JP2004052633A - Integrated control device for plurality of systems - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an integrated control device for efficient failure diagnosis of a plurality of systems, achieving desired functions without involving a great memory load and an arithmetic load. <P>SOLUTION: A detector 70 stores a standard model which can be satisfied with an input-output relationship in each of the plurality of systems 60-66 by reentering parameters. The plurality of systems 60-66 are selected as controlled object systems in sequence in a time-shared manner. The parameters stored in the controlled object systems are assigned to the standard models to create a model for each system. Actual inputs and actual outputs to the controlled object systems are detected. Failure diagnosis is performed for individual systems, depending on whether a relationship between the actual input and the actual output corresponds to a relationship defined by the model for each system. <P>COPYRIGHT: (C)2004,JPO

Description

と書ける。
【００９５】
このとき制御の目標値をω_ｋとし
Ｊ＝Ｅ（ｙ_ｋ＋ｄ−ω_ｋ＋ｄ）^２　　　　　　　　　　　　　　　　　　（４）
なる評価関数を最少とする制御（最少分散制御）は以下の（５）、（６）式で求められる。但し、Ｅは、期待値をとる作用素である。
【００９６】
【数１】

【００９７】
ここで、Ｆ（ｑ^−１）、Ｇ（ｑ^−１）は次の関係式を満たす多項式である。
Ｆ（ｑ^−１）＝Ａ（ｑ^−１）Ｅ（ｑ^−１）＋ｑ^{− ｄ}Ｆ（ｑ^−１）　　　　　　　　　　　　　　　（７）
Ｇ（ｑ^−１）＝Ｂ（ｑ^−１）Ｅ（ｑ^−１）　　　　　　　　　　　　　　　　　　　　　（８）
但し、
Ｅ（ｑ^−１）＝１＋δ_１ｑ^−１＋δ_２ｑ^{− ２}＋δ_３ｑ^{− ３}＋・・・δ_ｄ−１ｑ^{− （ｄ−１）}（９）
Ｆ（ｑ^−１）＝ｆ_０＋ｆ_１ｑ^−１＋ｆ_２ｑ^{− ２}＋ｆ_３ｑ^{− ３}＋・・・δ_ｎ−１ｑ^{− （ｎ−１）}（１０）
【００９８】
上記の（４）〜（６）式が一般化最少分散制御の式である。
但し、（４）式のｙ_ｋ＋ｄ−ω_ｋ＋ｄが一般に不安定であるため、その代わりに
φ_ｋ＋ｄ＝Ｐｙ_ｋ＋ｄ＋Ｑｕ_ｋ−Ｒω_ｋ＋ｄ
で与えられるφ_ｋ＋ｄの分散を最少にする最適制御を用いることが多い。上記の実施の形態３では、その最適制御が用いられている。
【００９９】
この場合、（１）式で与えられるシステムについて、次の評価関数
Ｊ＝Ｅ（φ_ｋ＋ｄ）^２＝Ｅ（Ｐｙ_ｋ＋ｄ＋Ｑｕ_ｋ−Ｒω_ｋ＋ｄ）^２　　　　　　（１１）
を最少にする入力ｕ_ｋは次式により表すことができる。
【０１００】
【数２】

【０１０１】
但し、Ｅ（ｑ^−１）は（ｄ−１）次、Ｆ（ｑ^−１）は（Ｐの次数＋Ｃの次数−１）の次数を持つものとする。また、Ｐは以下の関係を満たすものとする。
Ｐ（ｑ^−１）Ｃ（ｑ^−１）＝Ａ（ｑ^−１）Ｅ（ｑ^−１）＋ｑ^{− ｄ}Ｆ（ｑ^−１）　　　　　　　　　　　　（１３）
ここで、ノイズのない場合は、Ｃ（ｑ^−１）＝１とすればよい。
上記の（１１）、（１２）式が、最適制御を表す式である。
【０１０２】
［最少二乗法によるシステムのパラメータ推定の説明］
次に、上述した実施の形態１乃至３において、個々のシステムに対応するパラメータの初期値、或いは最新値を推定する際に用いられる最少二乗法によるパラメータの演算手法について説明する。
【０１０３】
標準モデルに代入すべきパラメータが不確定の場合、システムに対する入出力データから未知のパラメータを推定しなければならない。ここでは、入出力データから直接Ｆ（ｑ^−１）、Ｇ（ｑ^−１）＝Ｂ（ｑ^−１）Ｅ（ｑ^−１）、更に、Ｃ（ｑ^−１）の係数を求めるパラメータ推定の手法について述べる。
【０１０４】
ここでは、以下の条件が成立するものとする。
φ_ｋ＋ｄ＝ｙ_ｋ＋ｄ
Ｐ（ｑ^−１）＝１
Ｑ（ｑ^−１）＝γ（１−ｑ^−１）
Ｒ（ｑ^−１）＝１
Ｃ　（ｑ^−１）＝１　　　　　　　　　　　　　　　　　　　　　　　　（１４）
【０１０５】
これから、Ｆ（ｑ^−１）、Ｇ（ｑ^−１）のパラメータをθとするとき、
Ｊ＝Ｅ（φ_ｋ−ψ_ｋ−ｄθ）^２λ^ｔ−ｋ　　　　　　　　　　　　　　（１５）
が最少となるようにパラメータθを推定する。但し、λは０＜λ≦１の関係を満たす既定値である。また、ψ_ｋおよびθは、それぞれ以下の関係を満たすものとする。
ψ_ｋ ^Ｔ＝［ｙ_ｋ，ｙ_{ｋ − １}，ｙ_{ｋ − ２}，・・・，ｕ_ｋ，ｕ_{ｋ − １}，ｕ_{ｋ − ２}，・・・］
θ^Ｔ＝［ｆ_０，ｆ_１，ｆ_２，・・・，ｇ_０，ｇ_１，ｇ_２，・・・］
但し、Ｆ（ｑ^−１）、Ｇ（ｑ^−１）は以下のように表せるものとする。
【０１０６】
【数３】

【０１０７】
上記の（１５）式のＪを最少とする最少二乗推定値θハットは、以下のように与えられる。
【０１０８】
【数４】

【０１０９】
上記（１６）式で表されるθが、最少二乗法により推定されたシステムのパラメータである。また、このようにして推定されたシステムのパラメータＦ（ｑ^−１）、Ｇ（ｑ^−１）＝Ｂ（ｑ^−１）Ｅ（ｑ^−１）を次式に代入することで、（１１）式の評価関数を最少にする望ましい入力を計算することができる。ここで、次式は、（１４）式として示した条件を（１２）式に代入することで得られる式である。
【０１１０】
【数５】

【０１１１】
［遅れｄ＝３の場合のパラメータ推定と最適制御の例］
パラメータ推定と最適制御に関する上記の説明を補足するため、システムの遅れｄ＝３の場合について、パラメータ推定の例、および最適制御の例を説明する。
【０１１２】
制御系設計を行うシステムの入出力関係が、上記（１）式の場合と同様に以下のように表されるものとする。
Ａ（ｑ^−１）ｙ_ｋ＝Ｂ（ｑ^−１）ｑ^{− ｄ}ｕ_ｋ＋Ｃ（ｑ^−１）ｅ_ｋ
ここで、ｕ_ｋは時刻ｋにおけるシステムへの入力、ｙ_ｋは時刻ｋにおけるシステムの出力、ｄはシステムの遅れ（既知）、ｅ_ｋは時刻ｋにおけるノイズ誤差である。但し、
Ａ（ｑ^−１）＝１＋α_１ｑ^−１＋α_２ｑ^{− ２}＋α_３ｑ^{− ３}　　　　（ｎ＝３として）
Ｂ（ｑ^−１）＝β_０　　　　　　　　　　　　　　（ｍ＝ｎ−ｄ＝３−３＝０）
Ｃ（ｑ^−１）＝１　　　　　　　　　　　　　　　（ノイズのない場合と仮定）
【０１１３】
いま、次式の関係より、
【数６】

ψ_ｋおよびθは、それぞれ以下のように表すことができる。
ψ_ｋ ^Ｔ＝［ｙ_ｋ，ｙ_ｋ−１，ｙ_ｋ−２，ｕ_ｋ，ｕ_ｋ−１，ｕ_ｋ−２］
θ^Ｔ＝［ｆ_０，ｆ_１，ｆ_２，ｇ_０，ｇ_１，ｇ_２］　　　　　φ_ｋ＋３＝ｙ_ｋ＋３
【０１１４】
更に、評価関数
Ｊ＝Ｅ（φ_ｋ−ψ_ｋ−ｄθ）^２λ^ｔ−ｋ＝Ｅ（ｙ_ｋ−ψ_ｋ−３θ）^２λ^ｔ−ｋ
を最少とする最少二乗推定値θハットは、（１６）〜（１８）式より、以下のように求められる。
【０１１５】
【数７】

【０１１６】
以下、上記の（Ｐ１）〜（Ｐ３）式について具体的に述べる。
これまでに、Ｐ（ｔ）は対称行列であることがわかっているので、ここでは、Ｐ（ｔ）を以下のように置いて（Ｐ３）式中の▲１▼〜▲４▼、並びに（Ｐ１）〜（Ｐ３）式を求める。
【０１１７】
【数８】

