JP3825173B2 - Power distribution system control system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To prevent a state from getting worse due to a calculation error, by judging the state of a current power distribution system before calculating power flow, and by preventing execution of calculation processing when the state is appropriate. SOLUTION: Before calculating power flow, system information and the detection data of a target system are inputted and stored (S1), and the presence or absence of a place that deviates from the adjustment tolerance and prescribed voltage width of a local device is confirmed according to the detection data (S2). Furthermore, it is judged whether there is no tolerance or a deviation place exists according to the detection data (S3), reactive power in each detection place is increased in the case of NO or it is judged whether the minimum voltage value is not reached or not (S4), and no adjustment is required for preventing calculation processing from being executed in the case of NO. On the other hand, in the case of YES, the degree of appropriateness of a current state for a target voltage and the reactive power is judged (S5), at the same time, it is judged whether the current state is appropriate or not (S6), and no adjustment is required for preventing the calculation processing from being executed in the appropriate case. In the inappropriate case, a target local device is determined (S7) and the calculation processing is started.

Description

【００３２】
ここで、ｍはバンク当たりの配電線数、ａ及びｂは重み係数、Ｖ（ｘ）は電圧の目的関数、Ｑ（ｘ）は無効電力の目的関数である。このうち、重み係数ａ、ｂは配電線の事情を勘案して配電線ごとに決定されるべき性格のものであるが、本実施形態では、検出情報から配電線の状態を把握して、重み係数ａ，ｂの値を図８のように自動的に選択する。
【００３３】
図８中の重み係数の選択は以下のようにしている。
（Ｉ）検出点で１箇所でも電圧逸脱があれば電圧逸脱時の重み係数とする。及び全検出点が「通常時状態」であっても、状態ＳＴ１及びＳＴ３の傾向にあれば電圧と無効電力を調整すべき状態と考えられるので、重み係数ａ及びｂとも０．５を採用する。また、電圧の目的関数Ｖ（ｘ）は次式のように定義した。
【００３４】
【数２】

【００３５】
ここで、Ｖｉはロ−カル機器２の動作前の各検出電圧、Ｖｃは中央装置１のＮＮで算出したロ−カル機器２が動作した場合の検出点の電圧予測値である。Ｖｏは配電線単位で決めるべき目標値（固定値）であるが、本実施形態では、各検出電圧が６３００〜６９００Ｖの範囲を逸脱している場合には６６００Ｖとし、逸脱していない場合には６４００Ｖとしている。また、無効電力の目的関数Ｑ（ｘ）は次式のように定義した。
【００３６】
【数３】

【００３７】
ここで、ｉは１配電線の検出数、Ｑｉはロ−カル機器２の動作前の各検出無効電力、Ｑｃは中央装置１のＮＮで算出したロ−カル機器２が動作した場合の検出点の無効電力予測値である。
【００３８】
次いで、制約条件について述べる。まず、電圧上下限値の制約条件では、配電線の送出電圧と検出箇所の電圧は各々次式を満たさなければならない。ここで、送出電圧はＬＲＴのタップ調整範囲内、検出箇所の電圧は需要家の端子電圧を１０１±６Ｖとする電圧でなければならない。
【００３９】
【数４】
Ｖｆｍｉｎ≦Ｖｆ≦Ｖｆｍａｘ
【数５】
Ｖｓｍｉｎ≦Ｖｓ≦Ｖｓｍａｘ
【００４０】
ここで、Ｖｆは高圧配電線（フィーダ）の電圧、Ｖｆｍａｘは当該電圧の最大値、Ｖｆｍｉｎは当該電圧の最小値である。また、Ｖｓは検出電圧、Ｖｓｍａｘは当該検出電圧の最大値、Ｖｓｍｉｎは当該検出電圧の最小値である。
【００４１】
次いで、ＩＣＲ９１の調整量の制約条件では、電圧調整量及び無効電力調整量は各々次式のように、ＩＣＲ９１の仕様から決まる調整能力の範囲を越えてはならない。
【００４２】
【数６】
Ｖ_ICRｍｉｎ≦Ｖ_ICR≦Ｖ_ICRｍａｘ
【数７】
Ｑ_ICRｍｉｎ≦Ｑ_ICR≦Ｑ_ICRｍａｘ
【００４３】
ここで、Ｖ_ICRはＩＣＲ９１の電圧調整値、Ｖ_ICRｍａｘはその上限値、Ｖ_ICRｍｉｎはその下限値である。また、Ｑ_ICRはＩＣＲ９１の無効電力（位相）調整値、Ｑ_ICRｍａｘはその上限値、Ｑ_ICRｍｉｎはその下限値である。
【００４４】
さらに、ＴＣＲ９２の調整量の制約条件では、無効電力調整量は次式のように、ＴＣＲ９２の仕様から決まる調整能力の範囲を越えてはならない。
【００４５】
【数８】
Ｑ_TCR≦Ｑ_TCRｍａｘ
【００４６】
ここで、Ｑ_TCRはＴＣＲ９２の無効電力調整値、Ｑ_TCRｍａｘはその上限値である。
【００４７】
次いで、ＧＡとＮＮ５００との組み合わせによる定式化について述べる。遺伝的アルゴリズム（ＧＡ）では、ストリング長が短く、ストリング操作によって死滅ストリングを生成しにくい方法が求められる。一方、ＩＣＲ９１は電圧調整と位相調整を行い、ＴＣＲ９２の無効電力調整は実際的にはＴＣＲ９１の設置点の力率角を調整することから、位相調整を行うことになるので、本実施形態では、上記条件を満足するストリング表現として、以下に示す電圧調整量のストリング表現と位相調整量のストリング表現を用いる。
【００４８】
電圧調整量のストリング表現において、ＩＣＲ９１の制御可能電圧調整量は±３３０Ｖであることより、最小制御可能電圧幅を５Ｖ／ステップとして、５Ｖを１ビットとして表現すると、３３０Ｖは２進数で１００００１０となり７ビットで表される。ここで、±の符号を１ビットとして考えると調整量の変化を図９のようにストリング表現できる。この場合、電圧調整幅は最大３３０Ｖであるため、これを越えるようなストリングは致死遺伝子とする。
【００４９】
また、位相調整量のストリング表現において、ＩＣＲ９１とＴＣＲ９２の制御可能位相調整量は±５度とし、最小制御可能位相幅を０．１度／ステップとすれば、０．１度を１ビットとして表現すると、５度は２進数で１１００１０となり６ビットで表される。ここで、±の符号を１ビットとして考えると調整量の変化を図１０のようにストリング表現できる。この場合、位相調整幅は最大５度であるため、これを越えるようなストリングは致死遺伝子とする。
【００５０】
さらに、ストリング長について説明する。ストリングの生成は配電線単位で実施する。この場合、当該配電線のロ−カル機器２の台数によりストリング長が異なることになる。このため、ロ−カル機器２の最大台数を設定しておき、対象となる機器の種類及び台数によってストリング長を変える方法を採用した。図１１にＩＣＲ３台、ＴＣＲ３台の場合のストリング長の例を示す。
【００５１】
初期ストリングの生成方法においては、各ロ−カル機器２の調整量の各ビットに対して、ランダムに２進数の０又は１を割り振る。調整幅を超えるような致死遺伝子が生成された場合には、再度割り振りを行いＮ個のストリングを生成する。なお、現在運用されている調整量は最適解ではないものの、準最適解と考えられることから初期ストリング集団の１つに入れておくこととした。
【００５２】
次いで、適用するニューラルネットワーク（ＮＮ）５００の構成について説明する。ストリング（調整量）情報とＧＡでは対象としない情報（各検出点での有効電力及び無効電力、送出電圧、送出位相）をＮＮ５００の処理部へ送り、あらかじめ対象配電線の環境を学習（覚え）させたネットワ−クを持つＮＮ５００の処理部で確率的潮流計算（通常の潮流計算と同レベルの出力を短時間に推論する方法）を実施し、その演算結果（各検出点での電圧及び無効電力）をＧＡ処理部に出力させることとしている。ＮＮ５００には学習機能を有するバックプロパゲ−ションモデルを用い、中間層数は１層とし配電線ごとに１つのネットワ−クを構成し、中間ユニット数は後述の検証結果により決定することとした。ＩＣＲ、ＴＣＲ各３台を設置した８区間配電線でのＮＮ５００の構成例を図１５に示す。なお、本実施形態では、中間層２００を１層としているが、本発明はこれに限らず、複数層設けてもよい。
【００５３】
図１５のＮＮ５００において、ＮＮ５００は、入力層１００と中間層２００と出力層３００とを備えて構成される。入力層１００は、２７個の入力層ユニット１０１−１乃至１０１−２７を備え、中間層２００は、複数Ｍ個の中間層ユニット２０１−１乃至２０１−Ｍを備え、出力層３００は、２８個の出力層ユニット３０１−１乃至３０１−２８を備える。ここで、各入力層ユニット１０１−１乃至１０１−２７には、下記のデータが入力される。
（ａ）ＩＣＲ_V1乃至ＩＣＲ_V3：３台のＩＣＲ９１の電圧調整量、
（ｂ）ＩＣＲ_Q1乃至ＩＣＲ_Q3：３台のＩＣＲ９１の無効電力（位相）調整量、
（ｃ）ＴＣＲ_Q1乃至ＴＣＲ_Q3：３台のＩＣＲ９１の無効電力（位相）調整量、
（ｄ）Ｐ₁乃至Ｐ₈：各検出装置３間の区間における負荷有効電力（固定値）、
（ｅ）Ｑ₁乃至Ｑ₈：各検出装置３間の区間における負荷無効電力（固定値）、
（ｆ）Ｖｓ：配電用変電所からの送出電圧（中央装置により検出される。）、及び
（ｇ）θｓ：配電用変電所からの送出位相（中央装置により検出される。）。
【００５４】
次いで、入力層ユニット１０１−１乃至１０１−１７はそれぞれ入力されたデータをＭ分配して各中間層ユニット２０１−１乃至２０１−Ｍに出力する。そして、中間層ユニット２０１−１乃至２０１−Ｍはそれぞれ、入力される２７個のデータに対して所定の重み係数で線形結合で重み付けされたデータを２８分配して出力層ユニット３０１−１乃至３０１−２８に出力する。さらに、出力層ユニット３０１−１乃至３０１−２８はそれぞれ入力されるＭ個のデータを加算して下記のデータを出力する。
（ａ）Ｖ_C1乃至Ｖ_C14：各検出位置における電圧値、及び
（ｂ）Ｑ_C1乃至Ｑ_C14：各検出位置における無効電力値。
【００５５】
ここで、各中間層ユニット２０１−１乃至２０１−Ｍの出力ｙは、応答関数をａ（ｘ）、リンク荷重をｗ_iとすると次式で表わすことができる。
【００５６】
【数９】

