JP4146053B2 - Flow prediction method in dam or river - Google Patents

Description

【００３２】
ここで、得られた流下時間に応じて各河川１〜３の流域平均雨量を算出する。本実施形態において、雨量情報が１時間単位のデータであった場合、小数点以下を無視すると、河川１は河川３より流下時間が約１時間長く、河川２は河川３より約２時間長いものとして、流域平均雨量を以下の式により算出する。
流域平均雨量＝（河川１の流域面積×河川１の流域の１時間前の雨量＋河川２の流域面積×河川２の流域の２時間前の雨量＋河川３の流域面積×河川３の流域の現在の雨量）／全河川の流域面積
つまり、流域面積を重みとした加重平均により流域平均雨量を求める。こうして得られた流域平均雨量を用いることで、河川１〜３からダムへほぼ同時に流入する水量を予測することができ、時間遅れや齟齬のない流量予測が可能になる。
このようにして算出した流域平均雨量は、例えば参考形態１や参考形態２において累積雨量や累積雨量変化分を求める場合に適用することができる。
【００３３】
次に、請求項２の発明の実施形態を説明する。
本発明は、ダム流量との相関が高い累積時間ひいては累積雨量を算出するためのものである。累積時間が長過ぎても短過ぎても流量との相関が低く、流量の予測に用いることができない。本発明では累積雨量と流量との相関係数を用いることにより、適切な累積時間を算出するようにした。
【００３４】
以下において、〜時間累積雨量とは、降雨開始からの累積雨量ではなく、河川流域のｎ時間前からｍ時間前までの毎時間の累積雨量をいう。例えば、１時間累積雨量とはｎ時間前から（ｎ−１）時間前までの累積雨量であり、２時間累積雨量とはｎ時間前から（ｎ−２）時間前までの累積雨量である。従って、ｎ時間前を基準とした２時間累積雨量は１時間累積雨量を含むことになる。
【００３５】
図６（ａ），（ｂ）は、雨量の累積時間を算出するための累積雨量及び流量の時間変化を示す図である。一般に、累積雨量と流量との関係は降雨状況により変化する。すなわち、地面が乾いている時、流量は長時間（累積時間が長い）の累積雨量との相関が高くなり、地面が湿っている時、流量は短時間（累積時間が短い）の累積雨量との相関が高くなる傾向にある。
なお、図６（ａ），（ｂ）における１時間累積雨量、２時間累積雨量、流量は何れも実績値であり、過去の１時間あたり（図６（ａ））または２時間あたり（図６（ｂ））の累積雨量の変化に遅れて流量の変化が表れることを示している。
【００３６】
本実施形態における累積時間の算出は、次の手順による。
（１）図６（ａ）に示す１時間累積雨量のパターンを、時間軸に沿って流量パターン方向へ１時間ずつ移動させながら、両パターンの相関係数を求める。つまり、図４と同様の手順であり、相関係数についても−１〜１の範囲で変化する線形度の指標として求める。この場合の１時間累積雨量のパターンの移動時間は、前述した流下時間に相当する。
【００３７】
（２）同様に図６（ｂ）の２時間累積雨量、図示されていない３時間累積雨量，４時間累積雨量，……，ｋ時間累積雨量の各パターンを時間軸に沿って１時間ずつ移動させ、流量パターンとの間の相関係数を求める。なお、累積時間につき何種類の累積雨量パターンを用いるか（前述したｋを何時間までとするか）は、経験等により適宜決定される。
（３）次いで、得られた相関係数の中から最大値を選び、その最大値に対応する累積時間を探索する。つまり、相関係数が最大となった〜時間累積雨量の「〜時間」がそのまま累積時間となる。
【００３８】
例えば、ある河川流域の１時間〜１２時間累積雨量を１時間ずつ移動させた際の相関係数が表２で示される場合、１０時間累積雨量において移動時間が１時間のときの相関係数「０．９３４」がすべての相関係数の中で最大である。そこで、この河川については、累積雨量を算出するための累積時間を１０時間とする。以後は、この１０時間の累積雨量を使用し、流下時間を考慮して流量予測を行う。
【００３９】
【表２】

【００４０】
ここで、図７は、請求項１および請求項２の発明の有効性を説明するために、ダム上流域に設置してある複数の雨量計の測定値を単純平均した平均雨量と、ダム流入量実績値とを重ねて示した図である。図７の表示の基礎とした流入量実績値と降雨量とは図１の例と同一である。
雨は降ったり止んだりばらつきが大きいため、雨量計の測定値を単純平均しただけではばらつきが消えない。図７から分かるように、流入量が増加傾向のときにも平均雨量が減少してしまう場合もあり、この平均雨量から流入量の増減傾向を把握するのは困難である。
【００４１】
これに対し、参考形態３により河川流域ごとの流下時間を自動的に算出し、請求項１の発明により、流域面積を重みとした加重平均から流域平均雨量を算出し、更に請求項２の発明により適切な累積時間による累積雨量に換算して表示すれば、累積雨量の変化と流入量の増減傾向との相関関係が明瞭になり、累積雨量から流入量の増減傾向を容易に把握可能として流量予測に役立てることができる。
【００４２】
最後に、前記請求項２の発明の実施形態を流入量予測モデルに適用した場合の予測結果を検討する。ここでは、流入量予測モデルとしてニューラルネットワークを使用した。ニューラルネットワークを流入量予測に適用する場合には、複数時間分の雨量情報が必要である。また、何時間分の雨量データを使用するかという入力時間の決定方法としては、運用者が経験的に決める方法（平成５年電気学会電力・エネルギー部門大会NO.185,一柳ほか)、雨量の流下時間で決める方法(特願平10-187108号等)、流量との相関が高い累積雨量の累積時間で決める方法（前記請求項３の発明）がある。
【００４３】
ここでは、２つの検討事例１，２について検証してみる。