【０１１８】
【数９】

【０１１９】
【数１０】

【０１２０】
【数１１】

【０１２１】
【数１２】

【０１２２】
以上▲１▼〜▲４▼より、（Ｐ１）〜（Ｐ３）式は以下のように記述することができる。まず、（Ｐ１）および（Ｐ２）は、次式（Ｐ４）式のように記述することができる。
【数１３】

【０１２３】
ここで、

【０１２４】
次に、（Ｐ３）式は次式（Ｐ５）式のように記述することができる。
【数１４】

【０１２５】
上記の（Ｐ４）と（Ｐ５）式で求められるθｔ＝［ｆ_０（ｔ），ｆ_１（ｔ），ｆ_２（ｔ），　ｇ_０（ｔ），ｇ_１（ｔ），ｇ_２（ｔ）］は、以下に示す（Ｐ６）式の条件を満たすように求められている。
ｙ（ｔ＋３）−｛ｆ_０（ｔ）ｙ（ｔ）＋ｆ_１（ｔ）ｙ（ｔ−１）＋ｆ_２（ｔ）ｙ（ｔ−２）＋　ｇ_０（ｔ）ｕ（ｔ）＋ｇ_１（ｔ）ｕ（ｔ−１）＋ｇ_２（ｔ）ｕ（ｔ−２）｝＝０　　（Ｐ６）
つまり、上記の（Ｐ４）と（Ｐ５）式で求められるθｔ＝［ｆ_０（ｔ），ｆ_１（ｔ），ｆ_２（ｔ），　ｇ_０（ｔ），ｇ_１（ｔ），ｇ_２（ｔ）］が最少二乗法により推定されたシステムのパラメータである。
【０１２６】
次に、上記の如く求められたシステムのパラメータθｔを用いた最適制御について述べる。
このような最適制御では、パラメータθｔを使って、以下の評価関数を満たすｕ（ｋ）が求められる。
Ｊ＝Ｅ［ｙ（ｋ＋３）−ω（ｋ＋３）＋γ｛ｕ（ｋ）−ｕ（ｋ−１）｝］^２　　　　　　　　　　（Ｐ７）
上記（Ｐ７）式において、ｙ（ｋ＋３）−ω（ｋ＋３）は、３つ先の出力と目標値との差である。また、ｕ（ｋ）−ｕ（ｋ−１）は入力の変動を表す項である。ここで、評価関数に入力の変動を表す項をいれるのは、制御を安定させるためである。
【０１２７】
（Ｐ７）式で、ｙ（ｋ＋３）−ω（ｋ＋３）＋γ｛ｕ（ｋ）−ｕ（ｋ−１）＝０が成立するものとして、その関係に（Ｐ６）式の関係を代入すると、以下の関係が得られる。
｛ｆ_０（ｋ）ｙ（ｋ）＋ｆ_１（ｋ）ｙ（ｋ−１）＋ｆ_２（ｋ）ｙ（ｋ−２）＋　ｇ_０（ｋ）ｕ（ｋ）＋ｇ_１（ｋ）ｕ（ｋ−１）＋ｇ_２（ｋ）ｕ（ｋ−２）｝−ω（ｋ＋３）＋γｕ（ｋ）−γｕ（ｋ−１）＝０
【０１２８】
上記の関係式は、以下のように書き直すことができる。
｛ｇ_０（ｋ）＋γ｝ｕ（ｋ）＝［ω（ｋ＋３）−ｆ_０（ｋ）ｙ（ｋ）−ｆ_１（ｋ）ｙ（ｋ−１）−ｆ_２（ｋ）ｙ（ｋ−２）−｛ｇ_１（ｋ）−γ｝ｕ（ｋ−１）−ｇ_２（ｋ）ｕ（ｋ−２）］
従って、ｕ（ｋ）は以下のように表すことができる。
【０１２９】
【数１５】

【０１３０】
以上より、遅れｄ＝３のシステムの場合において、３つ先の出力ｙ（ｔ＋３）を目標出力ω（ｔ＋３）にするための修正入力値ｕ（ｔ）は、次式（Ｐ８）のように求められる。この（Ｐ８）式により求められる値が、実施の形態３において最適制御を実現するための修正入力値である。
【０１３１】
【数１６】