【００５７】
次いで、適応度関数について述べる。ＮＮ５００からの出力データ（各検出位置における電圧値及び無効電力値）に基づいて、数１を用いて、目的関数ｆ（ｘ）を求め、ＧＡでの評価は次式による適応度関数ｇ（ｘ）を用いて行う。
【００５８】
【数１０】
ｇ（ｘ）＝１／ｆ（ｘ）
【００５９】
さらに、ＧＡでは、世代を重ねるにつれ評価値の上昇が飽和する特性を持っているため、次のいずれかの条件が成立した場合に終了とする。
（条件１）設定された計算時間に達した場合。
（条件２）最大評価値が規定する世代数にわたって更新されなかった場合。
【００６０】
また、ＧＡの選択淘汰処理においては、前世代集団中で適応度の高いストリングを次世代に規定個数残し、そのうちの数個はエリ−トとして突然変異禁止のフラグを立てる。この場合、適応度の高いストリングが無条件で次世代に残る反面、エリ−トストリングが集団中に急速に広まり局所解に陥る可能性があるので、後述の突然変異処理でこれを回避している。すなわち、エリート保存戦略をとり、集団中でもっとも適応度の高い個体をそのまま次世代に残す方法を採用している。適応度の高い個体が無条件で次の世代に残る反面、そのエリート個体が集団中に急速に広まり局所解に陥る恐れが出てくる。実際のエリート保存戦略は、エリート個体に対する突然変異は禁止であるが。本実施形態では、任意に設定された個体以外は突然変異を可能にする。
【００６１】
また、ＧＡの交叉処理においては、選択されたストリングの中からランダムに２つのストリング選び、電圧及び位相調整量ごとに交叉点をランダムに決め、各々１点交叉を行い生成されたストリングのうち致死遺伝子を除外して規定個数のストリングを生成する。交叉の例を図１２に示す。また、図１４は、ＧＡにおけるストリングの多点交叉の一例を示す図であり、図１４に示すように、各ローカル装置２の各調整量ごとに１点交叉を行う。
【００６２】
さらに、ＧＡの突然変異処理において、突然変異は各ストリング位置の数値の変換である。このため、電圧及び位相調整量ごとに突然変異を起こす任意のロ−カル機器２をランダムに選び、調整量の任意のビットをビット反転させる方法をとる。図１３に突然変異の例を示す。
【００６３】
次いで、遺伝的アルゴリズム（ＧＡ）の適用方法について説明する。以下のような入力データに対して、ＧＡにより目的関数ｆ（ｘ）が最小となるようなローカル装置２の調整量を算出し（ＮＮ５００は、ＧＡ処理内部の適応度計算処理で使用する。）最適な調整量を出力する。
（Ａ）ＧＡの入力データ
（ａ）対象ローカル装置２：最適化を行う対象ローカル装置２である。
但し、ＩＣＲ９１の場合，電圧調整量のみ最適化するのか、位相調整も行うのかの情報を含む。
（ｂ）目的関数の重み係数：目的関数ｆ（ｘ）の重み係数ａ，ｂである。
（ｃ）目標電圧：目的関数ｆ（ｘ）を求める際の目標電圧である。
ここで、以下の入力情報は、同一時間におけるデータでなければならない。
（ａ）各ローカル装置２の調整量の初期値：各ローカル装置２の現在における調整量である。
（ｂ）インピーダンスマップ：各検出装置３間の区間における現在のインピーダンスである。
（ｃ）第１の検出装置３のＶ、Ｉ、Ｑ：現在の第１の検出装置３における電流、電圧、無効電力の測定値である。
（２）出力データ
（ａ）最適化を行ったローカル装置２の調整量：入力されたデータにおける最適な各ローカル装置２に対する調整量である。
【００６４】
本実施形態のＧＡにおいては、以下のように先験情報を利用している。規定電圧を逸脱していない場合、現在運用されている各ローカル装置２の調整量は最適解ではないものの、ある程度良い解であるといえ、最適解もその調整量の付近にあるはずである。これを利用して、初期生成個体に現在運用されている各ローカル装置２の調整量を反映させて探索空間を絞り込み計算時間を短縮する方法を用いる。なお、規定電圧を逸脱している場合には以下の手法を用いない。
【００６５】
ここで、初期生成方法（初期状態の反映）においては、初期生成個体のうちの１つに、調整量の初期値（現在運用されている調整量）をそのまま反映した個体を入れる。現在運用されている調整量は最適解ではないものの、ある程度良い解であるといえるため（規定電圧幅を逸脱していない場合）、初期生成に反映させることにより効率の良い探索をすることをできるという利点を有する。
【００６６】
図５は、図４のサブルーチンである適応度計算処理（ステップＳ１２）を示すフローチャートである。図５に示すように、まず、ステップＳ２１においてパラメータｎに１を代入し、ステップＳ２２においてｎ番目の個体の遺伝子内容から各ローカル装置２の調整量を求める。次いで、ステップＳ２３においてその調整量における各検出位置での電圧値Ｖ及び無効電力値Ｑをニューラルネットワーク５００を用いて求める。そして、ステップＳ２４において目的関数ｆ（ｘ）の算出するとともに、適応度関数を算出する。さらに、ステップＳ２５においてパラメータｎが個体数に達したか否かが判断され、達していないときは（ＮＯ）ステップＳ２６においてパラメータｎを１だけインクリメントしてステップＳ２２に戻って上記の処理を繰り返す。一方、ステップＳ２５でｎが個体数に達しているときは（ＹＥＳ）元のメインルーチンに戻る。
【００６７】
図６は、図４のサブルーチンである選択淘汰処理（ステップＳ１４）を示フローチャートである。図６に示すように、選択淘汰処理においては、ステップＳ３１において前世代の集団の中から、適応度の高い個体を規定個選択する。次いで、ステップＳ３２においてエリート個体として、適応度の高い数個体に対し、突然変異禁止のフラグを立てる。そして、元のメインルーチンに戻る。
【００６８】
ＧＡの終了判定においては、上記目的関数ｆ（ｘ）の逆数ｇ（ｘ）＝１／ｆ（ｘ）を適応度を示す目的関数として用い、予め設定された計算終了時間に達したら計算を終了する。また以下の条件を終了時間前に満たした場合、その時点で終了する。ＧＡで計算をすると、世代を重ねるにつれて指数関数に反比例するように目的関数の評価値の上昇の度合いが鈍くなってゆき、最終的には何世代計算させようとも評価値が変化しなくなってくる。これを利用して、「最大評価値が規定世代更新されなかったら」、最適解に達したものとして計算を終了する。なお、最適解が規定電圧幅を越えていても致死遺伝子とはしない。計算結果として、対象ローカル装置２の調整量を出力する。
【００６９】
次いで、初期のＮＮ５００の学習方法について説明する。まず、インピーダンスマップの作成を行う。図１６は、図１の配電系統制御システムで用いるある負荷状態におけるインピーダンスマップの生成処理を示すブロック図である。図１６に示すように、当該生成処理では、検出データと、各ローカル装置２の調整量から現在の負荷状態におけるインピーダンスマップを作成する。ここで、１日、数回の検出を行い、数パターンかのインピーダンスマップを作成する。次いで、各インピーダンスマップにおける学習データを作成する。図１７は、図１の配電系統制御システムで用いる各調整量におけるセンサにおける電圧値Ｖ及び無効電力値Ｑの計算処理を示すブロック図である。図１７に示すように、各インピーダンスマップに対して、各ローカル装置２を一様になるようなパターンで変化させたときの、各検出装置３における電圧値Ｖ及び無効電力値ＱをＮＮ５００で構成された従来技術の潮流シミュレータで求める。この電圧値Ｖ及び無効電力値Ｑを学習データとして用いる。以上説明したように、調整量が一様になるようパターンを設定し、出力された電圧値Ｖ及び無効電力値Ｑを学習データとする。
【００７０】
従来技術の潮流シミュレータは、ＮＮ５００と同様の入力データに基づいて、公知の電気回路の解析法であるアドミッタンス行列に基づくニュートンラプソン法を用いて潮流計算を行って、ＮＮ５００と同様の出力データを計算する装置である。この潮流シミュレータは、ＮＮ５００の学習のときに用いるのみならず、図４のステップＳ１７の処理において用いる。
【００７１】
図１８は、図１５のニューラルネットワーク（ＮＮ）５００の学習方法を示すブロック図である。上記で作成された学習データを基に学習を行う。なお、ＮＮ５００の学習は高圧配電線単位で別々に行うものとする。また、ＮＮ５００の追加学習では、初期学習で学習させたＮＮ５００だけでは、負荷の変動などに対処できないため最新の学習データをＮＮ５００に反映させる必要がある。従って、追加学習を行っている（図７のステップＳ５８及びＳ６６参照。）。
【００７２】
図１９は、図１５のニューラルネットワーク（ＮＮ）５００への学習データの取り入れ方法を示すブロック図である。図１９に示すように、入力層１００の各入力層ユニット１０１−１乃至１０１−２７に接続されたシフトレジスタ４０１と、出力層３００の各出力層ユニット３０１−１乃至３０１−２８に接続されたシフトレジスタ４０２を用いて、最新の学習データ（作成方法は初期学習と同じである。）を取り入れ、逐次ネットワークを学習させる方法を用いる。このとき、時間が経つにつれて学習データが雪だるま式に増えていくので、古いデータから順番に学習データから破棄するようにする。ここで、学習データ数は一定に保つ。なお、新規の学習データは、ローカル装置２の調整量を変化させた５〜１０データで１組となっている。また、各入力データと、教師データは一対になっている。
【００７３】
次いで、規定電圧幅を短期間内に逸脱することの確認処理について説明する。この処理では、以下のように、電圧逸脱の確認を行う。
（１）各高圧配電線毎にインピーダンスマップ（各検出装置３間の線路インピーダンス。）を確認する。
（２）各高圧配電線毎に、送出電圧（予測値）、線路インピーダンス、及び負荷電流（予測値）より、各検出位置での電圧を図２０に示すように算出する。
【００７４】
（計算例）
（ａ）線路インピーダンスのみの場合
【数１１】
Ｖ₂＝Ｖ₁：アドレス１の検出位置での電圧予測値
（ｂ）負荷有り（線路インピーダンス：Ｚ１）の場合
【数１２】
Ｖ₂＝Ｖ₁−Ｉ₁Ｚ₁
【数１３】
Ｚ₁＝（ｒ₁₂＋Ｘ₁₂）^1/2：アドレス１の検出位置での電圧予測値
【００７５】
（３）上記計算により算出した各検出位置での電圧値から、規定電圧幅を逸脱しているものはないか確認する。
（４）全アドレス（全検出位置）確認後、逸脱箇所有りと認識した高圧配電線がある場合、その対象高圧配電線とアドレス、及び予測値（電圧、電流、インピーダンス：負荷Ｚは無効電力の予測値と合わせて算出する。）をもって、マクロシミュレーション処理（図４のステップＳ１７）に進む。
【００７６】
次いで、図３のステップＳ７の対象ローカル装置２又はＬＲＴ９６の選定処理について説明する。この処理は（ａ）検出データ及び予測より算出したデータから電圧、無効電力値をチェックした結果、ローカル装置２の調整量を算出する場合と、（ｂ）ローカル装置２の調整量の算出処理において、ローカル装置２の選定見直し要の場合の２通りが考えられる。
（Ａ）初期設定の場合：初期設定においても、電圧逸脱是正の場合とその他（裕度無し、又は無効電力増加、又は電圧下限）の場合の２通りが考えられる。
（Ｉ）電圧逸脱是正の場合は、朝の就業開始、昼休み、夕方の就業停止等に着目して、そのような比較的短時間の負荷変動に対しても対応することを考えて早急に是正するために、予め対象とするローカル装置２を限定する。まず、電圧逸脱点をアドレスより認識し、その逸脱点より上位にある電圧調整機能を持つＩＣＲ９１を対象ローカル装置２として各高圧配電線毎に限定する。電圧逸脱点より上位にＩＣＲ９１がある場合、その高圧配電線についてはＬＲＴ９６選定として次の処理に進む。
（II）電圧逸脱是正以外の場合は、各高圧配電線毎にすべてのローカル装置２を対象として認識する。
【００７７】
（Ｂ）初期設定以外の場合
（ａ）この処理は、一度、電圧逸脱是正のため、あるローカル装置２を限定して調整量を求めて電圧逸脱是正を行おうとしたが、限定したローカル装置２のみでは是正できないため、ローカル装置２を再選定する必要がある場合で、電圧逸脱是正の場合に限定される。
（ｂ）電圧逸脱是正のため、限定しているＩＣＲ９１より上位にあるＩＣＲ９１を対象ローカル装置２として当該高圧配電線に追加する。現在限定しているＩＣＲ９１より上位にＩＣＲ９１がない場合、その高圧配電線はＬＲＴ９６選定として次の処理に進む。
【００７８】
次いで、目的関数が最小となるローカル装置２の調整量を求める処理について説明する。この処理は、前処理から各高圧配電線毎に目的関数及び対象ローカル装置２が決定され、その条件下で、目的関数が最小となるローカル装置２の調整量を求めるものである。
（１）各対象高圧配電線毎に目的関数を最小とするローカル装置２の調整量を求める。
（２）対象ローカル装置２としてＬＲＴ９６が選定されていない場合は、対象ローカル装置２の調整量をランダムに設定し、その調整量での各検出位置の電圧と無効電力をＮＮ５００を用いて求める。上記値より目的関数ｆ（ｘ）を算出し、終了条件を満足するまでローカル装置２の調整量の更新を行って目的関数ｆ（ｘ）を算出する。終了条件を満足した時点で、その時のローカル装置調整量をｆ（ｘ）をＧＡによる最小とする解とする。
（３）目的関数ｆ（ｘ）を最小とするローカル装置２の調整量が求まった時点で、その時の各検出位置での電圧を確認し、規定電圧幅を逸脱している部分がないかチェックする。ここで電圧逸脱がない場合は、対象高圧配電線における対象ローカル装置２の調整量を決定する。電圧逸脱がある場合は、対象ローカル装置２の見直しを行うためローカル装置２の選定処理に進む。
（４）対象ローカル装置２としてＬＲＴ９６が選定されている場合は、補正電圧を算出し、その補正電圧で補正した場合の送出電圧を求め、送出電圧が現状よりどれだけ変化するか認識する。それより、そのＬＲＴ９６に接続されている高圧配電線の各検出電圧がどう変化するか算出する。そこで、電圧逸脱箇所が新たに発生しないか確認する。発生しない場合は、制約条件（送出電圧）を変更し、マクロシミュレーション処理を行う。発生する場合は、現状問題の無い高圧配電線を悪くすることになるため、ＬＲＴ９６の補正は行わず、制御不能箇所（電圧逸脱を回避できない区間）として認識して、上記（１）及び（２）の処理を実施する。そして、電圧逸脱区間が制御不能箇所であれば、そのまま対象ローカル装置２の調整量を決定する。
【００７９】
さらに、簡素化した系統で潮流計算し、求めた調整量でローカル装置２が動作した場合の各検出位置での電圧、無効電力を求め、そのデータは改善されているか否かを確認する処理（図４のステップＳ１７）について説明する。
（１）この処理では、前処理で決定した対象高圧配電線毎のローカル装置２の調整量と、先に作成している対象高圧配電線の系統図（インピーダンスマップ）をもとに潮流計算を行い、各検出位置での電圧及び無効電力を算出する。
（２）この結果と検出データより、下記の目的関数（改善率の平均）を算出し、電圧及び無効電力の改善率が５０％以下の場合データは改善されていると判断する。改善率がどちらか一方でも５０％を越えている場合は、改善率の悪い方を認識して、再度調整量の計算をやり直す。
（Ｉ）電圧の目的関数ｆｖ（ｘ）：
【００８０】
【数１４】

