第六章 地下水參數檢定系統應用於濁水溪流域
6.3 綜合討論
6.3.4 專家系統之推論邏輯說明
由以上全區均方根誤差以及 Q 值檢定過程可發現，約在第 72 次至第 78 次調整時，全區均方根誤差及部份區域 Q 值調整量均有較大幅變動，因 此將選定六合(1)之第 72 次至第 78 次調整觀察專家系統之推論進行解釋，
其推論過程如圖 6.3-11 所示。第 72 次調整方向為正，調整量不變。而第
73 次 及 第 74 次 調 整 ， 兩 次 誤 差 改 變 量 均 低 於 設 定 門 檻 ( 即 Rule calibrating_adjust 之判斷結果)，因此須放大調整量。至第 75 次調整，調整 量已不需再放大，接著判斷調整方向與上次相同，調整量不變。第 76 次調 整，調整方向改變，調整量減半。而第77 次調整時為抽乾的狀態，所以調 整方向再改為正，而因調整方向改變，所以調整量減半。第 78 次調整時，
調整方向又與上次相反，所以調整量再減半。這一段專家系統解釋輸出，
可以瞭解當誤差改變率低於設定門檻時，調整量會放大，而再持續調整幾 次後，調整量又會再逐漸縮小，使最終答案趨近於合理值。
iteration: 72 z: 1 x: 51 y: 49 sim_head: 36.4599
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 15.625) MSG << put-modified_q ED:2 (<Instance-thefact> 15.625)
iteration: 73 z: 1 x: 51 y: 49 sim_head: 36.5931
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 calibrating_increase: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 78.125) MSG << put-modified_q ED:2 (<Instance-thefact> 78.125)
iteration: 74 z: 1 x: 51 y: 49 sim_head: 37.2331
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 calibrating_increase: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 390.625) MSG << put-modified_q ED:2 (<Instance-thefact> 390.625)
iteration: 75 z: 1 x: 51 y: 49 sim_head: 29.5186
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 390.625) MSG << put-modified_q ED:2 (<Instance-thefact> 390.625)
iteration: 76 z: 1 x: 51 y: 49 sim_head: 56.8244
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
FIRE 15 negi_min_deltQ: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -195.3125) MSG << put-modified_q ED:2 (<Instance-thefact> -195.3125)
iteration: 77 z: 1 x: 51 y: 49 sim_head: -888
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
FIRE 15 posi_min_deltQ: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 97.65625) MSG << put-modified_q ED:2 (<Instance-thefact> 97.65625)
iteration: 78 z: 1 x: 51 y: 49 sim_head: 43.2029
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
FIRE 15 negi_min_deltQ: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -48.828125) MSG << put-modified_q ED:2 (<Instance-thefact> -48.828125)
圖 6.3-11 六合(1)之第 72~78 次推論過程
接著再觀察同一次調整，不同區域之變化差別，乃選定六合(1)、田中 (1)、西螺(1)、竹塘(1)、合興(1)、香田(1)、豐榮(1)，這 7 個區域之第 50 次 調整，即此系統之調整初期來進行觀察，其推論過程如圖 6.3-12 所示。可 知於第50 次調整時，各參數調整量均不變，保持與上次調整量相同，其主 因為於調整初期，各參數均持續向合理值接近而未有震盪情形，因此調整 量(ΔQ)修正至一個適當值後，便會維持於此穩定值。此外從專家系統第 50 次調整所解出之調整量也可看出，位於上游之六合(1)、田中(1)在調整初期 調整量較大，而位於中、下游之西螺(1)、竹塘(1)、合興(1)、香田(1)、豐榮 (1)，初期調整量較小，此亦證實上游會先大幅調整之傾向。
iteration: 50 number: 竹塘(1) z: 1 x: 37 y: 39 sim_head: 20.3345
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -7.8125) MSG << put-modified_q ED:2 (<Instance-thefact> -7.8125)
iteration: 50 number: 豐榮(1) z: 1 x: 25 y: 46 sim_head: 1.60761
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 3.90625) MSG << put-modified_q ED:2 (<Instance-thefact> 3.90625)
iteration: 50 number: 田中(1) z: 1 x: 54 y: 39 sim_head: 34.5468
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 62.5) MSG << put-modified_q ED:2 (<Instance-thefact> 62.5)
iteration: 50 number: 西螺(1) z: 1 x: 41 y: 46 sim_head: 24.6618
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -7.8125) MSG << put-modified_q ED:2 (<Instance-thefact> -7.8125)
iteration: 50 number: 香田(1) z: 1 x: 32 y: 37 sim_head: 18.9913
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -7.8125) MSG << put-modified_q ED:2 (<Instance-thefact> -7.8125)
iteration: 50 number: 合興(1) z: 1 x: 41 y: 35 sim_head: 18.3421
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_minus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> -7.8125) MSG << put-modified_q ED:2 (<Instance-thefact> -7.8125)
iteration: 50 number: 六合(1) z: 1 x: 51 y: 49 sim_head: 36.0316
FIRE 1 initialization: [thefact]
FIRE 10 calibrating_adjust: [thefact]
FIRE 11 belonging: [thefact]
FIRE 12 first_dry: [thefact]
FIRE 13 calibrating_direction: [thefact]
FIRE 14 calibrating_plus_correction: [thefact]
MSG >> put-modified_q ED:2 (<Instance-thefact> 62.5) MSG << put-modified_q ED:2 (<Instance-thefact> 62.5)
圖 6.3-12 選定區域之第 50 次推論過程
以上專家系統之推論過程可說明每次調整時所啟動的規則，因此使用 者僅需回溯推論過程，即可瞭解如此調整之根據及調整後之意義。往後若 對任何區域之調整過程有疑點時，均可使用此機制釋疑，並讓使用者能完 全掌握最後答案之演繹。