【０１３２】
実施の形態４．
次に、図９を参照して本発明の実施の形態４について説明する。図９は、本実施形態の装置が、複数のシステムのそれぞれについて故障診断を行うために用いる手法を説明するための図である。図９に示す構成は、検出装置で用いられる標準モデルが、線形近似式のモデルからニューラルネットに置き換えられている点を除き、図１（Ｂ）に示す構成と同様である。
【０１３３】
すなわち、本実施形態の装置は、実施の形態１乃至３の何れかの装置において、ＥＣＵ５０に、複数のシステムの入出力関係を推定するための標準モデルとして、上記（式１）に示す線形近似式に変えて、ニューラルネットの機能を組み込むことにより実現することができる。
【０１３４】
上述した実施の形態１乃至３では、線形近似式の標準モデルに代入すべき個々のシステムのパラメータを、最少二乗法により推定している。これに対して、ニューラルネットでは、個々のシステムに対するパラメータ（結合係数）を、バックプロパゲーションによる公知の手法で求めることができる。
【０１３５】
本実施形態の装置においては、上述した公知の手法で求めたパラメータ（結合係数）を予めＥＣＵ５０に記憶させておくと共に、ＥＣＵ５０に、図５に示すルーチンに沿った処理を実行させることにより、実施の形態１の場合と同様の機能を実現することができる。この場合、パラメータが初期値に設定されているニューラルネットによって実車上で推定された出力が、実車上で実測された出力に対してどの程度の誤差を有しているかに基づいて、個々のシステムの故障診断を行うことができる。
【０１３６】
また、本実施形態の装置においては、上述した公知の手法で求めた結合係数を予めＥＣＵ５０に記憶させておき、かつ、ＥＣＵ５０に、その手法で結合係数を求めるための機能を組み込んだうえで、図６に示すルーチンに沿った処理を実行させることもできる。この場合、本実施形態の装置によって、実施の形態２の場合と同様の機能を実現することができる。
【０１３７】
図６に示すルーチンでは、実車上で、標準モデルに代入するパラメータ（結合係数）の最新値を求めることが要求される。ＥＣＵ５０には、バックプロパゲーションにより結合係数を求める機能が組み込まれているため、その要求を満たすことができる。また、図６に示すルーチンでは、実車上で求めたパラメータの最新値と、その初期値との比較に基づいてシステムの状態を判断することが要求される。本実施形態において、ＥＣＵ５０は、中間層の状態表現（重み）を比較することにより、そのような要求を満たすことができる。
【０１３８】
更に、本実施形態の装置においては、上述した公知の手法で求めた結合係数を予めＥＣＵ５０に記憶させておき、かつ、ＥＣＵ５０に、その手法で結合係数を求めるための機能を組み込んだうえで、図７および図８に示すルーチンに沿った処理を実行させることもできる。この場合、本実施形態の装置によって、実施の形態３の場合と同様の機能を実現することができる。
【０１３９】
図７に示すルーチンでは、実車上で、標準モデルに代入するパラメータ（結合係数）の最新値を求めることが要求される。ＥＣＵ５０には、バックプロパゲーションにより結合係数を求める機能が組み込まれているため、その要求を満たすことができる。また、図７に示すルーチンでは、実車上で求めたパラメータの最新値と、その初期値との比較に基づいてシステムの状態を判断することが要求される。本実施形態において、ＥＣＵ５０は、中間層の状態表現（重み）を比較することにより、そのような要求を満たすことができる。
【０１４０】
更に、図８に示すルーチンでは、実車上で求めたパラメータの最新値θｔに基づいて、最適制御を実行することが要求される。本実施形態において、ＥＣＵ５０は、パラメータ（結合係数）の最新値を代入したニューラルネットにより、目標出力ωｔ＋ｄが得られるように修正入力値ｕ（ｔ）を求めることで、そのような要求を満たすことができる。
【０１４１】
ニューラルネットによれば、線形近似式では表現が困難な非線形な入出力関係も精度良く表現することが可能である。このため、本実施形態の装置によれば、実施の形態１乃至３の装置に比して、より多くのシステムとの適合を得ることができる。従って、本実施形態の装置によれば、実施の形態１乃至３の装置に比して、より高い汎用性を確保することができる。
【０１４２】
尚、線形近似式によるモデルをニューラルネットのモデルに置き換える手法は、例えば、特開平８−１１７１９９号公報、或いは、ＥＰ０６９９４１３Ａ１などに詳しく開示されている。
【０１４３】
ところで、上述した実施の形態１乃至４においては、制御の対象が車両に搭載されている複数のシステムに限定されているが、本発明はこれに限定されるものではない。すなわち、本発明は、車両上のシステムに限らず、プラント施設などに設置されている複数のシステムに対して適用してもよい。
【０１４４】
【発明の効果】
この発明は以上説明したように構成されているので、以下に示すような効果を奏する。
第１の発明によれば、複数のシステムのそれぞれを順次制御対象システムとすることで、個々のシステムに対する所望の制御を時分割で順次行うことができる。このため、本発明によれば、高い演算負荷を伴わずに全てのシステムにつき、所望の制御を実行することができる。また、本発明によれば、パラメータの入れ替えにより個々のシステムの状態に適合したものとなる標準モデルを用いて、個々のシステムに対する制御を行うことができる。このため、本発明によれば、多大なメモリ負荷を伴わずに全てのシステムにつき所望の制御を実行することができる。
【０１４５】
第２の発明によれば、制御対象システムに対応するパラメータの初期値を標準モデルに当てはめることにより、そのシステムの初期値対応モデルを生成することができる。更に、初期値対応モデルと、そのシステムに対する実入力および実出力とを用いることで、そのシステムの出力推定値を演算することができる。そして、このようにして得られた出力推定値が、実出力と大きく異なっているか否かに基づき、システムの故障の有無を診断することができる。
【０１４６】
第３の発明によれば、標準モデルと、制御対象システムに対する実入力および実出力とに基づいて、そのシステムの最新の状態に適合するパラメータの最新値を演算することができる。そして、このようにして得られたパラメータの最新値が、パラメータの初期値と大きく異なっているか否かに基づき、システムの故障の有無を診断することができる。
【０１４７】
第４の発明によれば、標準モデルと、制御対象システムに対する実入力および実出力とに基づいて、そのシステムの最新の状態に適合するパラメータの最新値を演算することができる。更に、このようにして得られたパラメータの最新値を標準モデルに当てはめることにより、そのシステムの最新状態に対応する最新値対応モデルを生成することができる。そして、本発明によれば、その最新値対応モデルと目標出力とに基づき、そのシステムに入力すべき修正入力を求め、その修正入力が実現されるように必要な制御の修正が行われる。このため、本発明によれば、システムの経時変化等に影響されることなく、常に最適な制御状態を維持することができる。
【０１４８】
第５の発明によれば、標準モデルが線形近似式であるため、種々の演算処理を簡単に行うことができる。
【０１４９】
第６の発明によれば、標準モデルがニューラルネットであるため、複数のシステム内に、入出力の関係が非線形となるようなシステムが含まれる場合にも、全てのシステムに対して高い制御精度を確保することができる。
【図面の簡単な説明】
【図１】本発明の実施の形態１で実行される制御の概要を説明するための図である。
【図２】任意のシステムについての入出力の関係を示す図である。
【図３】本発明の実施の形態１の具体的構成を説明するための図である。
【図４】本発明の実施の形態１の装置において形成されている制御系の構成を具体的に表した図である。
【図５】本発明の実施の形態１において実行される制御ルーチンのフローチャートである。
【図６】本発明の実施の形態２において実行される制御ルーチンのフローチャートである。
【図７】本発明の実施の形態３において実行される第１の制御ルーチンのフローチャートである。
【図８】本発明の実施の形態３において実行される第２の制御ルーチンのフローチャートである。
【図９】本発明の実施の形態４で実行される制御の概要を説明するための図である。
【符号の説明】
１０　内燃機関
１８　エアフロメータ
２１　スロットルセンサ
２４　触媒
２６　空燃比センサ
２８　Ｏ_２センサ
３４　回転数センサ
３６　バッテリ電流計
３８　バッテリ電圧計
４０　トルクセンサ
４２　車輪速センサ
５０　ＥＣＵ（Ｅｌｅｃｔｒｏｎｉｃ　Ｃｏｎｔｒｏｌ　Ｕｎｉｔ）
６０，６２，６４，６６　システム
７０　検出装置
ｘ，ｘ（ｔ），ｕ_ｋ　入力
ｙ，ｙ（ｔ），ｙ_ｋ　出力
ｆｎ，ｇｎ，ｆ１ｎ，ｇ１ｎ，ｆ２ｎ，ｇ２ｎ，θ　パラメータ
ｄ　システムの遅れ時間
ω_ｋ　目標出力[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an integrated control device for a plurality of systems, and more particularly to an integrated control device for a plurality of systems suitable for efficiently performing a failure diagnosis or optimal control of a plurality of systems mounted on a vehicle.
[0002]
[Prior art]
Japanese Patent Laying-Open No. 9-330210 discloses an apparatus for easily specifying a failed system on a vehicle equipped with a plurality of systems. This apparatus includes a simulator unit including a plurality of models corresponding to each system so that a failure of each system can be independently detected. According to this device, a failure of each system can be detected independently of the state of another system based on each model. Therefore, according to the above-described conventional device, when a failure occurs in any of the plurality of systems mounted on the vehicle, the failure can be easily and quickly identified.
[0003]
[Problems to be solved by the invention]
However, a large amount of memory is required to prepare separate and independent models for each of a plurality of systems as in the conventional apparatus described above. Furthermore, in order to execute fault diagnosis using a plurality of models independently of each other, it is necessary to give high arithmetic processing capability to a device that performs the processing. For these reasons, the conventional device described above has a problem that it is difficult to realize it at low cost.
[0004]
SUMMARY OF THE INVENTION The present invention has been made to solve the above-described problems, and an object of the present invention is to provide an integrated control device capable of efficiently controlling a plurality of systems without a large memory load and a large calculation load. Aim.
[0005]
[Means for Solving the Problems]
A first invention is an integrated control device for controlling a plurality of systems to achieve the above object,
Standard model storage means for storing a standard model capable of satisfying an input / output relationship in each of the plurality of systems by exchanging parameters;
A system selection unit that selects each of the plurality of systems as a control target system in a time-division manner,
Actual input / output detection means for detecting an actual input and an actual output in the controlled system,
Based on the actual input and the actual output detected for the control target system, based on the standard model, an individual system control unit that executes control on the control target model,
It is characterized by having.
[0006]
Further, the second invention is based on the first invention,
With parameter initial value storage means for storing an initial value of the parameter corresponding to each of the plurality of systems,
The individual system control means,
Initial value corresponding model generation means for applying an initial value of the parameter corresponding to the control target system to the standard model to generate an initial value corresponding model of the system,
An output-estimated-value calculating unit that calculates an output estimated value of the controlled system based on the initial-value-corresponding model and the actual input and the actual output;
Failure diagnosis means for performing failure diagnosis of the controlled system based on a comparison result between the output estimation value and the actual output;
It is characterized by including.
[0007]
Further, a third invention is the first invention, wherein:
With parameter initial value storage means for storing an initial value of the parameter corresponding to each of the plurality of systems,
The individual system control means,
The standard model, based on the actual input and the actual output, the latest parameter calculating means for calculating the latest value of the parameter that matches the latest state of the controlled system,
Failure diagnosis means for performing a failure diagnosis of the controlled system based on a comparison result between the initial value and the latest value of the parameter,
It is characterized by including.
[0008]
In a fourth aspect based on the first aspect,
The individual system control means,
The standard model, based on the actual input and the actual output, the latest parameter calculating means for calculating the latest value of the parameter that matches the latest state of the controlled system,
A latest value corresponding model generation unit configured to apply a latest value of the parameter corresponding to the control target system to the standard model to generate a latest value corresponding model of the system, further comprising:
For a system in which the latest value corresponding model has been obtained, target output detection means for detecting a target output,
Correction input calculation means for calculating a correction input to be supplied to the system in order to obtain the target output, based on the latest value corresponding model,
Control correcting means for correcting the control related to the actual input so that the actual input to the system matches the corrected input;
It is characterized by having.
[0009]
According to a fifth invention, in any one of the first to fourth inventions, the standard model is a linear approximation formula.
[0010]
According to a sixth aspect of the present invention, in any one of the first to fourth aspects, the standard model is a neural network.
[0011]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. Elements common to the drawings are denoted by the same reference numerals, and redundant description will be omitted.
[0012]
Embodiment 1 FIG.
[Overview of control]
FIG. 1 is a diagram for describing an outline of control according to Embodiment 1 of the present invention. More specifically, FIG. 1A is a diagram for explaining a general method for performing a failure diagnosis for each of a plurality of systems. On the other hand, FIG. 1B is a diagram for explaining a method used in the present embodiment to perform a failure diagnosis for each of a plurality of systems.
[0013]
Vehicles are equipped with multiple systems, each performing an important function. On a vehicle, when an abnormality occurs in any of these systems, it is desired that the abnormality be promptly detected in a state where the abnormality has been specified. Such a requirement is generally desired in a complex system (eg, a plant or the like) comprising a plurality of systems.
[0014]
When detecting faults of individual systems included in a complex system separately and independently for each system, it is general to prepare a configuration as shown in FIG.
That is, the complex system shown in FIG. 1A includes a plurality of systems (systems 1, 2,...). These systems emit outputs (outputs 1, 2,...) Individually for inputs (inputs 1, 2,...) Supplied to the respective systems.
[0015]
In the configuration shown in FIG. 1A, a detection device (detection devices 1, 2,...) Is prepared for each of these systems (systems 1, 2,...). The detection device 1 determines whether the input 1 supplied to the system 1 and the output 1 output from the system 1 are in an appropriate relationship, and based on the result, determines whether or not the system 1 has a failure. 1 device. Similarly, the detection device 2 determines whether the input 2 supplied to the system 2 and the output 2 emitted from the system 2 are in an appropriate relationship, and based on the result, whether there is a failure in the system 2 or not. Is a device that issues a determination 2 representing
[0016]
As shown in FIG. 1A, if a configuration is provided in which a detection device is prepared for each system, failure diagnosis of each system can be performed irrespective of the state of other systems. Therefore, according to such a configuration, it is possible to easily and reliably detect a failure of each system.
[0017]
However, if an individual detection device is prepared for each of all the systems included in the complex system, problems such as an increase in cost and an increase in the installation space of the detection device occur. On the other hand, since the failure diagnosis of the system is a process that does not need to be performed very frequently, a detection device is provided for each system to form a state in which the failure diagnosis for a plurality of systems can be performed simultaneously. It is not necessary.
[0018]
Therefore, in the present embodiment, as shown in FIG. 1B, only one detection device having versatility is prepared for a plurality of systems (systems 1, 2,...) Included in the complex system. The detection device sequentially diagnoses failures of a plurality of systems in a time-division manner. That is, in the present embodiment, each of the plurality of systems (systems 1, 2,...) Is sequentially selected as the control target system as time progresses. Then, for example, when the system 1 is selected as a control target system, the input 1 supplied to the system 1 and the output 1 generated by the system 1 are input to the detection device.
[0019]
The detection device used in the present embodiment has a standard model including a plurality of parameters. This standard model is a model that is set so that the input / output relationship can be represented for all systems (systems 1, 2,...) Included in the complex system by changing the above parameters.
[0020]
[Description of standard model]
Hereinafter, the contents of the standard model used in the present embodiment will be described in more detail with reference to FIG.
FIG. 2 shows an input / output relationship for an arbitrary system. In general, for all systems, the following linear approximation relational expression holds between the input x and the output y.
y (t + d) = f1 · y (t) + f2 · y (t−1) + f3 · y (t−2) +... + g1 · x (t) + g2 · x (t−1) + g3 X (t−2) +... (Formula 1)
However, in the above (Equation 1), t is an arbitrary sampling time, and (t−n) is a sampling time n times before the time t. D is the delay of the system, and (t + d) is the time at which the history before time t is reflected on the output y of the system. Further, fn and gn are parameters to be adapted to the individual system.
In the present embodiment, the detection device illustrated in FIG. 1B stores the linear approximation expressed by the above (Equation 1) as a standard model.
[0021]
The detection device shown in FIG. 1B also stores parameters fn and gn corresponding to all the systems (systems 1, 2,...). Then, for example, when the system 1 is selected as the control target system, the parameters f1n and g1n corresponding to the system 1 are substituted for the parameters fn and gn of the standard model (Equation 1). When the system 2 is selected as a control target system, the parameters f2n and g2n corresponding to the system 2 are substituted for the parameters fn and gn of the standard model (Equation 1).
[0022]
(Equation 2) and (Equation 3) shown below are models set for the system 1 and the system 2 according to the above rules, respectively.
y1 (t + d1) = f11 · y1 (t) + f12 · y1 (t−1) + f13 · y1 (t−2) +... + g11 · x1 (t) + g12 · x1 (t−1) + g13 X1 (t-2) +... (Equation 2)
y2 (t + d2) = f21 · y2 (t) + f22 · y2 (t−1) + f23 · y2 (t−2) +... + g21 · x2 (t) + g22 · x2 (t−1) + g23 X2 (t-2) +... (Equation 3)
The parameters fn and gn corresponding to each system, such as f1n and g1n or f2n and g2n, can be obtained by a method using the least squares method. The calculation method of those parameters will be described later in detail.
[0023]
[Explanation of time sharing (time sharing)]
The detection device illustrated in FIG. 1B sequentially selects each of a plurality of systems as a system to be controlled in a time-division manner. Then, a model as shown in (Expression 2) or (Expression 3) is sequentially generated according to the selected control target system.
[0024]
When the detection apparatus selects the system 1 as the control target system, and as a result, if the model of the above (Equation 2) is set, the output y1 (t + d1) calculated based on the model is output y1 (t + d1). It is determined whether or not the measured value matches the actually measured value of (t + d1). As a result, if both match, a determination 1 indicating that the system 1 is normal is output, while if they do not match, a determination 1 indicating that the system 1 is abnormal is output. Further, when the system 2 is selected as the control target system, the detection device performs the same processing using the model of the above (Equation 3).
[0025]
As described above, in the present embodiment, the detection device illustrated in FIG. 1B performs failure diagnosis of all systems (systems 1, 2,...) Included in the complex system using the standard model. , Can be sequentially executed in a time-division manner. Therefore, according to the configuration of the present embodiment, the control device of the complex system can be realized at low cost and in a small space as compared with the general configuration shown in FIG.
[0026]
[Description of Specific Configuration of Control Device]
Next, a specific configuration of the first embodiment of the present invention will be described with reference to FIG. The device according to the present embodiment is an integrated control device for efficiently controlling a complex system including a plurality of systems mounted on a vehicle.
[0027]
As shown in FIG. 3, the device of the present embodiment includes an internal combustion engine 10. An intake passage 12 and an exhaust passage 14 communicate with the internal combustion engine 10. An air flow meter 18 is arranged in the intake passage 12 downstream of the air filter 16. The air flow meter 18 is a sensor that detects an intake air amount Ga flowing through the intake passage 12. A throttle valve 20 is provided downstream of the air flow meter 18. In the vicinity of the throttle valve 20, a throttle sensor 21 for detecting the throttle opening TA is arranged. Further, a fuel injection valve 22 for injecting fuel to an intake port of the internal combustion engine 10 is disposed downstream of the throttle valve 20.
[0028]
A catalyst (start catalyst) 24 is disposed in the exhaust passage 14. When a lean exhaust gas having an air-fuel ratio is discharged into the exhaust passage 14, the catalyst 24 includes oxygen contained in the gas.₂NOx in the exhaust gas is purified (reduced) by storing the NOx. On the other hand, when the exhaust gas is rich, the stored O2₂This is a device that purifies (oxidizes) HC and CO in exhaust gas by discharging water.
[0029]
An air-fuel ratio sensor 26 is arranged upstream of the catalyst 24. The air-fuel ratio sensor 26 is a sensor that outputs an output according to the exhaust air-fuel ratio. The air-fuel ratio sensor 26 can detect the air-fuel ratio of the exhaust gas immediately after being discharged from the internal combustion engine 10, that is, the air-fuel ratio of the exhaust gas before flowing into the catalyst 24.
[0030]
Downstream of the catalyst 24, O₂A sensor 28 is arranged. O₂The sensor 28 is a sensor that outputs an output according to whether the air-fuel ratio is rich or lean. O₂According to the sensor 28, it can be determined whether a lean gas containing oxygen flows out of the catalyst 24 or a rich gas containing no oxygen flows out.
[0031]
A water temperature sensor 32 for detecting a cooling water temperature THW is attached to a cylinder block of the internal combustion engine 10. An NE sensor 34 for detecting the engine speed NE is arranged near the crankshaft (not shown) of the internal combustion engine 10.
[0032]
Furthermore, the device of the present embodiment includes a battery ammeter 36 for detecting a current Ibatt flowing into and out of a battery (not shown) and a battery voltmeter 38 for detecting a battery voltage Vbatt. In addition, the device of the present embodiment includes a torque sensor 40 for detecting a wheel torque transmitted (input) to each wheel, and a wheel speed sensor 42 for detecting a wheel speed of each wheel. I have.
[0033]
As shown in FIG. 3, the device of the present embodiment includes an ECU (Electronic Control Unit) 50. The various sensors described above and the fuel injection valve 22 are electrically connected to the ECU 50. The ECU 50 controls the fuel injection timing and the fuel injection amount based on the sensor outputs, and executes failure diagnosis of various systems included in the device of the present embodiment.
[0034]
[Description of specific configuration of control system]
FIG. 4 is a diagram specifically showing the configuration of a control system formed in the apparatus of the present embodiment, that is, the concept of the control system shown in FIG. 1B is applied to the vehicle system shown in FIG. FIG. 3 is a diagram showing a configuration of a control system specifically realized by the above.
[0035]
The configuration shown in FIG. 3 includes the internal combustion engine 10 and the air-fuel ratio sensor 26. Focusing on these inputs and outputs, the air-fuel ratio of the air-fuel mixture flowing into the internal combustion engine 10 is an input, and the sensor value of the air-fuel ratio sensor 26 (hereinafter, referred to as an “A / F sensor value”) is an output. Can be seen as a system. The air-fuel ratio of the air-fuel mixture can be obtained by calculation based on the intake air amount Ga, the fuel injection amount, and the amount of fuel adhering to the intake port and the cylinder. Hereinafter, the calculated value is referred to as “calculated A / F”. In FIG. 4, the above-described system including the internal combustion engine 10 and the air-fuel ratio sensor 26 is represented as a system 60.
[0036]
The configuration shown in FIG.₂A sensor 28 is included. The air-fuel ratio downstream of the catalyst 24 is changed from rich to lean as the lean exhaust gas further containing oxygen continues to flow after the catalyst 24 has fully absorbed the storage capacity of oxygen. Further, after the catalyst 24 releases all oxygen, the rich exhaust gas containing unburned components such as HC and CO continues to flow out, so that the exhaust gas is changed from lean to rich. Therefore, catalyst 24 and O₂Focusing on the input and output of the sensor 28, the integrated amount of oxygen or unburned components (HC, CO) in the exhaust gas flowing into the catalyst 24 (the integrated amount of the difference between the exhaust air-fuel ratio and the stoichiometric air-fuel ratio) is input. And O₂This can be viewed as a system where the sensor value of sensor 28 is the output. Then, the input can be obtained based on the A / F sensor value. Therefore, catalyst 24 and O₂The above system, including sensor 28, can be represented as system 62 in FIG.
[0037]
The two systems described above are systems included in the engine system in the configuration shown in FIG. In the engine system, in addition to these systems, there are systems that can be handled in the same manner as the systems 60 and 62 by focusing on the relationship between input and output, but here, for convenience of explanation, Description of the system is omitted.
[0038]
The configuration shown in FIG. 3 includes a battery (not shown). Battery voltage Vbatt is determined according to the amount of current flowing into the battery and the amount of current flowing out of the battery, that is, according to battery current Ibatt. Therefore, when focusing on the input and output of the battery, it can be regarded as a system in which the battery current Ibatt is an input and the battery voltage Vbatt is an output. In FIG. 4, the above-described system including a battery is represented as system 64.