【００８１】
ここで、
Ｖｉ：各検出位置での検出データ（電圧）、
Ｖｆ：潮流計算で算出した各検出位置での電圧、
Ｖｏ：目標電圧、
ｎ：検出数。
（II）無効電力の目的関数ｆｑ（ｘ）：
【００８２】
【数１５】

【００８３】
ここで、
Ｑｉ：各検出位置での検出データ（無効電力値）、
Ｑｆ：潮流計算で算出した各検出位置での無効電力値、
Ｑｏ：目標無効電力、
ｎ：検出数。
【００８４】
図２１は、図１の配電系統制御システムで用いる平均化処理における特異データの処理方法を示すグラフである。本実施形態の配電系統制御システムにおいて、中央装置１は各ローカル装置２及び検出装置３のセンサ（以下、センサという。）からデータを取り込み、その内容を把握しながら系統の最適化制御を行うが、高圧配電線Ｆ１，Ｆ２に点在するローカル装置２及び検出装置３から同時にデータを入力することができないため、入力時に時間差が発生する。この時間差をどう扱うかについて説明する。具体的には、中央装置１からのポーリングにて各センサのデータを収集する際、最初に収集した検出データと、最後に収集した検出データとでは、時間差があるため、単純に同一として扱うことはできない。従って、本実施形態では、この検出データを同時間に収集したと見なせるよう時間的な平均化処理を行う。例えば、収集した検出データを同時間のデータと見なすために、収集時間誤差を１％以下にすると仮定した場合、ポーリング周期の１００倍の時間で、各検出機器にて検出データを平均し、その平均値のデータを収集することにより、同時間（時間誤差１％以内）の検出データとして扱う。ただし、検出したデータが、前回検出データより一定幅を越えた場合は、特異データとし、そのデータは破棄する。
【００８５】
さらに、図３のステップＳ５及びＳ６の処理について説明する。遠隔監視機能より最適化処理要求があった場合にも、ＮＮ５００の誤差の影響で現状より悪くする可能性があるため、以下の適正条件（電圧の適正条件と無効電力の適正条件）が成立する場合には、現状がほぼ最適であると判断し、最適化処理、すなわち、図３のステップＳ７以降の処理を実行しない（ステップＳ６でＮＯ）ことを確認する。
【００８６】
【数１６】

【数１７】

【００８７】
ここで、
６０：ＮＮ５００が有する各検出点での電圧の誤差（Ｖ）、
１００：ＮＮ５００が有する各検出点での無効電力の誤差（ｋＶａｒ）、
Ｖｉ：現状の各検出位置での電圧Ｖｏ、
Ｖｏ：目標電圧，
Ｑｉ：現状の各検出位置での無効電力、
ｎ：検出数。
すなわち、数１６及び数１７によって示される適正条件は、各検出装置３によって検出された現在の上記各検出位置における電圧値又は無効電力値と、その所定の目標値との差の総和が所定のしきい値以上であることである。
【００８８】
【実施例】
以上のように構成された図１の配電系統制御システムの動作の検証を行うため、本発明者は以下のシミュレーションを行った。
【００８９】
まず、ＮＮ５００による潮流計算機能の単体検証について説明する。ＮＮ５００ではそのネットワ−クの構造により、目的関数の算出精度が大きく左右されるため、ネットワ−クの構造をいかに決定するかが問題となる。このため、中間ユニット数をパラメ−タとして構造についての検討及び検証を実施した。
【００９０】
図２２は、図１の配電系統制御システムのシミュレーションを行う検証用配電系統の構成を示すブロック図である。検証用配電系統は図２２に示すとおりで、ユニット数は入力層３５、出力層２４である。中間ユニット数は３０、４０、５０の３種類とした。また、学習デ−タと評価デ−タについては、送出電圧は６９００Ｖとし、各区間負荷は２５〜５０Ａの範囲で１Ａステップ、負荷力率は０．４〜０．９のの範囲で０．０１ステップ、ＩＣＲ９１の電圧調整量は±３３０Ｖの範囲で５Ｖステップ、ＩＣＲ９１及びＴＣＲ９２の位相調整量は±５度の範囲で０．１度ステップで各々乱数による値を求めた。これらのデ−タを入力値として従来の潮流計算を実行し、得られた各検出位置での電圧及び無効電力値を学習デ−タとし、学習回数は１００、２００、４００の３パタ−ンとした。また、評価デ−タ（従来の潮流計算で求めたデ−タ）も学習デ−タと同様の方法で生成した。
【００９１】
ＮＮ５００の評価は、従来の潮流計算結果による評価デ−タを真値とし、ＮＮ５００で算出した各検出位置での電圧及び無効電力と従来の潮流計算結果による評価デ−タとの誤差によった。電圧の許容誤差は６６００Ｖの１％相当の６０Ｖであり、無効電力の許容誤差は評価デ−タの作成条件下で発生する最大無効電力の１０％相当の３００ｋｖａｒとした。
【００９２】
また、ＮＮ５００の学習打ち切り誤差の決定においては、ＮＮ５００による潮流計算機能の単体検証に先駆け、学習打ち切り誤差を事前に決定するため、中間ユニット数３０、学習デ−タ数１００での逐次学習法による学習開始後の平均二乗誤差を求めた。この場合、平均二乗誤差は次式に示す定義とした。
【００９３】
【数１８】