検討事例１では、図８に示す予測モデルを使用し、入力層Iに現在時から１時間おきに６時間前までの流域平均雨量差分を入力する。なお、IIは中間層、IIIは予測値としての、現在時から１時間後の流入量変化分（前述した流量変化分であるＡ・Δｃ＋Ｂに相当）が出力される出力層である。６時間前までとしたのは、対象流域の流下時間及び運用者の経験上の値に基づくものである。
【００４４】
検討事例２では、図９に示す予測モデルを使用し、現在時から１時間おきに２０時間前までの流域平均雨量差分を入力する。２０時間という値は、請求項２の発明により算出した累積時間である。出水事例により累積時間は異なるが、ここでは、過去の事例の中で最長の時間を選択した。平均的な事例では、前述の表２のように１０時間程度の値となる。
図８，図９の二つの予測モデルの入出力因子をまとめると表３のようになり、これらの入出力因子のもとで過去の複数の事例を十分に学習させ、２つの出水事例を予測させた。
【００４５】
【表３】

【００４６】
表３において、流域平均雨量差分を求める場合の流域平均雨量は、河川が複数ある場合の流下時間による調整をしていない値である。つまり、各河川の同時刻の雨量にそれぞれの河川の流域面積を乗じたものをすべて加算し、その値を全流域面積によって除算した値である。
また、例えば１時間前までの流域平均雨量差分は、現在の流域平均雨量から１時間前の流域平均雨量を減じた値である。
【００４７】
図１０（ａ）は、図８の予測モデルによる出水事例１についての流入量予測値及び実績値を示し、図１０（ｂ）は同じく出水事例２についての流入量予測値及び実績値を示している。同様に、図１１（ａ），（ｂ）は図９の予測モデルによる出水事例１，出水事例２についての流入量予測値及び実績値である。
流入量を予測する具体的な方法は、予測モデルに流域平均雨量差分を入力して求めた１時間後の流入量変化分（予測値）を現在の流入量に加算して１時間後の流入量を予測する方法をとり、これを図１０，図１１の時間軸方向に沿って繰り返し行う。
【００４８】
検討事例１，２について予測誤差を評価した結果を表４及び表５に示す。表４は出水事例１，表５は出水事例２に関するものである。
予測誤差の評価指標としては絶対値平均誤差〔％〕及び相関係数（予測値と実績値との相関係数）を用いた。ここで、絶対値平均誤差は小さい方がよく、相関係数は１．０に近い方がよい。
【００４９】
【表４】

【００５０】
【表５】

【００５１】
表４，表５から明らかなように、流域平均雨量差分として請求項２の発明における累積時間相当の差分データを用いた検討事例２の方が、表４，表５の何れにおいても検討事例１より絶対値平均誤差が小さく、相関係数が１．０に近くなっている。すなわち、累積時間を雨量の入力時間として選択する方が良好な結果が得られることが確認された。
【００５３】
【発明の効果】
以上述べたように、請求項１，２記載の発明は予測モデル構築時に利用できる手法である。予測モデルの構築に当たっては、その流域の特性の把握が必要である。
【００５４】
請求項１記載の発明は流域平均雨量の算出方法に関するものである。複数河川が合流している場合には河川ごとに流量を予測する手法が知られているが、個別河川の予測は個々の予測値のばらつきのため精度が高くなく、対象ダムの流量予測値の精度向上には限界がある。一方、複数河川を全体的に予測する手法では、データのばらつきが少なくなるが、河川ごとに特性の違いがある場合には精度が高くない。
本発明によれば、河川ごとの流下時間を考慮して流域平均雨量を算出可能であり、この流域平均雨量を使用することで高精度な予測が可能になる。
【００５５】
請求項２の発明は、累積雨量を求める際の累積時間を算出するものである。前述した参考形態１，２においては、流量との相関の高い累積雨量がデータとして必要である。この請求項２の発明によれば、相関の高い累積雨量が得られるため、例えば参考形態１，２に適用すれば比較的高精度な予測が可能である。また、他の予測モデルにおいても、数時間分の降雨量を入力情報として使用することが多いが、請求項２の発明によれば、最適な降雨量の入力範囲（何時間分の降雨量を予測に用いるか）を決定することが可能となる。
以上詳述したように、請求項１，２の発明は、どのような形状の河川にも適用できる汎用的かつ自動化が容易な方法である。
【図面の簡単な説明】
【図１】 参考形態１による画面の表示例を示す図である。
【図２】 参考形態１による画面の表示例を示す図である。
【図３】 参考形態２による画面の表示例を示す図である。
【図４】 参考形態３の原理の説明図である。
【図５】 参考形態３における流下時間と相関係数との関係を示す図である。
【図６】 請求項２の発明の実施形態において、累積時間を算出するための累積雨量及び流量の時間変化を示す図である。
【図７】 複数の雨量計の測定値を単純平均した平均雨量とダム流入量実績値とを重ねて示した図である。
【図８】 請求項２の発明の実施形態が適用される予測モデルの概念図である。
【図９】 請求項２の発明の実施形態が適用される予測モデルの概念図である。
【図１０】 各出水事例についての予測結果を示す図である。
【図１１】 各出水事例についての予測結果を示す図である。
【図１２】 従来のダム流量予測方法の説明図である。
【図１３】 従来のダム流量予測方法の説明図である。[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a method for automatically predicting the flow rate of a dam or river on a computer in a system control station, a power supply command station, a dam management station, a hydroelectric power station, or the like.
[0002]