[0039]
The configuration shown in FIG. 3 includes wheels (tires) not shown. The wheel speed is determined according to the value of the torque input to the wheel. Therefore, focusing on the input and output of the wheel, the wheel can be regarded as a system in which the sensor value of the torque sensor 40 is an input and the sensor value of the wheel speed sensor 42 is an output. In FIG. 4, the above-described system including wheels is represented as system 66.
[0040]
The two systems described above are systems included in the vehicle system in the configuration shown in FIG. In the vehicle system, in addition to these systems, there are systems that can be handled in the same manner as the systems 64 and 66 by focusing on the relationship between input and output, but here, for convenience of explanation, these systems are used. Description of the system is omitted.
[0041]
The detection device 70 shown in FIG. In order to function as the detection device 70, the ECU 50 needs to acquire inputs and outputs for all of the systems 60 to 66.
[0042]
As described above, the calculation A / F that is input to the system 60 can be calculated based on the intake air amount Ga, the fuel injection amount, and the fuel adhesion amount. The intake air amount Ga can be detected by the air flow meter 18. Further, the fuel injection amount can be calculated based on the intake air amount Ga and the engine speed NE. Further, the fuel adhesion amount can also be calculated based on the intake air amount Ga, the engine speed Ne, the throttle opening TA, and the like. Therefore, the ECU 50 can obtain an input (calculation A / F) to the system 60.
[0043]
The A / F sensor value で which is the output of the system 60, that is, the output of the air-fuel ratio sensor 26 is supplied to the ECU 50. Therefore, the ECU 50 can also acquire the output of the system 60.
[0044]
The input of system 62 is the same A / F sensor value as the output of system 60. Therefore, the ECU 50 can acquire the input of the system 62. Also, the output O of the system 62₂The sensor value is supplied to the ECU 50. Therefore, the ECU 50 can also acquire the output of the system 62.
[0045]
Further, the ECU 50 is also supplied with the outputs of the battery ammeter 36 and the battery voltmeter 38, which are the inputs and outputs of the system 64, and the sensor values of the torque sensor 40 and the wheel speed sensor 42, which are the inputs and outputs of the system 66. ing. Therefore, the ECU 50 can acquire inputs and outputs for all of the systems 60 to 66 shown in FIG.
[0046]
The ECU 50 diagnoses, in a time-division manner, whether or not each of the systems 60 to 66 is functioning properly by the method described with reference to FIGS. 1B and 2. More specifically, the ECU 50 sequentially selects the systems 60 to 66 as control target systems and, based on the inputs and outputs for each, uses a standard model and parameters for each system to diagnose the failure of those systems. Is executed in a time sharing manner.
[0047]
[Description of ECU operation]
FIG. 5 is a flowchart of a control routine executed by the ECU 50 to realize the above functions.
In the routine shown in FIG. 5, first, a system number k for specifying a control target system is initialized to 1 (step 100).
It is assumed that the system numbers are assigned in advance to all the systems included in the configuration shown in FIG. 2 so as not to be duplicated.
[0048]
Next, the k-th system is selected as a control target system, and a command to start a failure diagnosis for the system is issued (step 102).
[0049]
When the failure diagnosis is started, first, input / output data for the system selected as the control target system is measured for a predetermined number of samples (step 104).
If the predetermined number of samples is, for example, three, in this step 104, three inputs x (t), x (t-1), x (t-2) and three outputs y (t), y (t) -1) and y (t-2) are measured.
[0050]
Next, parameters fk1, fk2,... Gk1, gk2,... Stored in the ECU 50 in advance as parameters corresponding to the control target system are read (step 106).
These parameters are values obtained by using a least-squares method described later so as to conform to the state (initial state) of each system when the vehicle is shipped from the factory. Hereinafter, these parameters stored in the ECU 50 in advance are referred to as “initial values of parameters”.
[0051]
In the routine shown in FIG. 5, the parameters fk1, fk2,... Gk1, gk2,... Read in step 106 are then substituted into the standard model shown in the above (Equation 1). A model of a linear approximation expression as shown in 2) or (Expression 3), that is, a model of a linear approximation expression representing the input / output relationship of the controlled system is generated. Hereinafter, this model is referred to as an “initial value corresponding model”.
Further, the actual input x (t), x (t−1),... And the actual output y (t), y (t−) measured in step 104 are added to the initial value corresponding model thus generated. By substituting 1),..., An estimated value of the output y (t + d) considering the system delay d is calculated (step 108).
[0052]
After the output y (t + d) at the time t + d is estimated by the processing in step 108, the output y (t + d) at the time t + d is actually measured. Then, the estimated value of the output y (t + d) is compared with the actually measured value (step 110).
[0053]
Next, it is determined whether or not the error of the estimated value of the output y (t + d) with respect to the actually measured value is equal to or smaller than a predetermined threshold (step 112).
[0054]
As a result, when it is determined that the error is equal to or smaller than the threshold value, it can be determined that the input / output relationship of the controlled system has not significantly changed from the input / output relationship in the initial state. In this case, it is determined that the control target system is normal, and the process of step 116 described later is executed.
[0055]
On the other hand, if it is determined in step 112 that the error between the measured value and the estimated value exceeds the threshold value, it is determined that the control target system has undergone a change that significantly changes the input / output relationship. it can. In this case, next, a process of detecting an abnormality and displaying a failure is performed on the system that has been set as the control target system in the current processing cycle (step 114).
[0056]
When the above-described series of processing is completed, it is determined whether or not the failure diagnosis processing has been properly completed for the k-th system (step 116).
[0057]
As a result, if it is determined that the processing for the k-th system has not been properly completed, the processing from step 102 onward is executed again. On the other hand, if it is determined that the process has been properly completed, it is next determined whether or not the system number k matches the total system number N (step 118).
[0058]
When it is determined that k = N is not established, it can be determined that a system for which the failure diagnosis has not yet been performed remains. In this case, after the system number k is incremented (step 120), the processing after step 102 is executed again.
[0059]
On the other hand, if it is determined in step 118 that k = N holds, it can be determined that the failure diagnosis has been performed for all the systems. In this case, the system number k is initialized to 1 in step 100, and the fault diagnosis for all the systems is restarted.
[0060]
As described above, according to the routine shown in FIG. 5, all the systems included in the configuration shown in FIG. 2 can be sequentially diagnosed in a time-division manner. Further, according to this routine, an initial value correspondence model used for failure diagnosis of each system can be generated by substituting the initial values of the parameters stored for each system into the standard model. Therefore, according to the apparatus of the present embodiment, all of a plurality of systems can be accurately diagnosed without a large calculation load and a large memory load.
[0061]
In the first embodiment described above, the memory that stores the standard model shown in the above (Equation 1) inside the ECU 50 corresponds to the “standard model storage unit” in the first invention. Also, in the first embodiment, the ECU 50 executes the processing of steps 100, 102, 118 and 120, whereby the “system selecting means” in the first invention executes the processing of step 104. Accordingly, the "actual input / output detecting means" in the first invention executes the processing of steps 108 to 114, thereby realizing the "individual system control means" in the first invention.
[0062]
In the first embodiment described above, the memory that stores the initial values of the parameters for the individual systems inside the ECU 50 corresponds to the “parameter initial value storage means” in the second invention. Further, in the first embodiment, the ECU 50 substitutes the initial values of the parameters in the standard model in step 108, thereby generating the “initial value corresponding model generating means” in the second invention. By calculating the output y (t + d) using the initial value correspondence model, the "output estimated value calculating means" in the second invention is realized. The "failure diagnosis means" in the second aspect of the present invention is realized by the ECU 50 executing the processing of steps 110 to 114.
[0063]
Embodiment 2 FIG.
Next, a second embodiment of the present invention will be described with reference to FIG. The device of the present embodiment can be realized by using the same configuration as the device of the above-described first embodiment, and causing the ECU 50 to execute the routine shown in FIG. 6 instead of the routine shown in FIG. it can.
[0064]
The apparatus according to the first embodiment described above is based on how much the estimated value of the output y (t + d) calculated using the initial value correspondence model includes an error with respect to the actually measured value of the output y (t + d). It is determined whether there is a failure in each system. On the other hand, the apparatus according to the present embodiment calculates the latest values of the parameters for the individual systems, that is, the latest parameters reflecting the effects of aging and failures on the vehicle, and the latest values are used as the initial values of the parameters. Based on the degree of deviation from the value, the presence or absence of a failure in each system is diagnosed.
[0065]
FIG. 6 is a flowchart of a control routine executed by the ECU 50 in the present embodiment in order to realize the above functions. In FIG. 6, steps that are the same as the steps shown in FIG. 5 are given the same reference numerals, and descriptions thereof are omitted or simplified.
[0066]
In the routine shown in FIG. 6, in step 104, a predetermined number of actual inputs x (t), x (t-1),... And actual outputs y (t), y (t-1),. Are measured, the latest values fk1 (t), fk2 (t),... Gk1 (t), gk2 (t),... Of the parameters at that time t for the system are estimated by the least squares method. Is performed (step 130).
[0067]
In the apparatus of the present embodiment, a standard model that can represent input / output relationships of all systems by exchanging parameters is defined. When the latest value of the parameter is properly obtained, the model obtained by substituting the latest value into the standard model (hereinafter referred to as “the latest value-capable model”) provides the current input / output relationship of the controlled system. It will be represented with high accuracy.
[0068]
In other words, the latest values of the parameters are the actual inputs x (t), x (t−1),... And the actual outputs y (t), y (t−1),. May be set to a value such that the estimated value of the output y (t + d) obtained by substituting the output y (t + d) accurately matches the actually measured value of the output y (t + d). In the above step 130, based on the standard model, the minimum input value is calculated using the actual inputs x (t), x (t-1),... And the actual outputs y (t), y (t-1),. By the multiplication method, the latest parameter value that minimizes the error between the estimated value of the output y and the actually measured value is estimated. The parameter estimation method using the least squares method will be described later in more detail.
[0069]
In the routine shown in FIG. 6, next, the latest value of the parameter estimated in step 130 is compared with the initial value of the parameter stored in the ECU 50 in advance (step 132).
[0070]
Next, it is determined whether or not the error between the latest value and the initial value of the parameter is equal to or less than a predetermined threshold (step 134).
[0071]
As a result, when it is determined that the error is equal to or smaller than the threshold value, it can be determined that the input / output relationship of the controlled system has not significantly changed from the input / output relationship in the initial state. In this case, it is determined that the control target system is normal, and thereafter, the process of step 116 is performed.
[0072]
On the other hand, if it is determined in step 134 that the error between the latest value and the initial value of the parameter exceeds the threshold value, it is determined that a change has occurred in the controlled system that significantly changes the input / output relationship. I can judge. In this case, the process of step 114 is thereafter performed for the control target system to execute the process of detecting an abnormality and displaying a failure.
[0073]
As described above, according to the routine shown in FIG. 6, it is possible to estimate the latest parameter value that matches the latest state of each system based on the actual input / output values and the standard model. Then, a failure of each system can be diagnosed based on an error between the latest value of the parameter and the initial value of the parameter.
[0074]
Further, according to the routine shown in FIG. 6, such a failure diagnosis can be sequentially executed for all the systems in a time-division manner. Therefore, according to the apparatus of the present embodiment, as in the case of the first embodiment, all of the plurality of systems can be accurately diagnosed without a large computational load and without a large memory load. can do.
[0075]
In the above-described second embodiment, the “individual system control unit” in the first invention is realized by the ECU 50 executing the processing of steps 130, 132, and 114.
[0076]
In the second embodiment described above, the memory that stores the initial values of the parameters for each system in the ECU 50 corresponds to the “parameter initial value storage means” in the third invention. Further, the “latest parameter calculation means” in the third invention by executing the processing of step 130 by the ECU 50 executes the processing of steps 132, 134 and 114 in the third invention. Failure diagnosis means "are each realized.
[0077]
Embodiment 3 FIG.
Next, a third embodiment of the present invention will be described with reference to FIGS. The device according to the present embodiment uses the same configuration as the device according to the above-described first embodiment, and causes the ECU 50 to execute the routine shown in FIG. 7 and the routine shown in FIG. 8 instead of the routine shown in FIG. It can be realized by doing.
[0078]
The apparatus according to the above-described second embodiment estimates the latest values of parameters for each system on the vehicle as appropriate, and diagnoses the failure of the system based on the estimated values. If the latest value of the parameter deviates from its initial value, it can be determined that a non-negligible change has occurred in the state of the system. When such a change occurs in the system, a situation in which a desired target output cannot be obtained may occur if control designed in accordance with an initial state is performed. Therefore, the apparatus according to the present embodiment is configured to modify the control for the system based on the latest value of a system in which a large change is recognized in the latest value of the parameter based on the latest value so that the target output can be obtained with high accuracy. I have.
[0079]
FIG. 7 is a flowchart of a control routine executed by the ECU 50 in the present embodiment to realize the above functions. In FIG. 7, steps that are the same as the steps shown in FIG. 5 or FIG. 6 are given the same reference numerals, and descriptions thereof will be omitted or simplified.
[0080]
The routine shown in FIG. 7 is the same as the routine shown in FIG. 6 except that step 114 shown in FIG. 6 is changed to step 140. That is, in the routine shown in FIG. 7, after the latest value of the parameter corresponding to the control target system is estimated by the same method as in the second embodiment (step 130), the estimated value greatly differs from the initial value. If there is (Steps 132 and 134), the start of the optimal control is instructed for the system selected as the control target system (Step 140).
[0081]
FIG. 8 is a flowchart of a control routine executed by the ECU 50 to realize the optimal control. This routine is a routine that is repeatedly executed continuously when the start of the optimal control is instructed in the above-described step 140, with respect to the system from which the instruction is issued.
[0082]
In the routine shown in FIG. 8, first, the latest value θt of the parameter estimated by the routine shown in FIG. Where θt^T= [Fk1 (t), fk2 (t), ..., gk1 (t), gk2 (t), ...] (step 150).
[0083]
Next, a target output ωt + d to be realized in this system is read. Here, ωt + d is an output value to be realized at a time later than the current time t by the system delay time d (step 152).
[0084]
Next, based on the latest value corresponding model obtained by substituting the latest value θt of the parameter into the standard model and the target output ωt + d, the current input x (t) necessary to realize the target output ωt + d is , Is calculated as the corrected input value u (t) (step 154).
A specific method for obtaining the corrected input value u (t) from the latest parameter value θt will be described later in detail together with a method for obtaining the latest parameter value θt.
[0085]
For example, when the system 60 (the system including the internal combustion engine 10 and the air-fuel ratio sensor 26) shown in FIG. 4 is selected as the system to execute the optimal control, the corrected input value u (t) is determined in step 154. As a result, a calculation A / F for obtaining a target A / F sensor value is obtained. In addition, the system 62 (catalyst 24 and O₂If the system including the sensor 28) is a system to execute the optimal control, the target O₂An A / F sensor value (exhaust air-fuel ratio) for obtaining a sensor value is obtained as a corrected input value u (t).
[0086]
After the corrected input value u (t) is obtained as described above, a corrected value of the control parameter for realizing the corrected input value u (t) is calculated (step 156).
For example, in the case where the system 60 is to execute the optimal control in FIG. 4, the control parameters (intake air amount Ga, fuel injection amount, throttle opening TA, engine speed NE, etc.) that affect the calculation A / F ) Is calculated. When the system 62 shown in FIG. 4 is a system that should execute optimal control, a correction value of a control parameter (intake air amount Ga or fuel injection amount) that affects the A / F sensor value is calculated.
[0087]
When the corrected value of the control parameter is calculated in step 156, control of various actuators and the like (such as the fuel injection valve 22) is performed according to the corrected value. As a result, the corrected input value u (t) is input to the system determined to execute the optimal control, and as a result, the target output is generated by the system.
[0088]
As described above, according to the routine shown in FIG. 7, it is determined whether the control target system needs the optimal control based on whether the latest value of the parameter greatly deviates from the initial value. Can be. Further, according to the routine shown in FIG. 7, such a determination can be sequentially performed for all the systems in a time sharing manner using only a single standard model. For this reason, according to the apparatus of the present embodiment, it is possible to select a system that requires optimal control without a large calculation load and without a large memory load.
[0089]
Further, according to the routine shown in FIG. 8, the control parameters are corrected so that the corrected input value u (t) is input to the systems requiring the optimal control, and as a result, those systems are changed under the current situation. Can generate a target output. For this reason, according to the device of the present embodiment, all the systems can function stably over time without a large calculation load or memory load.
[0090]
In the third embodiment described above, the “individual system control means” in the first invention is realized by the ECU 50 executing the processes of steps 130 to 134 and 140 and the processes of steps 150 to 156. ing.
[0091]
In the third embodiment described above, the ECU 50 executes the process of step 130 so that the “latest parameter calculating means” in the fourth invention calculates the latest value θt of the parameter read in step 150. The “latest value corresponding model generating means” in the fourth aspect of the present invention performs the processing of step 152 by substituting the latest model corresponding to the standard model to generate the latest value corresponding model. The “detection means” executes the processing of step 154, and the “correction input calculation means” in the fourth invention executes the processing of step 156, thereby “control correction means” in the fourth invention. Has been realized respectively.
[0092]
[Conceptual explanation of optimal control (minimum dispersion control)]
Hereinafter, the concept of the optimal control (minimum variance control) executed in the third embodiment (see FIG. 8) will be described. Here, for the sake of simplicity, a case where the dead time (delay d) of the system is known will be described.
[0093]
Now, the input / output relationship of the system
A (q^-1) Y_k= B (q^-1) Q^{− d}u_k+ C (q^-1) E_k(1)
Let it be represented by Where u_kIs the input to the system at time k, y_kIs the output of the system at time k, d is the delay of the system (known), e_kIs the noise error at time k. However,
A (q^-1) = 1 + α₁q^-1+ Α₂q^{− 2}+ Α₃q^{− 3}+ ... α_nq^{− n}
B (q^-1) = Β₀+ Β₁q^-1+ Β₂q^{− 2}+ Β₃q^{− 3}+ ・ ・ ・ Β_mq^{− m}(M = nd)
C (q^-1) = 1 + γ₁q^-1+ Γ₂q^{− 2}+ Γ₃q^{− 3}+ ・ ・ ・ Γ_nq^{− n}(2)
Here, α in equation (2)₁~ Α_n, Β₁~ Β_m, Γ₁~ Γ_nAre the respective coefficients.
[0094]
Expression (1) is obtained by using expression (2).