【００９４】
図２３は、図１の配電系統制御システムのシミュレーション結果におけるニューラルネットワーク（ＮＮ）５００の学習曲線を示すグラフである。学習開始後の平均二乗誤差は、図２３に示すとおりで、学習打ち切り誤差は進行が緩慢となる１５分相当（学習回数６０００回）の０．０００２に設定した。
【００９５】
次いで、単体検証結果について説明する、ＮＮ５００と従来の潮流計算に異なるデ−タを９６個入力し、それぞれが算出した各検出位置での電圧及び無効電力の差を上述の評価基準を使用して評価した。学習回数１００、２００の場合には、中間ユニット数３０、４０、５０の全てにおいて評価基準値に達せず、両者とも機能が満足できないことが判明した。しかし、学習回数２００の場合は、１００の場合に対して電圧及び無効電力の両算出精度とも向上した。学習回数４００の場合には、中間ユニット数３０、４０、５０の全てにおいて評価基準値をほぼ達成するが、中間ユニット数３０の場合が４０、５０の場合に比べ良好な結果となっている。
【００９６】
図２４は、図１の配電系統制御システムのシミュレーション結果であって、中間ユニット数が３０であるときの単体検証結果である電圧の誤差範囲に対するデータ数を示すグラフであり、図２５は、図１の配電系統制御システムのシミュレーション結果であって、中間ユニット数が３０であるときの単体検証結果である無効電力の誤差範囲に対するデータ数を示すグラフである。また、図２６は、図１の配電系統制御システムのシミュレーションにおける実験の入力条件を示す表である。
【００９７】
すなわち、図２４及び図２５に中間ユニット数３０での単体検証結果の例を示す。また、図２６のように入力条件を同一とした場合、従来の潮流計算の出力値とＮＮ潮流計算の出力値との比較は図２７に示す。図２７に示すように、ＮＮ５００による潮流計算機能が実用可能レベルであることがわかる。さらに、従来の潮流計算手法で、ＩＣＲ９１の機能を付加すれば１配電線当たり１回の処理速度は約３０ｓであるのに対し、ＮＮ５００では１０ｍｓ以下と高速に算出できることも確認できた。従って、実質的にリアルタイムな動作を行うことができる。
【００９８】
次いで、ＧＡ機能の検証について説明する。ＧＡのパラメ−タとして１配電線当たりのストリング数を１００及び２００の２種類とし、選択淘汰率を５０％として、世代による目的関数の値と収束度から適切なストリング数を選択することとした。図２８及び図２９はそれぞれ、図１の配電系統制御システムのシミュレーションにおける検証配線系統のうちの配電系統ＦＦ１，ＦＦ２を示すブロック図である。すなわち、検証用配電系統は図２８及び図２９に示すとおりで、配電線当たりのＩＣＲ設置数は配電系統１の場合が２台、配電系統２の場合が３台とし、各点の負荷は図３０に示すとおりとした。また、この検証を行うにあたって、図４のステップＳ１２におけるＮＮ５００による適応度計算処理においても図３０の固定負荷値を用い、ロ−カル機器２の調整量のみを乱数にて作成し学習デ−タ数４００で事前学習を行わせた。
【００９９】
図２８及び図２９に示す配電系統ＦＦ１及びＦＦ２を対象に、ストリング数１００および２００で１００世代まで実施した目的関数の変化は図３１のとおりである。図３１から明らかなように、目的関数の最終収束値は両者とも同一であるが、ストリング数２００の場合の方が１００の場合に比べ収束する世代数が少ないことから、ストリング数は２００を選定することとした。また、配電系統ＦＦ１及びＦＦ２ともに全ての検出位置において、電圧及び無効電力とも現状より改善された。なお、ストリング数は２００での配電系統ＦＦ１の電圧の目的関数の平均値は０．４７、無効電力の目的関数の平均値は０．９２、配電系統ＦＦ２の電圧の目的関数の平均値は０．４９、無効電力の目的関数の平均値は０．４８であった。
【０１００】
さらに、ＧＡ及びＮＮ機能の総合検証について説明する。ＧＡ及びＮＮ機能の総合検証は、ＧＡ及びＮＮ機能を搭載した中央装置１と模擬ロ−カル機器２とをデータ通信回線４を介して回線接続して、中央装置１からのポ−リングフレ−ム及び目標値指令フレ−ムに対して、模擬ロ−カル機器２からの計測フレ−ムを返信させ、これらのデ−タの送受信によりＧＡ及びＮＮ機能を確認した。
【０１０１】
このときの検証条件としては、配電系統は、図２９の検証用配電系統ＦＦ２に対応させ、負荷パタ−ンは図３０の検証用配電系統ＦＦ２のものを使用した。この場合の初期条件は、「検出位置で電圧が下限値を逸脱」とした。また、目標電圧は６６５０Ｖ、電圧下限許容値は６３００Ｖとし、ＧＡの終了条件は非更新世代数３００とした。
【０１０２】
図３２は、図１の配電系統制御システムのシミュレーションにおける検証前後の電圧の改善効果を示すグラフであり、図３３は、図１の配電系統制御システムのシミュレーションにおける検証前後の無効電力の改善効果を示すグラフである。図３２及び図３３から明らかなように、検証前後の各検出位置での電圧および無効電力を示している。電圧及び無効電力ともにＧＡ及びＮＮの最適化機能が効果的に働いており、システムの妥当性が確認できた。
【０１０３】
以上説明したように、本実施形態によれば、ＮＮ５００とＧＡを用いて配電系統制御システムを構成したので、数ケ所の検出データに基づいて配電系統全体の電圧及び無効電力を、従来例に比較して、効率的にかつ適応的でより高精度で、しかも高速でリアルタイムに調整することができ、これによって、系統全体の電力損失を最小にすることができる。また、ＮＮ５００とＧＡを用いて得られた調整量の結果に基づいて、従来の確定的潮流計算方法により制御の確実性を確認した後、各ローカル装置２を制御しているので、より高精度で確実に配電系統の制御を行うことができる。さらに、各ローカル装置２及び検出装置３のセンサからの検出データは一定時間の平均データを使用して、これに基づいてＮＮ５００とＧＡによる配電系統制御を行うので、各センサ間、並びに中央装置１と各センサ間の時間的誤差を小さくすることができ、より高精度で確実に配電系統の制御を行うことができる。
【０１０４】
また、本実施形態によれば、ＮＮ５００を用いて当該配電系統の潮流計算を実行しているので、複雑な配電系統の相互影響の潮流シミュレーションを実現することができ、従来例に比較して、より高精度で、しかも高速でリアルタイムに実行することができ、これによって、系統全体の電力損失を最小にすることができる。さらに、本実施形態によれば、ＮＮ５００とＧＡによる調整量の計算の前に、現在の配電系統が所定の適正条件を満足しているときに、ＮＮ５００とＧＡによる調整量の計算を行わずに、現状を維持するように構成したので、ＮＮ５００による誤差による影響を最小限にして、現状の配電系統の状態を悪化させる処理を回避することにより、誤制御を防止し、制御効率を向上させることができる。
【０１０５】
なお、本装置は配電系統だけでなく、もちろん種々の制御装置を有するあらゆる回路網の潮流計算又は状態計算に適用できるものである。
【０１０６】
【発明の効果】
以上詳述したように本発明に係る請求項１記載の配電系統制御システムによれば、配電用変圧器に接続された母線と上記母線に接続された複数の高圧配電線とを有する配電系統の各高圧配電線の電圧の変動幅を設定範囲に収めるように制御する配電系統制御システムにおいて、
上記配電系統を統括してその動作を制御する中央装置と、
上記各高圧配電線に対して少なくとも１つ設けられ、上記中央装置からの制御信号に基づいて、上記各高圧配電線の電圧値及び無効電力値をそれぞれ所定の調整量で調整するローカル装置と、
上記各高圧配電線のローカル装置の設置箇所から離れた少なくとも１つの検出位置で上記高圧配電線の電圧及び電流をそれぞれ検出する電圧センサ及び電流センサを備え、検出された電圧値及び電流値の検出データを上記中央装置に送信する検出装置とを備え、
上記中央装置は、
入力層と、少なくとも１層の中間層と、出力層を備え、予め所定の教師データに基づいて学習され、上記各ローカル装置の電圧調整量及び位相調整量と、上記各検出装置間の区間における所定の負荷有効電力及び負荷無効電力と、上記配電用変圧器からの送出電圧及び送出位相とを含む入力データを入力して、上記各検出位置における電圧値及び無効電力値を含む出力データを出力することにより潮流計算を行うニューラルネットワークを用いて、所定の初期の調整量と、上記各検出装置間の区間における所定の負荷有効電力及び負荷無効電力と、上記配電用変圧器からの送出電圧及び送出位相とに基づいて、上記各検出位置における電圧値及び無効電力値を計算する計算手段と、
上記計算手段の処理の前に、上記各検出装置によって検出された電圧値及び電流値の検出データに基づいて、現在の配電系統が適正状態にあるか否かを、所定の適正条件を用いて判断し、適正状態であるときは、上記計算手段による処理を実行しないように制御する制御手段とを備える。
従って、ニューラルネットワークによる計算誤差による影響を最小限にして、現状の配電系統の状態を悪化させる処理を回避することにより、誤制御を防止し、制御効率を向上させることができる。
【０１０７】
また、請求項２記載の配電系統制御システムによれば、請求項１記載の配電系統制御システムにおいて、上記適正条件は、上記各検出装置によって検出された現在の上記各検出位置における電圧値又は無効電力値と、その所定の目標値との差の総和が所定のしきい値以下である。
従って、ニューラルネットワークによる計算誤差による影響を最小限にして、現状の配電系統の状態を悪化させる処理を回避することにより、誤制御を防止し、制御効率を向上させることができる。
【図面の簡単な説明】
【図１】 本発明に係る一実施形態である配電系統制御システムの構成を示すブロック図である。
【図２】 図１の配電系統制御システムの具体例の一例を示すブロック図である。
【図３】 図１の中央装置１によって実行される中央装置処理の第１の部分を示すフローチャートである。
【図４】 図１の中央装置１によって実行される中央装置処理の第２の部分を示すフローチャートである。
【図５】 図４のサブルーチンである適応度計算処理を示すフローチャートである。
【図６】 図４のサブルーチンである選択淘汰処理を示すフローチャートである。
【図７】 図１の中央装置１によって実行される中央装置割り込み処理を示すフローチャートである。
【図８】 図１の配電系統制御システムで用いる目的関数における重み係数の設定表を示す図である。
【図９】 図１の配電系統制御システムで用いる電圧調整量のストリング表現の一例を示す図である。
【図１０】 図１の配電系統制御システムで用いる位相調整量のストリング表現の一例を示す図である。
【図１１】 図１の配電系統制御システムで用いるストリングの個体の一例を示す図である。
【図１２】 図１の配電系統制御システムで用いるストリングの交叉の一例を示す図である。
【図１３】 図１の配電系統制御システムで用いるストリングの突然変異の一例を示す図である。
【図１４】 図１の配電系統制御システムで用いるストリングの多重交叉の一例を示す図である。
【図１５】 図１の配電系統制御システムで用いるニューラルネットワーク（ＮＮ）５００の構成を示すブロック図である。
【図１６】 図１の配電系統制御システムで用いるある負荷状態におけるインピーダンスマップの生成処理を示すブロック図である。
【図１７】 図１の配電系統制御システムで用いる各調整量におけるセンサにおける電圧値Ｖ及び無効電力値Ｑの計算処理を示すブロック図である。
【図１８】 図１５のニューラルネットワーク（ＮＮ）５００の学習方法を示すブロック図である。
【図１９】 図１５のニューラルネットワーク（ＮＮ）５００への学習データの取り入れ方法を示すブロック図である。
【図２０】 図１の配電系統制御システムにおける負荷予測からの電圧算出例を示す図である。
【図２１】 図１の配電系統制御システムで用いる平均化処理における特異データの処理方法を示すグラフである。
【図２２】 図１の配電系統制御システムのシミュレーションを行う検証用配電系統の構成を示すブロック図である。
【図２３】 図１の配電系統制御システムのシミュレーション結果におけるニューラルネットワーク（ＮＮ）５００の学習曲線を示すグラフである。
【図２４】 図１の配電系統制御システムのシミュレーション結果であって、中間ユニット数が３０であるときの単体検証結果である電圧の誤差範囲に対するデータ数を示すグラフである。
【図２５】 図１の配電系統制御システムのシミュレーション結果であって、中間ユニット数が３０であるときの単体検証結果である無効電力の誤差範囲に対するデータ数を示すグラフである。
【図２６】 図１の配電系統制御システムのシミュレーションにおける実験の入力条件を示す表である。
【図２７】 図１の配電系統制御システムのシミュレーション結果であって、配電系統亘長に対する電圧及び無効電力を示すグラフである。
【図２８】 図１の配電系統制御システムのシミュレーションにおける検証配線系統のうちの配電系統ＦＦ１を示すブロック図である。
【図２９】 図１の配電系統制御システムのシミュレーションにおける検証配線系統のうちの配電系統ＦＦ２を示すブロック図である。
【図３０】 図１の配電系統制御システムのシミュレーションにおける検証配線系統のうちの配電系統ＦＦ１及びＦＦ２における各負荷点の負荷値を示す表である。
【図３１】 図１の配電系統制御システムのシミュレーションにおけるストリング数の評価曲線を示すグラフである。
【図３２】 図１の配電系統制御システムのシミュレーションにおける検証前後の電圧の改善効果を示すグラフである。
【図３３】 図１の配電系統制御システムのシミュレーションにおける検証前後の無効電力の改善効果を示すグラフである。
【符号の説明】
１…中央装置、
２…ローカル装置、
２Ａ…ローカル機器、
２Ｂ…電圧センサ、
２Ｃ…電流センサ、
２Ｄ…機器制御装置、
２Ｅ…通信制御及び制御目標設定装置、
３…検出装置、
３Ａ…電圧センサ、
３Ｂ…電流センサ、
３Ｃ…通信子局、
４…データ通信回線、
５…データ送受信装置、
９０…ＬＲＴ目標補正部、
９１…ＩＣＲ、
９２…ＴＣＲ、
９３…センサ、
９４…配電自動化用伝送路、
９５…電圧調整継電器、
９６…ＬＲＴ，
１００…入力層、
１０１−１乃至１０１−２７…入力層ユニット、
２００…中間層、
２０１−１乃至２０１−Ｍ…中間層ユニット、
３００…出力層、
３０１−１乃至３０１−２８…出力層ユニット、
４０１，４０２…シフトレジスタ、
ＳＳ…配電用変圧器、
Ｂ…母線、
Ｆ１，Ｆ２…高圧配電線。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a distribution system control system that controls the voltage and reactive power of each part of a distribution system.
[0002]
[Prior art]
Generally, a distribution system includes a bus connected to a distribution transformer installed in a substation, and a plurality of high-voltage distribution lines connected to the bus, and each high-voltage distribution line is used to compensate for a voltage drop. A voltage regulator is connected as a local device.
[0003]
In a conventional distribution system, all voltage information and current information detected at detection points set in each part of the system are collected in a monitoring device, information on each part of the system is displayed, and voltage is artificially determined based on the state of each part. Whether or not to make the adjustment, and the value of the adjustment value was determined, and the tap of the distribution transformer was selected using the result. In addition, the voltage regulator connected to each high-voltage distribution line is manually set with a preset value based on the configuration of the distribution system, and the voltage of the high-voltage distribution line detected at the installation location of each voltage regulator. The voltage regulator was controlled to keep the value at a set value.
[0004]
Due to the recent increase in demand for interconnection of distributed power sources, power quality management in distribution systems can be adequately handled by conventional control systems that assume only a fixed unidirectional power flow and slow fluctuations. Expected to be impossible. In addition, from the viewpoint of global warming, a minimum power loss of the system is required from the viewpoint of energy saving more than ever. In other words, the current voltage regulation of the distribution system is mainly performed by a voltage regulator at load in the distribution substation (hereinafter referred to as LRT (Load Ratio Transformer)). Further, a step voltage regulator for voltage drop compensation (hereinafter referred to as SVR (Step Voltage Regulator)) is added along the line.
[0005]
Although it is necessary to perform tidal current calculations for this distribution system control system, the tidal current simulator of the prior art performs tidal current calculations using the Newton-Raphson method based on the admittance matrix, which is a well-known electric circuit analysis method. The voltage value and reactive power value at the detection position are calculated. However, the prior art simulator has a problem that the calculation accuracy is relatively low and the calculation speed is relatively slow. In order to solve this problem, the present inventor has proposed a conventional technical document “Hiroyuki Fudo et al.,“ Development of a distribution system voltage / reactive power control method using GA / NN ”, IEICE Technical Committee Materials, Power Technology and "Power System Technology Joint Study Group, PE-97-75, PSE-97-75, October 7, 1997" proposes a method for controlling a distribution system using a genetic algorithm and a neural network. .
[0006]
[Problems to be solved by the invention]
However, in this method, when the voltage and reactive power at each detection position are calculated using a neural network, there are not a few errors, and even if the power distribution system is in an appropriate state, the state tends to deteriorate There was a problem that there was.
[0007]
The object of the present invention is to solve the above-mentioned problems and to avoid a situation in which the state deteriorates due to an error when the voltage and reactive power at each detection position are calculated using a neural network. It is to provide a distribution system control system.
[0008]
[Means for Solving the Problems]
The distribution system control system according to claim 1 of the present invention is a voltage fluctuation of each high-voltage distribution line of a distribution system having a bus connected to a distribution transformer and a plurality of high-voltage distribution lines connected to the bus. In the distribution system control system that controls the width to be within the set range,
A central device that controls the operation of the power distribution system, and
At least one local device that is provided for each of the high-voltage distribution lines, and adjusts the voltage value and reactive power value of each of the high-voltage distribution lines by a predetermined adjustment amount based on a control signal from the central device,
A voltage sensor and a current sensor for detecting the voltage and current of the high-voltage distribution line at at least one detection position remote from the installation location of the local device of each high-voltage distribution line, and detection of the detected voltage value and current value; A detection device for transmitting data to the central device,
The central device is
An input layer, at least one intermediate layer, and an output layer are learned in advance based on predetermined teacher data, and the voltage adjustment amount and phase adjustment amount of each local device, and the section between each detection device Inputs input data including predetermined load active power and load reactive power, and transmission voltage and transmission phase from the distribution transformer, and outputs output data including voltage value and reactive power value at each detection position. By using a neural network that performs power flow calculation, a predetermined initial adjustment amount, a predetermined load active power and a load reactive power in a section between the detection devices, a transmission voltage from the distribution transformer, and Calculation means for calculating a voltage value and a reactive power value at each of the detection positions based on the transmission phase;
Prior to the processing of the calculation means, based on the detection data of the voltage value and the current value detected by each of the detection devices, whether or not the current distribution system is in an appropriate state using a predetermined appropriate condition It is characterized in that it comprises control means for controlling so as not to execute the processing by the calculating means when it is judged and in an appropriate state.
[0009]
The distribution system control system according to claim 2 is the distribution system control system according to claim 1, wherein the appropriate condition is a voltage value or reactive power value at each of the current detection positions detected by the detection devices. And the sum of the differences from the predetermined target value is less than or equal to a predetermined threshold value.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Embodiments according to the present invention will be described below with reference to the drawings.
[0011]
FIG. 1 is a block diagram showing a configuration of a power distribution system control system according to an embodiment of the present invention. In FIG. 1, SS is a distribution transformer installed in the distribution substation, B is a bus connected to the secondary side of the transformer SS, F1, F2,... Are connected to the bus B. High-voltage distribution line. Although not shown, low-voltage distribution lines (not shown) are connected to the high-voltage distribution lines F1, F2,.
[0012]
Reference numeral 1 denotes a central device that supervises a power distribution system. The central device 1 is provided, for example, in a distribution substation. In the present embodiment, at least one local device 2 is provided for each high-voltage distribution line in order to perform voltage adjustment and reactive power adjustment for each of the high-voltage distribution lines F1, F2,. Each of the high-voltage distribution lines F1, F2,... Is also provided with a detection device 3 that detects the three-phase voltage and current of the high-voltage distribution line at at least one detection point set away from the local device 2. Yes. In addition, a data communication line 4 is provided to perform communication between the central device 1 and each local device 2 and detection device 3. On the central device 1 side, a data transmission / reception device 5 for receiving data transmitted from each local device 2 and detection device 3 to the central device 1 and transmitting data from the central device 1 to each local device 2 is provided. It has been.
[0013]
Each local device 2 receives the voltage and current of a local device 2A having a voltage regulator and a reactive power regulator and the corresponding high voltage distribution lines F1 and F2 at detection points set in the vicinity of the installation location of the local device 2A. Internal voltage sensor 2B and internal current sensor 2C to be detected, device control device 2D for controlling voltage regulator and reactive power regulator constituting local device 2A, control target values of voltage regulator and reactive power regulator, respectively A communication control and control target setting device 2E that performs setting processing and communication between the central device 1 is provided. The internal voltage sensor 2B is composed of, for example, an instrument transformer, and the internal current sensor 2C is composed of a current transformer.
[0014]
Here, a well-known voltage regulator (LRT) can be used as the voltage regulator of the local device 2A. As the reactive power adjuster, a shunt reactor or a power capacitor that is selectively connected to the line via a switch, a synchronous phase adjuster, or the like can be used. The communication control and control target setting device 2E receives the voltage information and current information detected by the internal voltage sensor 2B and the internal current sensor 2C, and the state information indicating the state of the local device 2 via the data communication line 4, respectively. 1 is a communication device including a transmitter that transmits to 1 and a receiver that receives a signal transmitted from the central device 1, and voltage control that is a control target value for the voltage and reactive power of the corresponding high-voltage distribution line. A control target setting device for setting the target value and the reactive power control target value. In addition, the device control device 2D has a voltage regulator and reactive power adjustment so as to keep the voltage and reactive power of each part of the corresponding high-voltage distribution line at the voltage control target value and reactive power control target value set by the control target setting device, respectively. Control the instrument.
[0015]
The detection device 3 includes a voltage sensor 3A and a current sensor 3B that detect the voltage and current of the high-voltage distribution line at the installation location, and the voltage information and current information detected by the sensors 3A and 3B, respectively, through the data communication line 4. And a communication slave station 3C that transmits the data to the data transmission / reception device 5 of the central device 1 via the communication device. The voltage sensor 3A is composed of, for example, an instrument transformer, and the current sensor 3B is composed of a current transformer.
[0016]
The central device 1 executes the following processing.
(A) To grasp system information (address and connection of detection device, ON / OFF of VS, type of local device) of the distribution system.
(B) Based on the detection data from the local device 2 and the detection device 3, the load current at each detection position is predicted, and the voltage at each detection position is calculated from the impedance map and the detection information.
(C) The detection data is collected from the devices 2 and 3 by polling the local device 2 and the detection device 3, and checks such as voltage deviation and reactive power increase are performed.
(D) Details Using a genetic algorithm and a neural network, which will be described later, the local device 2 for determining the optimum state is determined and the adjustment amount is calculated.
(E) An impedance map is created based on the detection data and system information.
(F) The power flow is calculated using the impedance map and the detection information, and the target values of the voltage and reactive power at the position of each local device 2 are calculated.
(G) The power flow is calculated using the impedance map, and after confirming the calculated adjustment amount, a macro command is issued to the local device 2.
[0017]
Further, the communication control and control target setting device 2E of the local device 2 executes the following processing.
(1) The line voltage and the phase current are input from the self voltage sensor 3A and the current sensor 3B, the input detection information is averaged, and a temporal average value at a predetermined time is calculated.
(2) In response to polling from the central apparatus 1, the latest information (voltage, current, reactive power) averaged at the present time and the current adjustment amount are returned as a response.
(3) The current adjustment amount is constantly monitored, it is determined whether there is a tolerance for the adjustment range, and the response is returned to the central apparatus 1 by a polling response.
(4) The adjustment amount is calculated with respect to the macro command (target value) from the central apparatus 1, and it is determined whether or not it can be handled.
(5) The set target value of the local device 2A is determined and passed to the device control apparatus 2D together with the detection information.
[0018]
Further, the communication slave station 3C of the detection device 3 executes the following processing.
(1) The line voltage and the phase current are input from the self voltage sensor 3A and the current sensor 3B, the input detection information is averaged, and a temporal average value at a predetermined time is calculated.
(2) In response to polling from the central apparatus 1, the latest information (voltage, current, reactive power) averaged at the present time is returned as a response.
[0019]
Next, a specific example of the power distribution system control system will be described. In this system, the state of each part of the system is grasped by a distribution automation system, and based on this, the line adjusting devices distributed and arranged in each part of the system are controlled. FIG. 2 shows the concept of a system including a central device part to be examined. The system includes a central device 1 that performs overall control, an LRT target correction unit 90 in a substation, and an inverter-controlled regulator (hereinafter referred to as ICR (Inverter Controlled Regulator)) as a line adjustment device installed in a distribution line. 91, a thyristor-controlled reactor (hereinafter referred to as TCR) 92, and a sensor 93 that is installed at a key point of the distribution line and sequentially captures voltage, current, and power factor information. The transmission path 94 is linked. In this configuration, the LRT target correction unit 90 has a bank collective voltage control function for appropriately increasing or decreasing the output voltage of the LRT 96 controlled by the voltage adjusting relay 95. The ICR 91 has a voltage adjustment function at the installation point and a reactive power compensation function for the power supply side line, and the TCR 92 has a delayed reactive power injection function mainly for improving overcompensation by the SC. These devices 91, 92, and 96 receive control target values (the LRT target correction unit 90 is a voltage correction target value, ICR 91 is a voltage and reactive power control target value, and TCR 92 is a reactive power control target value) from the central apparatus 1. The feedback control is performed by itself so as to follow the target value. Therefore, in the central apparatus 1, it is important to generate target values for realizing optimum voltage and reactive power control as a power distribution system and to instruct individual local devices 2 such as ICR 91 and TCR 92. It becomes.
[0020]
Next, the voltage and reactive power control method will be described. The optimization of voltage and reactive power control is a combinatorial optimization problem. In the present embodiment, an optimal optimal solution for voltage and reactive power control in the distribution system is obtained at high speed using a genetic algorithm (hereinafter referred to as GA) and a neural network (hereinafter referred to as NN).
[0021]
3 and 4 are flowcharts showing the central device processing executed by the central device 1 of FIG. In FIG. 3, first, in step S <b> 1, system information and detection data of the target system from the local device 2 and the detection device 3 are input and stored. Next, the presence or absence of a location that deviates from the adjustment margin of the local device 2 and the specified voltage width is confirmed from the detection data input in step S2. Further, in step S3, it is determined whether or not there is a margin or a deviation point from the above data. If YES, the process proceeds to step S7. If NO, the process proceeds to step S4. In step S4, it is determined whether the reactive power at each detection position increases or the voltage is not the lower limit value. If NO, there is no need to adjust, so the process returns to step S1, while if YES, the process proceeds to step S5. In step S5, the appropriateness of the current state with respect to the target voltage and reactive power is determined. In step S6, it is determined whether or not it is inappropriate. If it is not inappropriate (NO), there is no need for adjustment, so step S1. On the other hand, if it is inappropriate (YES), the target local device 2 is designated in step S7. That is, the local device 2 to be controlled is determined from the system configuration (the number of strings in the GA is determined), and the process proceeds to step S11 in FIG.
[0022]
In the central device processing of FIG. 4, steps S11 to S15 are control processing for determining an adjustment amount using GA and NN. In particular, NN is used in step S12. In step S11 of FIG. 4, an initial string (adjustment amount) group generation process is performed to generate a tidal current calculation data group, that is, each adjustment amount of the target local device 2 is randomly generated within the adjustment range and adjusted. Generate a specified number of quantity groups. Next, fitness calculation processing is executed in step S12, and the voltage V and reactive power Q at each detection position when the local device 2 of the target system operates with the given adjustment amount are calculated by power flow calculation (that is, specified A combination value of numbers is calculated), and a function value of a comprehensive objective function (or evaluation function) f (x) described later is calculated based on the calculated voltage V and reactive power Q. Here, the relationship between the adjustment amount and the tidal current distribution is learned in advance by the NN, and the adjustment amount is obtained at high speed by an inference method using the NN. In step S13, it is determined whether or not the termination condition is satisfied. If satisfied, the process proceeds to step S16. If not satisfied (NO), the selection process is performed in step S14. The reciprocal of each objective function f (x) is calculated and selected in descending order of the reciprocal value indicating the fitness, and a plurality of predetermined adjustment amounts at that time are selected as strings having a high fitness. Next, in step S15, crossover and mutation processing is executed to newly generate a predetermined number of adjustment amount groups (corrected strings) in which crossover points are randomly determined from the selected adjustment amounts. Then, bit inversion is performed in the strings in the predetermined plurality of low-fitness strings in the generated population, the mutation process is performed, and then the process returns to step S12 to execute the fitness calculation process. .
[0023]
If the predetermined end condition is satisfied in step S13, each adjustment amount of the target local device 2 is determined in step S16, and the conventional power flow calculation is performed based on the adjustment amount for each device obtained in GA and NN in step S17. In the macro simulation of the power distribution system, the calculated target voltage and target reactive power can be transmitted as the target value of the local equipment by finally confirming using the conventional power flow calculation method (method for obtaining an exact solution). After the validity of the adjustment amount is confirmed, in step S18, a macro command including the adjustment amount determined in step S16 is transmitted to the communication control and control target setting device 2E of the local device 2 via the data communication line 4. Return to step S1 in FIG.
[0024]
FIG. 7 is a flowchart showing a central device interrupt process executed by the central device 1 of FIG. In FIG. 7, first, in step S41, it is determined whether there is a system change in the distribution system, and in step S42, it is determined whether there is new detection data from the local device 2 and the detection device 3. . When YES is determined in the step S41, the processing from the step S51 to the step S58 is executed, while when YES is determined in the step S42, the processing from the step S61 to the step S66 is executed.
[0025]
System information is input in step S51, switch ON / OFF information is input in step S52, and detection data is input in step S53. In step S54, an impedance map is created based on the above information. Then, in step S55, power flow calculation is performed based on the impedance map and the detection data, learning data is created, NN is learned and updated based on the learning data in step S56, and then the process returns to step S41.
[0026]
Moreover, system information is input in step S61, switch on / off information is input in step S62, and detection data is input in step S63. Next, in step S64, an impedance map is created based on the above information, and in step S65, power flow calculation is performed based on the impedance map and the detected data, and learning data is created. Furthermore, after learning the NN 500 based on the learning data created in step S66 and updating the NN 500, the process returns to step S41. In addition, you may comprise so that the process from step S61 to S66 may be performed, for example, whenever new detection data is periodically input.
[0027]
Next, an impedance map creation process in steps S54 and S64 will be described. In this process, the impedance between detection positions is calculated based on the input system information and detection data (voltage, current, reactive power) from each detection position, and is used for power flow calculation. Here, the creation of the impedance map is calculated when system change information is input and when detection data is input. Further, the impedance map recognizes the positional relationship between the detection devices based on the system information, and calculates the impedance between the detection devices based on the input detection data.
[0028]
Furthermore, the learning data creation processing of the neural network by the power flow calculation in steps S57 and S65 will be described. In this process, an impedance map is created when detection data is input, and the power flow is calculated using the impedance map and detection data to calculate the voltage and reactive power at each detection position.
[0029]
Next, additional learning of the network is performed based on the adjustment amount (input) of each local device 2 input as detection data, the voltage calculated by power flow calculation, and reactive power (teacher data at the output).
[0030]
Next, the formulation of the voltage and reactive power optimization problem is described. The problem of realizing optimization of voltage and reactive power in bank units of a distribution substation is considered as a combination optimization problem of adjustment amounts, and can be formulated as follows. That is, for the purpose of optimizing the system voltage and minimizing the reactive power, a comprehensive objective function f (x) defined by the following equation including a voltage objective function and a reactive power objective function is used.
[0031]
[Expression 1]