[Prior art]
Prediction of dam inflow and upstream river flow (hereinafter referred to simply as flow when necessary) is important for the safety and economics of dam operation. Predicting the flow rate several hours ahead is necessary to appropriately determine the dam discharge rate during rainfall, which can help ensure the safety of the basin. In addition, forecasting several hours to several days ahead, especially the next day's flow rate forecast, can make a power generation plan appropriately and contribute to the effective use of hydro energy and the improvement of economic efficiency.
[0003]
Conventionally, this type of flow rate prediction is often performed according to the experience and intuitive knowledge of a skilled operator. For this reason, various methods using a tank model, a storage function method, a neural network, and the like have been proposed as an example of automating prediction work up to several hours ahead, and the prediction accuracy has been improved year by year.
By the way, when creating a dam discharge prediction model, it is indispensable to analyze the characteristics of rivers and basins, such as topography and geology. In other words, it is indispensable to analyze how rainfall in the upstream reaches the discharge. . In particular, the flow time, which is the time until the rainfall in the upstream area is reflected in the flow rate of the dam, etc. (the time until the flow rate increases after the rain) is an important factor. The difference between the peak time of the quantity and the peak time of the flow rate was determined and used as the flow-down time.
[0004]
Usually, a plurality of rain gauges are used for flow prediction. Since the information of this rain gauge has a large variation, it is general to average and use the values measured by a plurality of rain gauges. At this time, when there are a plurality of rivers upstream, a method of calculating and using the average rainfall for each river as will be described later (FIG. 12 to be described later), and a method of using the average basin rainfall for all the basins. (FIG. 13 described later) is known.
[0005]
[Problems to be solved by the invention]
By the way, operators who predict the flow rate of dams need a great deal of expertise and many years of experience, but in recent years, the number of skilled operators who have this knowledge is steadily decreasing. On the other hand, flow rate prediction is the foundation of dam operation, and improvement and automation of the prediction accuracy are desired.