Can be written.
[0095]
At this time, the control target value is ω_kage
J = E (y_{k + d}−ω_{k + d})²(4)
Control (minimum dispersion control) that minimizes the evaluation function is obtained by the following equations (5) and (6). Here, E is an operator that takes an expected value.
[0096]
(Equation 1)

[0097]
Here, F (q^-1), G (q^-1) Is a polynomial satisfying the following relational expression.
F (q^-1) = A (q^-1) E (q^-1) + Q^{− d}F (q^-1) (7)
G (q^-1) = B (q^-1) E (q^-1) (8)
However,
E (q^-1) = 1 + δ₁q^-1+ Δ₂q^{− 2}+ Δ₃q^{− 3}+ ... δ_d-1q^{− (D-1)}(9)
F (q^-1) = F₀+ F₁q^-1+ F₂q^{− 2}+ F₃q^{− 3}+ ... δ_n-1q^{− (N-1)}(10)
[0098]
The above equations (4) to (6) are equations for generalized minimum dispersion control.
Where y in equation (4)_{k + d}−ω_{k + d}Is generally unstable, so instead
φ_{k + d}= Py_{k + d}+ Qu_k−Rω_{k + d}
Φ given by_{k + d}In many cases, an optimal control that minimizes the variance is used. In the third embodiment, the optimal control is used.
[0099]
In this case, for the system given by equation (1), the following evaluation function
J = E (φ_{k + d})²= E (Py_{k + d}+ Qu_k−Rω_{k + d})²(11)
Input u to minimize_kCan be represented by the following equation.
[0100]
(Equation 2)