[0032]
Here, m is the number of distribution lines per bank, a and b are weighting factors, V (x) is a voltage objective function, and Q (x) is a reactive power objective function. Of these, the weighting factors a and b are of a character that should be determined for each distribution line in consideration of the circumstances of the distribution line, but in this embodiment, the weight of the distribution line is determined from the detection information. The values of the coefficients a and b are automatically selected as shown in FIG.
[0033]
The selection of the weighting factor in FIG. 8 is as follows.
(I) If there is a voltage deviation even at one detection point, the weighting factor is used when the voltage deviates. Even if all the detection points are in the “normal state”, it is considered that the voltage and the reactive power should be adjusted if there is a tendency of the states ST1 and ST3, and therefore the weighting factors a and b are both set to 0.5. . Moreover, the objective function V (x) of the voltage was defined as follows:
[0034]
[Expression 2]

[0035]
Here, Vi is each detected voltage before the operation of the local device 2, and Vc is a predicted voltage value of the detection point when the local device 2 calculated by the NN of the central device 1 operates. Vo is a target value (fixed value) to be determined in units of distribution lines. In this embodiment, when each detected voltage deviates from the range of 6300 to 6900 V, it is 6600 V, and when it does not deviate. 6400V. The objective function Q (x) of reactive power was defined as the following equation.
[0036]
[Equation 3]