In recent years, new technologies such as neural networks have significantly improved prediction accuracy, and even when existing tank models are used, it is now possible to make highly accurate predictions by making the configuration more complex. Yes.
However, these methods have high prediction accuracy, but due to the complexity of the configuration, it is difficult to understand for what reason the predicted value was obtained.
[0006]
In particular, short-term dam flow prediction is directly linked to dam gate operation, and it is required to clarify the reason for prediction and the reason for gate operation due to safety issues in the downstream area.
Therefore, a simple method that can be easily understood and a method that can clearly show the basis of the predicted value is required.
Also, when building a prediction model, it is necessary to analyze the characteristics of the target basin, and it is particularly important to calculate the flow time. In the conventional method, it cannot be said that the above-mentioned flow time is necessarily obtained, and the analysis accuracy is not high. In other words, in the conventional method, only the difference between the peak times of rainfall and flow is obtained, but if the rainfall peak time is not clear, it is impossible to calculate the flow time and the peak time is not clear. In some cases, automatic calculation by a computer was difficult.
[0007]
In addition, in the case of Fig. 12, in the dam in which multiple rivers flow, the flow rate is predicted for each river and the flow rate is summed to obtain the dam flow rate. In step (1), the average rainfall for each river 22 is obtained by the rain gauge. The flow rate of each dam 24 is predicted by adding the flow rate of each river 22 obtained by (1) in step (2). The flow meter 23 is used for obtaining the actual value.
According to this method, since the flow rate of the dam 24 is predicted based on the basin rainfall of each target river, an error in the predicted flow rate may increase due to variations in measured values of individual rain gauges.
[0008]
In order to solve this problem, there is a method in which the dam discharge is directly predicted after calculating the average rainfall, assuming that the flow time of all the catchments is the same. FIG. 13 shows the situation, and the dam flow rate is directly predicted from the average rainfall in the entire basin. However, in this case, if the flow time of each river 22 is different, the inflow time to the dam 24 is shifted, so that the prediction accuracy tends to be low.
[0009]
The present invention was made in order to solve the above-mentioned problems, and it is possible to grasp and predict the increase / decrease tendency of the flow rate with a very simple method, and also to accurately reflect the basin characteristics such as the downflow time of each river. It is intended to provide a flow rate prediction method that can be predicted with high accuracy by a model.
[0013]
[Means for Solving the Problems]
  Claim1The described invention uses a prediction model constructed by a computer, and in a flow prediction method for predicting the flow of a dam or a river while considering the flow down time, which is the time until rainfall in a plurality of upstream areas is reflected in the flow. ,In one flood case, the correlation coefficient of both patterns is calculated while moving the time change pattern of rainfall in each upstream area along the time axis in the direction of the time change pattern of the subsequent actual flow rate value. The time corresponding to the amount of movement of the time change pattern of rainfall is defined as the flow time of each upstream area, and the flow time of each upstream area is determined for multiple flooding cases, and the average value is calculated for the average time of flow of each upstream area. And each previous time by the time based on the average flow time in the upstream area.Average rainfall in each basin and eachUpThe basin average rainfall is calculated using the basin area, and the basin average rainfall is used to predict the flow rate.
[0014]
  Claim2The described invention uses a prediction model constructed by a computer, and in a flow prediction method for predicting the flow of a dam or a river from a cumulative rainfall for a predetermined period in an upstream region, a temporal change pattern of the cumulative rainfall for a plurality of types of cumulative time Then, all the correlation coefficients of both patterns are calculated while moving each along the time axis in the direction of the time change pattern of the actual flow rate value after that, and the accumulation of the time change pattern of the accumulated rainfall when the correlation coefficient becomes maximum Time is searched and the accumulated rainfall of the accumulated time is used for the flow prediction.
[0015]
DETAILED DESCRIPTION OF THE INVENTION
  Below, according to the figureReference form 1Will be explained.
  In general, even if there is rainfall when the ground in the upstream area is dry, it will not flow into the dam because it will be absorbed by the ground, and even if the ground is wet, such as after a long rain, Dam inflow increases. Thus, the relationship between the amount of dam inflow and the cumulative amount of rainfall due to long-term (several hours to several days) rainfall is more important than the relationship with instantaneous rainfall.
  ThereforeReference form 1Then, paying attention to the relationship between accumulated rainfall and the flow rate of dams, etc., we grasped the trend of increase and decrease of flow rate and used it for prediction.