[0101]
However, E (q^-1) Is (d-1) order, F (q^-1) Has an order of (the order of P + the order of C−1). Further, P satisfies the following relationship.
P (q^-1) C (q^-1) = A (q^-1) E (q^-1) + Q^{− d}F (q^-1) (13)
Here, when there is no noise, C (q^-1) = 1.
Equations (11) and (12) above are equations representing optimal control.
[0102]
[Explanation of system parameter estimation by least squares method]
Next, a description will be given of a method of calculating parameters by the least-squares method used in estimating an initial value or a latest value of a parameter corresponding to each system in the first to third embodiments.
[0103]
If the parameters to be substituted into the standard model are uncertain, unknown parameters must be estimated from input / output data for the system. Here, F (q^-1), G (q^-1) = B (q^-1) E (q^-1) And C (q^-1A method for estimating parameters for obtaining the coefficient of ()) will be described.
[0104]
Here, it is assumed that the following conditions are satisfied.
φ_{k + d}= Y_{k + d}
P (q^-1) = 1
Q (q^-1) = Γ (1−q^-1)
R (q^-1) = 1
C (q^-1) = 1 (14)
[0105]
From now on, F (q^-1), G (q^-1) Is θ,
J = E (φ_k−ψ_kdθ)²λ^tk(15)
Is estimated so that is minimized. Here, λ is a predetermined value satisfying the relationship of 0 <λ ≦ 1. Also, ψ_kAnd θ each satisfy the following relationship.
ψ_k ^T= [Y_k, Y_{k − 1}, Y_{k − 2}, ..., u_k, U_{k − 1}, U_{k − 2}, ...]
θ^T= [F₀, F₁, F₂, ..., g₀, G₁, G₂, ...]
However, F (q^-1), G (q^-1) Can be expressed as follows.
[0106]
(Equation 3)