[0037]
Here, i is the number of detections of one distribution line, Qi is each detected reactive power before the operation of the local device 2, and Qc is a detection point when the local device 2 calculated by the NN of the central device 1 is operated. This is a predicted reactive power value.
[0038]
Next, constraint conditions will be described. First, under the constraint condition of the voltage upper and lower limit values, the transmission voltage of the distribution line and the voltage at the detection location must each satisfy the following formulas. Here, the output voltage must be within the LRT tap adjustment range, and the voltage at the detection location must be a voltage that sets the terminal voltage of the consumer to 101 ± 6V.
[0039]
[Expression 4]
Vfmin ≦ Vf ≦ Vfmax
[Equation 5]
Vsmin ≦ Vs ≦ Vsmax
[0040]
Here, Vf is the voltage of the high-voltage distribution line (feeder), Vfmax is the maximum value of the voltage, and Vfmin is the minimum value of the voltage. Vs is the detection voltage, Vsmax is the maximum value of the detection voltage, and Vsmin is the minimum value of the detection voltage.
[0041]
Next, under the restriction condition of the adjustment amount of the ICR 91, the voltage adjustment amount and the reactive power adjustment amount must not exceed the range of the adjustment capability determined from the specification of the ICR 91 as shown in the following equations.
[0042]
[Formula 6]
V_ICRmin ≦ V_ICR≦ V_ICRmax
[Expression 7]
Q_ICRmin ≦ Q_ICR≦ Q_ICRmax
[0043]
Where V_ICRIs the voltage adjustment value of ICR91, V_ICRmax is the upper limit, V_ICRmin is the lower limit thereof. Q_ICRIs the reactive power (phase) adjustment value of ICR91, Q_ICRmax is its upper limit, Q_ICRmin is the lower limit thereof.
[0044]
Further, under the restriction condition of the adjustment amount of the TCR 92, the reactive power adjustment amount must not exceed the range of the adjustment capability determined from the specification of the TCR 92 as shown in the following equation.
[0045]
[Equation 8]
Q_TCR≦ Q_TCRmax
[0046]
Where Q_TCRIs the reactive power adjustment value of TCR92, Q_TCRmax is the upper limit value.
[0047]
Next, formulation by a combination of GA and NN500 will be described. In the genetic algorithm (GA), a method is required in which the string length is short and it is difficult to generate a dead string by string manipulation. On the other hand, the ICR 91 performs voltage adjustment and phase adjustment, and the reactive power adjustment of the TCR 92 actually adjusts the power factor angle of the installation point of the TCR 91, so the phase adjustment is performed. As string expressions satisfying the above conditions, the following string expression of voltage adjustment amount and string expression of phase adjustment amount are used.
[0048]
In the string representation of the voltage adjustment amount, the controllable voltage adjustment amount of the ICR 91 is ± 330 V. Therefore, when the minimum controllable voltage width is 5 V / step and 5 V is expressed as 1 bit, 330 V becomes 1000010 in binary number. Represented in bits. Here, if the sign of ± is considered as 1 bit, the change in the adjustment amount can be expressed as a string as shown in FIG. In this case, since the voltage adjustment width is 330 V at the maximum, a string exceeding this is a lethal gene.
[0049]
Also, in the string representation of the phase adjustment amount, if the controllable phase adjustment amount of ICR91 and TCR92 is ± 5 degrees and the minimum controllable phase width is 0.1 degrees / step, 0.1 degree is represented as 1 bit. Then, 5 degrees is 110010 in binary and is represented by 6 bits. Here, if the sign of ± is considered as 1 bit, the change in the adjustment amount can be expressed as a string as shown in FIG. In this case, since the phase adjustment width is 5 degrees at the maximum, a string exceeding this is a lethal gene.
[0050]
Further, the string length will be described. String generation is performed on a distribution line basis. In this case, the string length varies depending on the number of local devices 2 of the distribution line. For this reason, a method is adopted in which the maximum number of local devices 2 is set and the string length is changed according to the type and number of the target devices. FIG. 11 shows an example of string length in the case of three ICRs and three TCRs.
[0051]
In the initial string generation method, binary numbers 0 or 1 are randomly assigned to each bit of the adjustment amount of each local device 2. When a lethal gene exceeding the adjustment range is generated, reassignment is performed to generate N strings. Although the adjustment amount currently in operation is not an optimal solution, it is considered to be a sub-optimal solution, so that it is included in one of the initial string groups.
[0052]
Next, the configuration of the applied neural network (NN) 500 will be described. String (adjustment amount) information and information not targeted by GA (active power and reactive power at each detection point, transmission voltage, transmission phase) are sent to the processing unit of NN500, and the environment of the target distribution line is learned (remembered) in advance. Probabilistic power flow calculation (method of inferring output at the same level as normal power flow calculation in a short time) is performed in the NN500 processing unit with a network that has been made, and the calculation results (voltage and invalid at each detection point) Power) is output to the GA processing unit. The back propagation model having a learning function is used for the NN 500, the number of intermediate layers is one, one network is formed for each distribution line, and the number of intermediate units is determined based on the verification results described later. FIG. 15 shows a configuration example of the NN 500 with an 8-section distribution line in which three ICRs and TCRs are installed. In the present embodiment, the intermediate layer 200 is a single layer, but the present invention is not limited to this, and a plurality of layers may be provided.
[0053]
In the NN 500 of FIG. 15, the NN 500 includes an input layer 100, an intermediate layer 200, and an output layer 300. The input layer 100 includes 27 input layer units 101-1 to 101-27, the intermediate layer 200 includes a plurality of M intermediate layer units 201-1 to 201-M, and the output layer 300 includes 28 output layers. Output layer units 301-1 to 301-28. Here, the following data is input to each of the input layer units 101-1 to 101-27.
(A) ICR_V1Thru ICR_V3: Voltage adjustment amount of three ICR91,
(B) ICR_Q1Thru ICR_Q3: Reactive power (phase) adjustment amount of three ICR91s,
(C) TCR_Q1To TCR_Q3: Reactive power (phase) adjustment amount of three ICR91s,
(D) P₁Thru P₈: Load effective power (fixed value) in the section between the detection devices 3
(E) Q₁Thru Q₈: Load reactive power (fixed value) in the section between the detection devices 3
(F) Vs: voltage sent from distribution substation (detected by central unit), and
(G) θs: a transmission phase from the distribution substation (detected by the central device).
[0054]
Next, each of the input layer units 101-1 to 101-17 distributes the input data to M and outputs the divided data to the intermediate layer units 201-1 to 201-M. The intermediate layer units 201-1 to 201-M distribute 28 data weighted by linear combination with a predetermined weighting factor to the 27 pieces of input data, respectively, and output layer units 301-1 to 301-301. Output to -28. Further, the output layer units 301-1 to 301-28 add the M pieces of input data and output the following data.
(A) V_C1To V_C14: Voltage value at each detection position, and
(B) Q_C1Thru Q_C14: Reactive power value at each detection position.
[0055]
Here, the output y of each of the intermediate layer units 201-1 to 201-M has a response function of a (x) and a link load of w._iThen, it can be expressed by the following equation.
[0056]
[Equation 9]

[0057]
Next, the fitness function will be described. Based on the output data from the NN 500 (voltage value and reactive power value at each detection position), the objective function f (x) is obtained using Equation 1, and the evaluation by the GA is an fitness function g (x ).
[0058]
[Expression 10]
g (x) = 1 / f (x)
[0059]
Furthermore, since GA has a characteristic that the evaluation value increases as generations overlap, the process ends when one of the following conditions is satisfied.
(Condition 1) When the set calculation time is reached.
(Condition 2) The maximum evaluation value is not updated over the number of generations specified.
[0060]
In addition, in the GA selection process, a predetermined number of highly adaptable strings are left in the next generation group, and several of them are flagged as mutations to prohibit mutation. In this case, while a string with high fitness remains unconditionally for the next generation, elite strings may spread rapidly throughout the group and fall into a local solution. Yes. In other words, an elite preservation strategy is adopted, and a method in which individuals with the highest fitness in the group are left as they are in the next generation is adopted. While highly adaptable individuals remain unconditionally in the next generation, the elite individuals can spread rapidly throughout the population and fall into a local solution. The actual elite conservation strategy prohibits mutation of elite individuals. In this embodiment, mutation is enabled except for individuals that are arbitrarily set.
[0061]
Also, in the GA crossover process, two strings are selected at random from the selected strings, the crossover points are randomly determined for each voltage and phase adjustment amount, and one point crossover is performed for each of the strings generated. Generating a specified number of strings excluding genes. An example of crossover is shown in FIG. FIG. 14 is a diagram showing an example of multi-point crossover of strings in GA. As shown in FIG. 14, one-point crossover is performed for each adjustment amount of each local device 2.
[0062]
Furthermore, in GA mutation processing, mutation is the conversion of the numerical value at each string position. Therefore, an arbitrary local device 2 that causes mutation for each voltage and phase adjustment amount is selected at random, and an arbitrary bit of the adjustment amount is bit-inverted. FIG. 13 shows an example of mutation.
[0063]
Next, a method for applying the genetic algorithm (GA) will be described. For the following input data, an adjustment amount of the local device 2 that minimizes the objective function f (x) is calculated by GA (NN 500 is used in fitness calculation processing inside GA processing). Outputs the optimum adjustment amount.
(A) GA input data
(A) Target local device 2: Target local device 2 to be optimized.
However, in the case of ICR91, information on whether only the voltage adjustment amount is optimized or whether phase adjustment is also performed is included.
(B) Objective function weight coefficient: The weight coefficients a and b of the objective function f (x).
(C) Target voltage: a target voltage for obtaining the objective function f (x).
Here, the following input information must be data at the same time.
(A) Initial value of the adjustment amount of each local device 2: The current adjustment amount of each local device 2.
(B) Impedance map: current impedance in a section between the detection devices 3.
(C) V, I, Q of the first detection device 3: current, voltage, and reactive power measurement values in the first detection device 3.
(2) Output data
(A) Adjustment amount of the optimized local device 2: It is the optimal adjustment amount for each local device 2 in the input data.
[0064]
In the GA of the present embodiment, a priori information is used as follows. If it does not deviate from the specified voltage, the adjustment amount of each local device 2 currently operated is not an optimal solution, but it can be said that it is a good solution to some extent, and the optimal solution should be in the vicinity of the adjustment amount. By using this, a method of narrowing down the search space and reducing the calculation time by reflecting the adjustment amount of each local device 2 currently operated on the initially generated individual is used. Note that the following method is not used when the voltage deviates from the specified voltage.
[0065]
Here, in the initial generation method (reflection of the initial state), an individual that directly reflects the initial value of the adjustment amount (the adjustment amount currently in operation) is put in one of the initial generation individuals. Although the amount of adjustment currently in operation is not the optimal solution, it can be said that it is a good solution to some extent (when it does not deviate from the specified voltage range), so an efficient search can be performed by reflecting it in the initial generation. Has the advantage.
[0066]
FIG. 5 is a flowchart showing fitness calculation processing (step S12) which is a subroutine of FIG. As shown in FIG. 5, first, 1 is substituted for parameter n in step S21, and the adjustment amount of each local device 2 is obtained from the gene content of the nth individual in step S22. Next, in step S23, the voltage value V and the reactive power value Q at each detection position in the adjustment amount are obtained using the neural network 500. In step S24, the objective function f (x) is calculated and the fitness function is calculated. Further, in step S25, it is determined whether or not the parameter n has reached the number of individuals. If not (NO), the parameter n is incremented by 1 in step S26, the process returns to step S22, and the above processing is repeated. On the other hand, when n has reached the number of individuals in step S25 (YES), the process returns to the original main routine.
[0067]
FIG. 6 is a flowchart showing the selection process (step S14) which is a subroutine of FIG. As shown in FIG. 6, in the selection process, a predetermined number of individuals with high fitness are selected from the previous generation group in step S31. Next, in step S32, a mutation prohibition flag is set for several individuals having high fitness as elite individuals. Then, the process returns to the original main routine.
[0068]
In the GA end determination, the reciprocal g (x) = 1 / f (x) of the objective function f (x) is used as an objective function indicating fitness, and the calculation ends when a preset calculation end time is reached. To do. If the following conditions are met before the end time, the process ends at that time. When calculating with GA, as the number of generations increases, the degree of increase in the evaluation value of the objective function becomes dull so as to be inversely proportional to the exponential function, and eventually the evaluation value does not change no matter how many generations are calculated. . By utilizing this, if “the maximum evaluation value is not updated by the specified generation”, the calculation is terminated assuming that the optimal solution has been reached. Even if the optimal solution exceeds the specified voltage range, it is not a lethal gene. The adjustment amount of the target local device 2 is output as the calculation result.
[0069]
Next, an initial learning method of the NN 500 will be described. First, an impedance map is created. FIG. 16 is a block diagram showing an impedance map generation process in a certain load state used in the distribution system control system of FIG. As shown in FIG. 16, in the generation process, an impedance map in the current load state is created from the detection data and the adjustment amount of each local device 2. Here, detection is performed several times a day, and an impedance map having several patterns is created. Next, learning data in each impedance map is created. FIG. 17 is a block diagram showing calculation processing of the voltage value V and the reactive power value Q in the sensor at each adjustment amount used in the distribution system control system of FIG. As shown in FIG. 17, the voltage value V and the reactive power value Q in each detection device 3 when each local device 2 is changed in a uniform pattern with respect to each impedance map is configured by NN500. Obtained with the conventional tidal simulator. This voltage value V and reactive power value Q are used as learning data. As described above, a pattern is set so that the adjustment amount is uniform, and the output voltage value V and reactive power value Q are used as learning data.
[0070]
Based on the same input data as NN500, the tidal current simulator of the prior art uses the Newton-Raphson method based on the admittance matrix, which is a well-known electric circuit analysis method, to calculate the output data similar to NN500. It is a device to do. This tidal current simulator is used not only when learning the NN 500 but also in the process of step S17 in FIG.
[0071]
FIG. 18 is a block diagram showing a learning method of the neural network (NN) 500 of FIG. Learning is performed based on the learning data created above. Note that learning of the NN 500 is performed separately for each high-voltage distribution line. Further, in the additional learning of the NN 500, the NN 500 needs to reflect the latest learning data because only the NN 500 learned in the initial learning cannot cope with load fluctuations. Therefore, additional learning is performed (see steps S58 and S66 in FIG. 7).
[0072]
FIG. 19 is a block diagram showing a method of incorporating learning data into the neural network (NN) 500 of FIG. As shown in FIG. 19, the shift register 401 connected to each input layer unit 101-1 to 101-27 of the input layer 100 and the output layer unit 301-1 to 301-28 of the output layer 300 are connected. Using the shift register 402, the latest learning data (the creation method is the same as the initial learning) is taken in, and a method of learning the network sequentially is used. At this time, since the learning data increases in a snowball type as time passes, the learning data is discarded from the oldest data in order. Here, the number of learning data is kept constant. The new learning data is a set of 5 to 10 data in which the adjustment amount of the local device 2 is changed. Each input data and teacher data are paired.
[0073]
Next, a confirmation process for deviating from the specified voltage width within a short period will be described. In this process, the voltage deviation is confirmed as follows.
(1) Check the impedance map (line impedance between each detection device 3) for each high-voltage distribution line.
(2) For each high-voltage distribution line, the voltage at each detection position is calculated from the send voltage (predicted value), line impedance, and load current (predicted value) as shown in FIG.
[0074]
(Calculation example)
(A) In case of line impedance only
## EQU11 ##
V₂= V₁: Predicted voltage at the detection position of address 1
(B) With load (line impedance: Z1)
[Expression 12]
V₂= V₁-I₁Z₁
[Formula 13]
Z₁= (R₁₂+ X₁₂)^1/2: Predicted voltage at the detection position of address 1
[0075]
(3) Confirm that there is no deviation from the specified voltage range from the voltage value at each detection position calculated by the above calculation.
(4) After confirming all addresses (all detection positions), if there is a high-voltage distribution line that is recognized as having a deviation, the target high-voltage distribution line and address, and predicted values (voltage, current, impedance: load Z is reactive power Then, the process proceeds to the macro simulation process (step S17 in FIG. 4).
[0076]
Next, the process of selecting the target local device 2 or LRT 96 in step S7 in FIG. 3 will be described. In this processing, (a) when the adjustment amount of the local device 2 is calculated as a result of checking the voltage and reactive power value from the detection data and the data calculated from the prediction, and (b) in the calculation processing of the adjustment amount of the local device 2 There are two possible cases in which the selection review of the local device 2 is necessary.
(A) In the case of initial setting: In the initial setting, there are two cases of voltage deviation correction and other cases (no margin, reactive power increase, or voltage lower limit).
(I) In the case of voltage deviation correction, pay attention to the start of work in the morning, lunch break, stop of work in the evening, etc., and correct it as soon as possible in consideration of dealing with such relatively short load fluctuations. In order to do so, the target local devices 2 are limited in advance. First, the voltage deviation point is recognized from the address, and the ICR 91 having a voltage adjustment function higher than the deviation point is limited as the target local device 2 for each high-voltage distribution line. If the ICR 91 is above the voltage departure point, the LRT 96 is selected for the high-voltage distribution line and the process proceeds to the next process.
(II) In cases other than voltage deviation correction, all local devices 2 are recognized for each high-voltage distribution line.
[0077]
(B) Other than initial setting
(A) In this process, in order to correct a voltage deviation, a certain local device 2 is limited to obtain an adjustment amount and an attempt is made to correct the voltage deviation. However, the correction cannot be performed by the limited local device 2 alone. This is limited to the case where it is necessary to re-select 2 and correct the voltage deviation.
(B) To correct the voltage deviation, the ICR 91 that is higher than the limited ICR 91 is added to the high-voltage distribution line as the target local device 2. If there is no ICR 91 higher than the currently limited ICR 91, the high-voltage distribution line proceeds to the next process as an LRT 96 selection.
[0078]
Next, a process for obtaining the adjustment amount of the local device 2 that minimizes the objective function will be described. In this process, the objective function and the target local device 2 are determined for each high-voltage distribution line from the preprocessing, and the adjustment amount of the local device 2 that minimizes the objective function is obtained under the conditions.
(1) The adjustment amount of the local device 2 that minimizes the objective function is obtained for each target high-voltage distribution line.
(2) When the LRT 96 is not selected as the target local device 2, the adjustment amount of the target local device 2 is set at random, and the voltage and reactive power at each detection position with the adjustment amount are obtained using the NN 500. The objective function f (x) is calculated from the above value, and the objective function f (x) is calculated by updating the adjustment amount of the local device 2 until the end condition is satisfied. When the end condition is satisfied, the local device adjustment amount at that time is set as a solution that minimizes f (x) by GA.
(3) When the adjustment amount of the local device 2 that minimizes the objective function f (x) is obtained, the voltage at each detection position at that time is confirmed, and it is checked whether there is a part that deviates from the specified voltage width. To do. If there is no voltage deviation, the adjustment amount of the target local device 2 in the target high-voltage distribution line is determined. If there is a voltage deviation, the process proceeds to the selection process of the local device 2 in order to review the target local device 2.
(4) When the LRT 96 is selected as the target local device 2, a correction voltage is calculated, a transmission voltage when the correction voltage is corrected is obtained, and how much the transmission voltage changes from the current state is recognized. Then, how each detected voltage of the high voltage distribution line connected to the LRT 96 changes is calculated. Therefore, it is confirmed whether or not a voltage deviation point newly occurs. If it does not occur, change the constraint condition (sending voltage) and perform the macro simulation process. If this occurs, the high-voltage distribution line having no problem at present will be deteriorated. Therefore, the correction of the LRT 96 is not performed, and it is recognized as an uncontrollable part (section where voltage deviation cannot be avoided), and the above (1) and (2 ) Is executed. If the voltage deviation section is an uncontrollable part, the adjustment amount of the target local device 2 is determined as it is.
[0079]
Further, a process of calculating the power flow with a simplified system, obtaining the voltage and reactive power at each detection position when the local device 2 operates with the obtained adjustment amount, and confirming whether or not the data has been improved ( Step S17) in FIG. 4 will be described.
(1) In this process, the tidal current calculation is performed based on the adjustment amount of the local device 2 for each target high-voltage distribution line determined in the pre-processing and the system diagram (impedance map) of the target high-voltage distribution line previously created. And calculate the voltage and reactive power at each detection position.
(2) From the result and detection data, the following objective function (average improvement rate) is calculated. If the improvement rate of voltage and reactive power is 50% or less, it is determined that the data has been improved. If either of the improvement rates exceeds 50%, the one with the worse improvement rate is recognized and the adjustment amount is calculated again.
(I) Voltage objective function fv (x):
[0080]
[Expression 14]

