself of the software realizes the functions of the above-described embodiment, and the program code itself and a unit for supplying the program code to the computer, for example, storing the program code The recorded recording medium constitutes the present invention. As a recording medium for recording such a program code, for example, a flexible disk, a hard disk, an optical disk, a magneto-optical disk, a CD-ROM, a magnetic tape, a nonvolatile memory card, a ROM, and the like can be used.
[0169]
When the computer executes the supplied program code, not only the functions of the above-described embodiments are realized, but also the OS (operating system) or other application software running on the computer. Such a program code is also included in an embodiment of the present invention when the functions of the above-described embodiment are implemented in cooperation with the above.
[0170]
Further, after the supplied program code is stored in a memory provided in a function expansion board of a computer or a function expansion unit connected to the computer, a CPU provided in the function expansion board or the function expansion unit based on the instruction of the program code. The present invention also includes a case where the functions of the above-described embodiments are implemented by performing part or all of actual processing.
[0171]
【The invention's effect】
As described above, according to the present invention, the time at which the prediction is started is included from among a plurality of clusters generated based on the actual information of the time-series waveform in a predetermined time until the prediction of the time-series waveform is started. A cluster one cluster before is extracted, a cluster appearing next to the extracted cluster is identified based on the past learning result, and a cluster appearing after the identified cluster is regarded as the past learning result. Since the clusters are sequentially selected based on the probability, the cluster at the time when the prediction is started can be identified with high accuracy based on the probability statistical basis. This makes it possible to predict the transition of the cluster after the time when the prediction is started with high accuracy. Therefore, it is possible to predict the future trend of the time-series waveform that changes unsteadily with high accuracy.
[Brief description of the drawings]
FIG. 1 is a diagram showing an embodiment of the present invention, showing an embodiment of the present invention, and showing an example of a configuration of a power supply and demand adjustment system.
FIG. 2 is a block diagram illustrating an embodiment of the present invention and illustrating an example of a configuration of a power demand prediction device.
FIG. 3 is a diagram illustrating an embodiment of the present invention and illustrating a process when a plurality of clusters are generated.
FIG. 4 is a diagram illustrating an embodiment of the present invention and illustrating a process when learning a relationship between a plurality of clusters.
FIG. 5 is a diagram illustrating the embodiment of the present invention and describing a process when estimating power demand.
FIG. 6 is a flowchart illustrating the operation of the power demand prediction device according to the embodiment of the present invention.
FIG. 7 shows an embodiment of the present invention, and shows a result of investigating prediction accuracy when power demand is predicted by the prediction method according to the present embodiment and when power demand is predicted by another method; FIG.
FIG. 8 is a diagram illustrating the embodiment of the present invention and illustrating a method of predicting power demand by arbitrarily identifying a cluster at a prediction start time by human intervention.
FIG. 9 shows the embodiment of the present invention, and is a diagram of an example in which a result when power demand up to 20 hours ahead is predicted every hour and an error between actual data and the actual data are shown in a table format.
FIG. 10 is a block diagram illustrating an embodiment of the present invention and illustrating an example of a configuration of a power adjustment device.
FIG. 11 is a diagram illustrating the embodiment of the present invention and illustrating an example of power control characteristics when the required power is changed.
FIG. 12 is a diagram illustrating the embodiment of the present invention and illustrating attribute information of a polygon structure.
FIG. 13 shows the embodiment of the present invention and is a diagram illustrating a relationship between polarities of areas in a polygon structure and a relationship between polarities of hypotenuses in a triangular polygon structure.
FIG. 14 shows an embodiment of the present invention, and is a diagram illustrating an example of power control characteristics when there is no change in required power.
FIG. 15 is a flowchart illustrating an embodiment of the present invention and describing an operation of the power adjustment device.
[Explanation of symbols]
REFERENCE SIGNS LIST 1 power supply and demand adjustment system 2 customer group 3 power system network 4 internet communication network 5 electricity provider 6 customer 7 demand power measurement device 8 power demand prediction device 9 power adjustment device 10 power plant 11 supply power measurement device 21, 101 data Processing units 22 and 102 Information storage unit 22a Learning target time-series information storage area 22b Time-series information storage area before prediction start time 22c Learning information storage area 22d Prediction result time-series information storage area 23 Cluster learning unit 23a Cluster division unit 23b Cluster classification Unit 24a initial cluster identification unit 24b cluster selection unit 24 cluster predictor