[0107]
The least squares estimation value θ hat minimizing J in the above equation (15) is given as follows.
[0108]
(Equation 4)

[0109]
Θ represented by the above equation (16) is a parameter of the system estimated by the least squares method. In addition, the system parameter F (q^-1), G (q^-1) = B (q^-1) E (q^-1) Is substituted into the following equation, it is possible to calculate a desirable input that minimizes the evaluation function of the equation (11). Here, the following equation is an equation obtained by substituting the condition shown as equation (14) into equation (12).
[0110]
(Equation 5)

[0111]
[Example of parameter estimation and optimal control when delay d = 3]
To supplement the above description regarding parameter estimation and optimal control, an example of parameter estimation and an example of optimal control will be described for a system delay d = 3.
[0112]
It is assumed that the input / output relationship of the system for designing the control system is expressed as follows, as in the case of the above equation (1).
A (q^-1) Y_k= B (q^-1) Q^{− d}u_k+ C (q^-1) E_k
Where u_kIs the input to the system at time k, y_kIs the output of the system at time k, d is the delay of the system (known), e_kIs the noise error at time k. However,
A (q^-1) = 1 + α₁q^-1+ Α₂q^{− 2}+ Α₃q^{− 3}(Assuming n = 3)
B (q^-1) = Β₀(M = nd = 3-3 = 0)
C (q^-1) = 1 (assuming no noise)
[0113]
Now, from the relationship of the following formula,
(Equation 6)

ψ_kAnd θ can be expressed as follows, respectively.
ψ_k ^T= [Y_k, Y_k-1, Y_k-2, U_k, U_k-1, U_k-2]
θ^T= [F₀, F₁, F₂, G₀, G₁, G₂] Φ_{k + 3}= Y_{k + 3}
[0114]
Furthermore, the evaluation function
J = E (φ_k−ψ_kdθ)²λ^tk= E (y_k−ψ_k-3θ)²λ^tk
Is obtained from the equations (16) to (18) as follows.
[0115]
(Equation 7)

[0116]
Hereinafter, the above equations (P1) to (P3) will be specifically described.
Since it has been known that P (t) is a symmetric matrix, P (t) is set as follows, and (1) to (4) in (P3) and ( Equations (P1) to (P3) are obtained.
[0117]
(Equation 8)

[0118]
(Equation 9)

[0119]
(Equation 10)

[0120]
[Equation 11]

[0121]
(Equation 12)

[0122]
From the above (1) to (4), the equations (P1) to (P3) can be described as follows. First, (P1) and (P2) can be described as the following equation (P4).
(Equation 13)

[0123]
here,

[0124]
Next, the equation (P3) can be described as the following equation (P5).
[Equation 14]

[0125]
Θt = [f obtained by the above equations (P4) and (P5)₀(T), f₁(T), f₂(T), g₀(T), g₁(T), g₂(T)] is determined so as to satisfy the condition of the following equation (P6).
y (t + 3)-｛f₀(T) y (t) + f₁(T) y (t-1) + f₂(T) y (t-2) + g₀(T) u (t) + g₁(T) u (t-1) + g₂(T) u (t−2)｝ = 0 (P6)
That is, θt = [f obtained by the above equations (P4) and (P5)₀(T), f₁(T), f₂(T), g₀(T), g₁(T), g₂(T)] are system parameters estimated by the least squares method.
[0126]
Next, the optimal control using the system parameter θt obtained as described above will be described.
In such optimal control, u (k) that satisfies the following evaluation function is obtained using the parameter θt.
J = E [y (k + 3) -ω (k + 3) + γ {u (k) -u (k-1)}]²(P7)
In the above equation (P7), y (k + 3) -ω (k + 3) is the difference between the output three ahead and the target value. U (k) -u (k-1) is a term representing the change of the input. Here, the reason that the term indicating the change of the input is included in the evaluation function is to stabilize the control.
[0127]
Assuming that y (k + 3) −ω (k + 3) + γ ｛u (k) −u (k−1) = 0 holds in the equation (P7), substituting the relation of the equation (P6) into the relation gives the following: Is obtained.
｛F₀(K) y (k) + f₁(K) y (k-1) + f₂(K) y (k-2) + g₀(K) u (k) + g₁(K) u (k-1) + g₂(K) u (k−2)｝ − ω (k + 3) + γu (k) −γu (k−1) = 0
[0128]
The above relational expression can be rewritten as follows.
｛G₀(K) + γ｝ u (k) = [ω (k + 3) −f₀(K) y (k) -f₁(K) y (k-1) -f₂(K) y (k-2)-｛g₁(K) -γ｝ u (k-1) -g₂(K) u (k-2)]
Therefore, u (k) can be expressed as follows.
[0129]
(Equation 15)

[0130]
From the above, in the case of the system with the delay d = 3, the corrected input value u (t) for setting the output y (t + 3) three ahead to the target output ω (t + 3) is expressed by the following equation (P8). Desired. The value obtained by the equation (P8) is a corrected input value for realizing optimal control in the third embodiment.
[0131]
(Equation 16)