[0081]
here,
Vi: detection data (voltage) at each detection position,
Vf: voltage at each detection position calculated by power flow calculation,
Vo: target voltage,
n: Number of detections.
(II) Reactive power objective function fq (x):
[0082]
[Expression 15]

[0083]
here,
Qi: detection data (reactive power value) at each detection position,
Qf: reactive power value at each detection position calculated by power flow calculation,
Qo: target reactive power,
n: Number of detections.
[0084]
FIG. 21 is a graph showing a method of processing singular data in the averaging process used in the distribution system control system of FIG. In the power distribution system control system of the present embodiment, the central device 1 takes in data from the sensors of the local devices 2 and the detection devices 3 (hereinafter referred to as sensors) and performs optimization control of the system while grasping the contents thereof. Since data cannot be input simultaneously from the local device 2 and the detection device 3 scattered in the high-voltage distribution lines F1 and F2, a time difference occurs at the time of input. How to handle this time difference will be described. Specifically, when collecting the data of each sensor by polling from the central device 1, the detection data collected first and the detection data collected last have a time difference, so they are simply treated as the same. I can't. Therefore, in the present embodiment, temporal averaging processing is performed so that the detected data can be regarded as being collected at the same time. For example, if it is assumed that the collection time error is 1% or less in order to regard the collected detection data as data at the same time, the detection data is averaged at each detection device in 100 times the polling cycle, By collecting average data, it is handled as detection data at the same time (with a time error of 1% or less). However, if the detected data exceeds a certain width from the previously detected data, it is regarded as singular data and the data is discarded.
[0085]
Further, the processing of steps S5 and S6 in FIG. 3 will be described. Even when there is an optimization processing request from the remote monitoring function, the following appropriate conditions (appropriate condition for voltage and appropriate condition for reactive power) hold because there is a possibility that it will be worse than the current situation due to the influence of NN500 error. In this case, it is determined that the current state is almost optimal, and it is confirmed that the optimization process, that is, the process after step S7 in FIG. 3 is not executed (NO in step S6).
[0086]
[Expression 16]

[Expression 17]

[0087]
here,
60: Error (V) of voltage at each detection point of NN500
100: Reactive power error (kVar) at each detection point of NN500,
Vi: Voltage Vo at each current detection position,
Vo: target voltage,
Qi: reactive power at each current detection position,
n: Number of detections.
That is, the appropriate condition shown by the equations 16 and 17 is that the sum of the difference between the current voltage value or reactive power value detected by each detection device 3 at each current detection position and the predetermined target value is a predetermined value. It is above the threshold.
[0088]
【Example】
In order to verify the operation of the power distribution system control system of FIG. 1 configured as described above, the present inventor performed the following simulation.
[0089]
First, single verification of the power flow calculation function by the NN 500 will be described. In the NN 500, the calculation accuracy of the objective function is greatly influenced by the network structure, so how to determine the network structure becomes a problem. Therefore, the structure was examined and verified using the number of intermediate units as a parameter.
[0090]
FIG. 22 is a block diagram illustrating a configuration of a verification distribution system that performs simulation of the distribution system control system of FIG. 1. The verification distribution system is as shown in FIG. 22, and the number of units is the input layer 35 and the output layer 24. The number of intermediate units was three types of 30, 40, and 50. As for the learning data and the evaluation data, the sending voltage is 6900V, the load of each section is 1A step in the range of 25-50A, and the load power factor is 0.00 in the range of 0.4-0.9. In step 01, the voltage adjustment amount of ICR91 was 5V step in the range of ± 330V, and the phase adjustment amounts of ICR91 and TCR92 were each determined by random number in the range of ± 5 ° and 0.1 degree step. The conventional power flow calculation is executed using these data as input values, and the obtained voltage and reactive power value at each detection position is used as learning data, and the number of learning is three patterns of 100, 200, and 400. It was. Evaluation data (data obtained by conventional tidal current calculation) was also generated by the same method as the learning data.
[0091]
The evaluation of the NN500 was based on the error between the voltage and reactive power at each detection position calculated by the NN500 and the evaluation data based on the conventional power flow calculation result, with the evaluation data based on the conventional power flow calculation result as a true value. . The allowable error of voltage is 60V corresponding to 1% of 6600V, and the allowable error of reactive power is set to 300 kvar corresponding to 10% of the maximum reactive power generated under the conditions for creating the evaluation data.
[0092]
Further, in the determination of the learning truncation error of the NN 500, prior to the unit verification of the power flow calculation function by the NN 500, the learning truncation error is determined in advance by a sequential learning method with 30 intermediate units and 100 learning data. The mean square error after learning was calculated. In this case, the mean square error is defined as follows.
[0093]
[Formula 18]