Claims (10)

A time-series waveform prediction device for predicting a future trend of a time-series waveform given from a device to be predicted,
A cluster generating unit that generates a plurality of clusters based on actual information indicating a trend of the device to be predicted within a predetermined time until the start of the prediction; and a cluster generating unit that generates a plurality of clusters based on the cluster information. A cluster identification means for extracting a cluster immediately before the cluster including the time at which the prediction is started, and identifying a cluster appearing next to the extracted cluster based on a past learning result;
Cluster selecting means for sequentially selecting clusters appearing after the cluster identified by the cluster identifying means based on the past learning result;
A time-series waveform prediction unit that predicts a future trend of the time-series waveform based on the cluster identified by the cluster identification unit and the cluster selected by the cluster selection unit. Time series waveform prediction device. Cluster learning means for generating a plurality of clusters based on the time-series waveform result information given from the device to be predicted, and learning, for each of the plurality of generated clusters, what cluster appears next; The apparatus for predicting a time-series waveform according to claim 1, comprising: 3. The time-series waveform according to claim 1, wherein the prediction target device is a device that consumes power, and the time-series waveform is a waveform related to power consumed by the prediction target device. 4. Forecasting device. A time-series waveform prediction device according to claim 1 or 2,
An adjustment device for adjusting the operation of the device to be predicted so as to obtain a time-series waveform predicted by the time-series waveform prediction device. A time-series waveform prediction device according to claim 3,
A power adjusting device that adjusts a power generation operation of a power generation facility that supplies power to the device to be predicted such that a time-series waveform predicted by the time-series waveform predicting device is obtained. Adjustment system. The power adjusting device is configured to generate a power generation operation of the power generation equipment based on information on the amount of power requested to be supplied from a predetermined power company and information on the predicted power amount predicted by the time-series waveform prediction device. The power supply and demand adjustment system according to claim 5, wherein the power supply and demand adjustment system is adjusted. A time-series waveform prediction method for predicting a future trend of a time-series waveform given from a device to be predicted,
A cluster generation process of generating a plurality of clusters based on actual information indicating a trend of the device to be predicted within a predetermined time until the start of the prediction; A cluster identification process of extracting a cluster immediately before the cluster including the time at which the prediction started, and identifying a cluster appearing next to the extracted cluster based on a past learning result;
Cluster selection processing for sequentially selecting clusters appearing after the cluster identified by the cluster identification processing based on the past learning result;
Performing a prediction process of a time-series waveform for predicting a future trend of the time-series waveform based on the cluster identified by the cluster identification process and the cluster selected by the cluster selection process. Time series waveform prediction method. Cluster learning processing for generating a plurality of clusters based on the actual information of the time-series waveform given from the device to be predicted, and learning, for each of the plurality of generated clusters, what cluster appears next; 8. The method of predicting a time-series waveform according to claim 7, wherein: When predicting the future trend of the time-series waveform given from the device to be predicted,
A cluster generation process of generating a plurality of clusters based on actual information indicating a trend of the device to be predicted within a predetermined time until the start of the prediction; A cluster identification process of extracting a cluster immediately before the cluster including the time at which the prediction started, and identifying a cluster appearing next to the extracted cluster based on a past learning result;
Cluster selection processing for sequentially selecting clusters appearing after the cluster identified by the cluster identification processing based on the past learning result;
And causing the computer to execute a time-series waveform prediction process for predicting a future trend of the time-series waveform based on the cluster identified by the cluster identification process and the cluster selected by the cluster selection process. Characteristic computer program. A computer-readable recording medium on which the computer program according to claim 9 is recorded.

Images