[0132]
Embodiment 4 FIG.
Next, a fourth embodiment of the present invention will be described with reference to FIG. FIG. 9 is a diagram for explaining a method used by the device of the present embodiment to perform failure diagnosis for each of a plurality of systems. The configuration shown in FIG. 9 is the same as the configuration shown in FIG. 1B except that the standard model used in the detection device is replaced with a neural network from a linear approximation model.
[0133]
That is, the device of the present embodiment is the same as the device of any of the first to third embodiments, except that the ECU 50 uses the linear approximation shown in the above (Equation 1) as a standard model for estimating the input / output relationship of a plurality of systems. It can be realized by incorporating the function of the neural network instead of the formula.
[0134]
In the above-described first to third embodiments, the parameters of each system to be substituted into the standard model of the linear approximation formula are estimated by the least square method. On the other hand, in the neural network, a parameter (coupling coefficient) for each system can be obtained by a known method based on back propagation.
[0135]
In the apparatus of the present embodiment, the parameters (coupling coefficients) obtained by the above-described known method are stored in the ECU 50 in advance, and the ECU 50 executes the processing according to the routine shown in FIG. Functions similar to those of the first embodiment can be realized. In this case, based on how much error the output estimated on the actual vehicle by the neural network whose parameters are set to the initial value has relative to the output actually measured on the actual vehicle, Can be diagnosed.
[0136]
Further, in the apparatus of the present embodiment, the coupling coefficient obtained by the above-described known method is stored in the ECU 50 in advance, and a function for obtaining the coupling coefficient by the method is incorporated in the ECU 50. The processing according to the routine shown in FIG. 6 can also be executed. In this case, the same function as that of the second embodiment can be realized by the device of the present embodiment.
[0137]
In the routine shown in FIG. 6, it is required to obtain the latest value of a parameter (coupling coefficient) to be substituted into the standard model on the actual vehicle. Since the function of obtaining the coupling coefficient by back propagation is incorporated in the ECU 50, the requirement can be satisfied. Further, in the routine shown in FIG. 6, it is required to determine the state of the system based on a comparison between the latest value of the parameter obtained on the actual vehicle and its initial value. In the present embodiment, the ECU 50 can satisfy such a request by comparing the state expression (weight) of the intermediate layer.
[0138]
Further, in the device of the present embodiment, the coupling coefficient obtained by the above-described known method is stored in the ECU 50 in advance, and a function for obtaining the coupling coefficient by the method is incorporated in the ECU 50. Processing according to the routine shown in FIGS. 7 and 8 can also be executed. In this case, the device of the present embodiment can realize the same function as that of the third embodiment.
[0139]
In the routine shown in FIG. 7, it is required to obtain the latest value of a parameter (coupling coefficient) to be substituted into the standard model on the actual vehicle. Since the function of obtaining the coupling coefficient by back propagation is incorporated in the ECU 50, the requirement can be satisfied. Further, in the routine shown in FIG. 7, it is required to determine the state of the system based on a comparison between the latest value of the parameter obtained on the actual vehicle and its initial value. In the present embodiment, the ECU 50 can satisfy such a request by comparing the state expression (weight) of the intermediate layer.
[0140]
Further, in the routine shown in FIG. 8, it is required to execute the optimum control based on the latest value θt of the parameter obtained on the actual vehicle. In the present embodiment, the ECU 50 satisfies such a requirement by obtaining the corrected input value u (t) so as to obtain the target output ωt + d by using a neural network into which the latest value of the parameter (coupling coefficient) is substituted. Can be.
[0141]
According to the neural network, it is possible to accurately represent a nonlinear input / output relationship that is difficult to represent with a linear approximation formula. For this reason, according to the device of the present embodiment, it is possible to obtain compatibility with more systems than the devices of the first to third embodiments. Therefore, according to the device of the present embodiment, higher versatility can be secured as compared with the devices of the first to third embodiments.
[0142]
Note that a method of replacing a model based on a linear approximation formula with a model of a neural network is disclosed in detail in, for example, Japanese Patent Application Laid-Open No. 8-117199 or EP0699413A1.
[0143]
By the way, in Embodiments 1 to 4 described above, the control target is limited to a plurality of systems mounted on the vehicle, but the present invention is not limited to this. That is, the present invention is not limited to a system on a vehicle, but may be applied to a plurality of systems installed in a plant facility or the like.
[0144]
【The invention's effect】
Since the present invention is configured as described above, it has the following effects.
According to the first aspect, by sequentially setting each of the plurality of systems as a system to be controlled, desired control for each system can be sequentially performed in a time-division manner. For this reason, according to the present invention, desired control can be performed for all systems without a high calculation load. Further, according to the present invention, it is possible to control each system by using a standard model which is adapted to the state of each system by exchanging parameters. Therefore, according to the present invention, it is possible to execute desired control for all systems without a large memory load.
[0145]
According to the second invention, an initial value corresponding model of the system to be controlled can be generated by applying the initial value of the parameter corresponding to the control target system to the standard model. Further, by using the initial value correspondence model and the actual input and actual output to the system, an output estimation value of the system can be calculated. Then, it is possible to diagnose whether or not the system has a failure based on whether or not the output estimation value obtained in this way is significantly different from the actual output.
[0146]
According to the third aspect, based on the standard model and the actual input and actual output to the control target system, it is possible to calculate the latest value of the parameter that matches the latest state of the system. Then, based on whether or not the latest value of the parameter obtained in this way is significantly different from the initial value of the parameter, it is possible to diagnose the presence or absence of a system failure.
[0147]
According to the fourth invention, based on the standard model and the actual input and actual output to the control target system, the latest value of the parameter suitable for the latest state of the system can be calculated. Further, by applying the latest values of the parameters obtained in this way to the standard model, a latest value corresponding model corresponding to the latest state of the system can be generated. According to the present invention, a correction input to be input to the system is obtained based on the latest value corresponding model and the target output, and necessary control correction is performed so that the correction input is realized. Therefore, according to the present invention, it is possible to always maintain the optimum control state without being affected by the aging of the system.
[0148]
According to the fifth aspect, since the standard model is a linear approximation, various arithmetic processing can be easily performed.
[0149]
According to the sixth aspect, since the standard model is a neural network, even when a plurality of systems include a system in which the input / output relationship is non-linear, high control accuracy is achieved for all the systems. Can be secured.
[Brief description of the drawings]
FIG. 1 is a diagram for describing an outline of control executed in a first embodiment of the present invention.
FIG. 2 is a diagram showing an input / output relationship for an arbitrary system.
FIG. 3 is a diagram for explaining a specific configuration of the first embodiment of the present invention.
FIG. 4 is a diagram specifically illustrating a configuration of a control system formed in the device according to the first embodiment of the present invention.
FIG. 5 is a flowchart of a control routine executed in the first embodiment of the present invention.
FIG. 6 is a flowchart of a control routine executed in a second embodiment of the present invention.
FIG. 7 is a flowchart of a first control routine executed in a third embodiment of the present invention.
FIG. 8 is a flowchart of a second control routine executed in Embodiment 3 of the present invention.
FIG. 9 is a diagram for describing an outline of control executed in a fourth embodiment of the present invention.
[Explanation of symbols]
10 internal combustion engine
18 air flow meter
21 ° throttle sensor
24 catalyst
26 air-fuel ratio sensor
28 O₂Sensor
34 ° rotation speed sensor
36 battery ammeter
38 battery voltmeter
40 ° torque sensor
42 wheel speed sensor
50 ECU (Electronic Control Unit)
60, 62, 64, 66 system
70 ° detector
x, x (t), u_kInput
y, y (t), y_kOutput
fn, gn, f1n, g1n, f2n, g2n, θ parameters
d System delay time
ω_kTarget output

Claims (6)

An integrated control device for controlling a plurality of systems,
Standard model storage means for storing a standard model capable of satisfying an input / output relationship in each of the plurality of systems by exchanging parameters;
System selecting means for selecting each of the plurality of systems in a time-division manner and sequentially as a control target system;
Actual input / output detection means for detecting an actual input and an actual output in the controlled system,
Based on the actual input and the actual output detected for the control target system, based on the standard model, an individual system control unit that executes control on the control target model,
An integrated control device for a plurality of systems, comprising: With parameter initial value storage means for storing an initial value of the parameter corresponding to each of the plurality of systems,
The individual system control means,
Initial value corresponding model generation means for applying an initial value of the parameter corresponding to the control target system to the standard model to generate an initial value corresponding model of the system,
An output-estimated-value calculating unit that calculates an output estimated value of the controlled system based on the initial-value-corresponding model and the actual input and the actual output;
Failure diagnosis means for performing failure diagnosis of the controlled system based on a comparison result between the output estimation value and the actual output;
The integrated control device for a plurality of systems according to claim 1, further comprising: With parameter initial value storage means for storing an initial value of the parameter corresponding to each of the plurality of systems,
The individual system control means,
The standard model, based on the actual input and the actual output, the latest parameter calculating means for calculating the latest value of the parameter that matches the latest state of the controlled system,
Failure diagnosis means for performing a failure diagnosis of the controlled system based on a comparison result between the initial value and the latest value of the parameter,
The integrated control device for a plurality of systems according to claim 1, further comprising: The individual system control means,
The standard model, based on the actual input and the actual output, the latest parameter calculating means for calculating the latest value of the parameter that matches the latest state of the controlled system,
A latest value corresponding model generation unit configured to apply a latest value of the parameter corresponding to the control target system to the standard model to generate a latest value corresponding model of the system, further comprising:
For a system in which the latest value corresponding model has been obtained, target output detection means for detecting a target output,
Correction input calculation means for calculating a correction input to be supplied to the system in order to obtain the target output, based on the latest value corresponding model,
Control correcting means for correcting the control related to the actual input so that the actual input to the system matches the corrected input;
The integrated control device for a plurality of systems according to claim 1, further comprising: The composite control device for a plurality of systems according to any one of claims 1 to 4, wherein the standard model is a linear approximation formula. The multi-system combined control apparatus according to any one of claims 1 to 4, wherein the standard model is a neural network.

Images