[0094]
FIG. 23 is a graph showing a learning curve of the neural network (NN) 500 in the simulation result of the distribution system control system of FIG. The mean square error after the start of learning is as shown in FIG. 23, and the learning truncation error is set to 0.0002 corresponding to 15 minutes (6000 learning times) in which the progress is slow.
[0095]
Next, 96 different data are input to the NN500 and the conventional power flow calculation to explain the unit verification results, and the difference between the voltage and reactive power calculated at each detection position is calculated using the above evaluation criteria. evaluated. In the case of learning times 100 and 200, it was found that the evaluation reference value was not reached in all of the intermediate units 30, 40, and 50, and the functions of both were not satisfactory. However, in the case of learning number 200, both voltage and reactive power calculation accuracy improved compared to 100. In the case of the learning count 400, the evaluation reference values are almost achieved in all of the intermediate unit numbers 30, 40, and 50. However, the case of the intermediate unit number 30 is better than the cases of 40 and 50.
[0096]
FIG. 24 is a graph showing the simulation result of the distribution system control system of FIG. 1 and showing the number of data with respect to the voltage error range, which is a unit verification result when the number of intermediate units is 30, and FIG. It is a simulation result of 1 distribution system control system, Comprising: It is a graph which shows the number of data with respect to the error range of the reactive power which is a unit verification result when the number of intermediate units is 30. Moreover, FIG. 26 is a table | surface which shows the input conditions of experiment in the simulation of the power distribution system control system of FIG.
[0097]
That is, FIG. 24 and FIG. 25 show examples of single unit verification results with 30 intermediate units. Further, when the input conditions are the same as shown in FIG. 26, a comparison between the output value of the conventional power flow calculation and the output value of the NN power flow calculation is shown in FIG. As shown in FIG. 27, it can be seen that the power flow calculation function by the NN 500 is at a practical level. Furthermore, with the conventional power flow calculation method, if the function of ICR91 is added, the processing speed per one distribution line is about 30 s, whereas NN500 can be calculated as high as 10 ms or less. Accordingly, a substantially real-time operation can be performed.
[0098]
Next, verification of the GA function will be described. As GA parameters, the number of strings per distribution line is 100 and 200, the selection ratio is 50%, and the appropriate number of strings is selected from the value of the objective function and the convergence degree according to the generation. . 28 and 29 are block diagrams showing the distribution systems FF1 and FF2 of the verification wiring system in the simulation of the distribution system control system of FIG. That is, the verification distribution system is as shown in FIGS. 28 and 29, and the number of ICRs installed per distribution line is 2 for distribution system 1 and 3 for distribution system 2, and the load at each point is shown in FIG. 30. Further, in performing this verification, in the fitness calculation processing by the NN 500 in step S12 of FIG. 4, the fixed load value of FIG. 30 is used, and only the adjustment amount of the local device 2 is created with random numbers to obtain learning data. Pre-learning was performed using the number 400.
[0099]
FIG. 31 shows changes in the objective function implemented up to 100 generations with 100 and 200 strings for the power distribution systems FF1 and FF2 shown in FIGS. As is clear from FIG. 31, the final convergence value of the objective function is the same, but the number of generations converged is smaller in the case of 200 strings than in the case of 100 strings, so 200 is selected as the number of strings. It was decided to. In addition, both the distribution systems FF1 and FF2 have improved both voltage and reactive power from the current state at all detection positions. It should be noted that when the number of strings is 200, the average value of the objective function of the voltage of the distribution system FF1 is 0.47, the average value of the objective function of the reactive power is 0.92, and the average value of the objective function of the voltage of the distribution system FF2 is 0 .49, the average value of the reactive power objective function was 0.48.
[0100]
Furthermore, comprehensive verification of GA and NN functions will be described. Comprehensive verification of the GA and NN functions is performed by connecting the central device 1 equipped with the GA and NN functions and the simulated local device 2 via the data communication line 4 to connect the polling frame from the central device 1. And the measurement frame from the simulated local device 2 was returned to the target value command frame, and the GA and NN functions were confirmed by transmitting and receiving these data.
[0101]
As the verification conditions at this time, the power distribution system corresponds to the verification power distribution system FF2 of FIG. 29, and the load pattern used is that of the verification power distribution system FF2 of FIG. The initial condition in this case was “the voltage deviates from the lower limit value at the detection position”. The target voltage was 6650 V, the voltage lower limit allowable value was 6300 V, and the GA termination condition was 300 non-updated generations.
[0102]
32 is a graph showing the improvement effect of the voltage before and after the verification in the simulation of the distribution system control system of FIG. 1, and FIG. 33 shows the improvement effect of the reactive power before and after the verification in the simulation of the distribution system control system of FIG. It is a graph to show. As is clear from FIGS. 32 and 33, the voltage and reactive power at each detection position before and after verification are shown. The optimization functions of GA and NN worked effectively for both voltage and reactive power, and the validity of the system could be confirmed.
[0103]
As described above, according to the present embodiment, since the distribution system control system is configured using the NN 500 and the GA, the voltage and reactive power of the entire distribution system are compared with the conventional example based on the detection data at several locations. Thus, it can be adjusted efficiently, adaptively, with higher accuracy, and at high speed in real time, thereby minimizing the power loss of the entire system. Moreover, since each local apparatus 2 is controlled after confirming the certainty of control by the conventional deterministic power flow calculation method based on the result of the adjustment amount obtained by using NN500 and GA, more accurate Thus, the distribution system can be controlled reliably. Furthermore, since the detection data from the sensors of each local device 2 and the detection device 3 uses average data for a certain period of time, and based on this, the distribution system is controlled by the NN 500 and the GA. The time error between the sensors can be reduced, and the distribution system can be controlled with higher accuracy and reliability.
[0104]
Moreover, according to this embodiment, since the power flow calculation of the said distribution system is performed using NN500, the power flow simulation of the mutual influence of a complicated power distribution system can be implement | achieved, Compared with a prior art example, It can be executed with higher accuracy and at higher speed in real time, thereby minimizing the power loss of the entire system. Furthermore, according to the present embodiment, before the calculation of the adjustment amount by the NN 500 and the GA, when the current distribution system satisfies a predetermined appropriate condition, the calculation of the adjustment amount by the NN 500 and the GA is not performed. Since it is configured to maintain the current status, it is possible to prevent erroneous control and improve control efficiency by minimizing the effects of errors caused by NN500 and avoiding processing that deteriorates the status of the current distribution system. Can do.
[0105]
In addition, this apparatus can be applied not only to the power distribution system but also to power flow calculation or state calculation of any circuit network having various control devices.
[0106]
【The invention's effect】
As described above in detail, according to the distribution system control system according to claim 1 of the present invention, a distribution system having a bus connected to the distribution transformer and a plurality of high-voltage distribution lines connected to the bus. In the distribution system control system that controls the fluctuation range of the voltage of each high-voltage distribution line within the set range,
A central device that controls the operation of the power distribution system, and
At least one local device that is provided for each of the high-voltage distribution lines, and adjusts the voltage value and reactive power value of each of the high-voltage distribution lines by a predetermined adjustment amount based on a control signal from the central device,
A voltage sensor and a current sensor for detecting the voltage and current of the high-voltage distribution line at at least one detection position remote from the installation location of the local device of each high-voltage distribution line, and detection of the detected voltage value and current value; A detection device for transmitting data to the central device,
The central device is
An input layer, at least one intermediate layer, and an output layer are learned in advance based on predetermined teacher data, and the voltage adjustment amount and phase adjustment amount of each local device, and the section between each detection device Inputs input data including predetermined load active power and load reactive power, and transmission voltage and transmission phase from the distribution transformer, and outputs output data including voltage value and reactive power value at each detection position. By using a neural network that performs power flow calculation, a predetermined initial adjustment amount, a predetermined load active power and a load reactive power in a section between the detection devices, a transmission voltage from the distribution transformer, and Calculation means for calculating a voltage value and a reactive power value at each of the detection positions based on the transmission phase;
Prior to the processing of the calculation means, based on the detection data of the voltage value and the current value detected by each of the detection devices, whether or not the current distribution system is in an appropriate state using a predetermined appropriate condition When it is determined that the state is in an appropriate state, control means for controlling so as not to execute the processing by the calculation means is provided.
Therefore, it is possible to prevent erroneous control and improve control efficiency by minimizing the influence of calculation errors caused by the neural network and avoiding processing that deteriorates the state of the current distribution system.
[0107]
Further, according to the distribution system control system according to claim 2, in the distribution system control system according to claim 1, the appropriate condition is the voltage value or invalidity at each of the current detection positions detected by the detection devices. The sum of the difference between the power value and the predetermined target value is less than or equal to a predetermined threshold value.
Therefore, it is possible to prevent erroneous control and improve control efficiency by minimizing the influence of calculation errors caused by the neural network and avoiding processing that deteriorates the state of the current distribution system.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a power distribution system control system according to an embodiment of the present invention.
FIG. 2 is a block diagram showing an example of a specific example of the power distribution system control system of FIG.
FIG. 3 is a flowchart showing a first part of central device processing executed by the central device 1 of FIG. 1;
FIG. 4 is a flowchart showing a second part of the central device process executed by the central device 1 of FIG. 1;
FIG. 5 is a flowchart showing fitness calculation processing that is a subroutine of FIG. 4;
FIG. 6 is a flowchart showing a selection process that is a subroutine of FIG. 4;
FIG. 7 is a flowchart showing a central device interrupt process executed by the central device 1 of FIG. 1;
8 is a diagram showing a setting table of weighting factors in an objective function used in the distribution system control system of FIG.
9 is a diagram showing an example of a string representation of a voltage adjustment amount used in the distribution system control system of FIG.
10 is a diagram showing an example of a string representation of a phase adjustment amount used in the power distribution system control system of FIG.
11 is a diagram showing an example of an individual string used in the power distribution system control system of FIG. 1. FIG.
12 is a diagram showing an example of string crossover used in the power distribution system control system of FIG. 1; FIG.
FIG. 13 is a diagram showing an example of string mutation used in the distribution system control system of FIG. 1;
14 is a diagram showing an example of string multiple crossover used in the distribution system control system of FIG. 1; FIG.
15 is a block diagram showing a configuration of a neural network (NN) 500 used in the power distribution system control system of FIG.
16 is a block diagram showing an impedance map generation process in a certain load state used in the power distribution system control system of FIG. 1. FIG.
17 is a block diagram showing calculation processing of a voltage value V and a reactive power value Q in a sensor at each adjustment amount used in the distribution system control system of FIG.
18 is a block diagram illustrating a learning method of the neural network (NN) 500 of FIG.
FIG. 19 is a block diagram showing a method for incorporating learning data into the neural network (NN) 500 of FIG. 15;
20 is a diagram illustrating a voltage calculation example from load prediction in the distribution system control system of FIG. 1;
FIG. 21 is a graph showing a method for processing singular data in the averaging process used in the distribution system control system of FIG. 1;
22 is a block diagram showing a configuration of a verification power distribution system that performs simulation of the power distribution system control system of FIG. 1; FIG.
23 is a graph showing a learning curve of a neural network (NN) 500 in the simulation result of the distribution system control system of FIG.
24 is a graph showing the simulation result of the distribution system control system of FIG. 1 and the number of data with respect to the voltage error range, which is a unit verification result when the number of intermediate units is 30. FIG.
25 is a graph showing the number of data with respect to an error range of reactive power, which is a result of unit verification when the number of intermediate units is 30, which is a simulation result of the power distribution system control system of FIG. 1;
26 is a table showing input conditions for experiments in the simulation of the distribution system control system of FIG. 1; FIG.
FIG. 27 is a graph showing simulation results of the distribution system control system of FIG. 1 and showing voltage and reactive power with respect to the distribution system length.
28 is a block diagram showing a power distribution system FF1 in the verification wiring system in the simulation of the power distribution system control system of FIG.
29 is a block diagram showing a power distribution system FF2 in the verification wiring system in the simulation of the power distribution system control system in FIG. 1. FIG.
30 is a table showing load values at load points in the distribution systems FF1 and FF2 in the verification wiring system in the simulation of the distribution system control system of FIG. 1;
FIG. 31 is a graph showing an evaluation curve of the number of strings in the simulation of the distribution system control system of FIG. 1;
32 is a graph showing the voltage improvement effect before and after verification in the simulation of the power distribution system control system of FIG. 1;
33 is a graph showing an improvement effect of reactive power before and after verification in the simulation of the distribution system control system of FIG.
[Explanation of symbols]
1 ... Central device,
2 ... Local device,
2A ... Local equipment,
2B ... Voltage sensor,
2C ... current sensor,
2D ... equipment control device,
2E: Communication control and control target setting device,
3 ... detection device,
3A ... Voltage sensor,
3B ... current sensor,
3C: Communication slave station,
4 ... Data communication line,
5. Data transmission / reception device,
90 ... LRT target correction unit,
91 ... ICR,
92 ... TCR,
93 ... sensor,
94 ... Transmission path for distribution automation,
95: Voltage regulating relay,
96 ... LRT,
100 ... input layer,
101-1 to 101-27 ... input layer unit,
200 ... intermediate layer,
201-1 to 201 -M ... intermediate layer unit,
300 ... output layer,
301-1 to 301-28 ... output layer unit,
401, 402 ... shift register,
SS: Distribution transformer,
B ...
F1, F2 ... High voltage distribution lines.

Claims (2)

In a distribution system control system for controlling the voltage fluctuation range of each high-voltage distribution line in a distribution system having a bus connected to a distribution transformer and a plurality of high-voltage distribution lines connected to the bus to be within a set range ,
A central device that controls the operation of the power distribution system, and
At least one local device that is provided for each of the high-voltage distribution lines, and adjusts the voltage value and reactive power value of each of the high-voltage distribution lines by a predetermined adjustment amount based on a control signal from the central device,
A voltage sensor and a current sensor for detecting the voltage and current of the high-voltage distribution line at at least one detection position remote from the installation location of the local device of each high-voltage distribution line, and detection of the detected voltage value and current value; A detection device for transmitting data to the central device,
The central device is
An input layer, at least one intermediate layer, and an output layer are learned in advance based on predetermined teacher data, and the voltage adjustment amount and phase adjustment amount of each local device, and the section between each detection device Inputs input data including predetermined load active power and load reactive power, and transmission voltage and transmission phase from the distribution transformer, and outputs output data including voltage value and reactive power value at each detection position. By using a neural network that performs power flow calculation, a predetermined initial adjustment amount, a predetermined load active power and a load reactive power in a section between the detection devices, a transmission voltage from the distribution transformer, and Calculation means for calculating a voltage value and a reactive power value at each of the detection positions based on the transmission phase;
Prior to the processing of the calculation means, based on the detection data of the voltage value and the current value detected by each of the detection devices, whether or not the current distribution system is in an appropriate state using a predetermined appropriate condition A distribution system control system comprising: a control unit that performs control so as not to execute the processing by the calculation unit when it is determined and is in an appropriate state. The appropriate condition is that the sum of the difference between the voltage value or reactive power value at each current detection position detected by each detection device and the predetermined target value is equal to or less than a predetermined threshold value. The power distribution system control system according to claim 1.

Images