[0016]
  The reference form 1 is mainly composed of two steps, and consists of step 1 for calculating the optimum accumulation time of the accumulated rainfall and step 2 for displaying the accumulated rainfall together with the inflow actual value.
  In step 1 for calculating the accumulated time, how many hours of rainfall should be accumulated from the past several hours to several months or similar past water discharge cases, for example, will be described later.2First of all, the invention is calculated. Next, the flow-down time is calculated for the accumulated rainfall determined according to the calculated accumulated time. Here, there are various methods for calculating the flow-down time.Reference form 3Good results can be obtained by using the present invention. This calculation of the flow time is not an essential requirement, but it is considered as the time when the dam inflow is shifted along the time axis when the accumulated rainfall and the dam inflow are displayed in superposition taking into account the flow time. .
[0017]
Next, the process proceeds to step 2 for displaying. Here, at least the cumulative rainfall during the same period is displayed in a superimposed manner together with the actual flow rate values for the past predetermined period with the current time as a reference. Moreover, the rainfall amount for every predetermined period from the past to the present, the rainfall amount forecast value from the present to several hours ahead, and the future accumulated rain amount forecast value are displayed on the screen as needed.
Even if only the accumulated rainfall for a given period up to the present time is displayed, it is possible to grasp the trend of increase or decrease of inflow about several hours ahead from the correlation between the accumulated rainfall and the actual inflow. By displaying the forecast value and the future cumulative rainfall forecast value calculated from the forecast rainfall value, it is possible to further grasp the trend of increase and decrease in the future inflow.
[0018]
  FIG. 1 shows a display example of a screen according to the reference mode 1.
  As shown in the figure, on the display screen, the actual flow rate of the past predetermined period up to the present time is displayed in a graph with a thick solid line, and the accumulated rainfall in the same period is displayed with a thin broken line. Here, on the vertical axis of the graph, actual values are normalized with an appropriate scale (the same applies hereinafter).
  Usually, depending on the water discharge case, the accumulated time with a high correlation with the inflow amount is different. So first, as mentioned above, the claims2The accumulated time is calculated by the procedure of the invention and the calculation result is displayed. In the example of FIG. 1, the accumulated time is 10 hours. In addition, the horizontal axis of FIG. 1 indicates the passage of time for the past 16 days.
  The specific procedure for calculating and displaying the accumulated time is as follows.
[0019]
(1) Calculation of accumulated time
  Accumulated time is calculated from the rainfall and flow rate (both are actual values) for the past several hours. Here, the actual rainfall value is claimed.1Is the average basin rainfall (basin average rainfall considering the downflow time) calculated by the invention of the present invention, and the accumulated time is claimed.2Determined by the invention. The calculation of the accumulated time is performed every fixed period such as every hour or every day.
[0020]
(2) Display
As shown in FIG. 1, the inflow amount actual value and the accumulated rainfall are overlapped and displayed on the screen. If the flow-down time is calculated as 1 hour, the dam inflow pattern may be moved to the right along the time axis by 1 hour relative to the accumulated rainfall pattern.
[0021]
  next,Reference form 2Will be explained.
  Mentioned aboveReference form 1Is effective for grasping the increase / decrease tendency of the flow rate, but is not suitable for predicting the flow rate itself. ThereforeReference form 2Then, using the fact that the correlation between the change in accumulated rainfall and the change in flow rate is high, the flow rate itself was predicted by a simple method.Reference form 2Indicates the basis for prediction for an unskilled operator, and uses the following simple formula.
[0022]
In general, in dams and rivers, if the amount of change in accumulated rainfall c is Δc, there is a relationship of change in flow rate = A · Δc + B with the change in flow rate. Paying attention to this relationship, the flow rate prediction value d is expressed by the following equation.
d = current flow rate + flow rate change = current flow rate + A · Δc + B
In addition, A and B in the formula are constants, and are calculated using the least squares method or the like based on past flooding cases at the time of construction of the prediction model, or the values calculated from the latest actual values are used for each prediction. May be.
Therefore, the predicted flow rate is obtained from the current flow rate and the change in flow rate using the above formula.
[0023]
  2 and 3 areReference form 2FIG. 2 is a display screen showing the actual inflow amount and cumulative rainfall corresponding to FIG. 1, and FIG.Reference form 2It is a display screen of the predicted value and the actual value of the inflow amount.
  As can be seen from these comparisons, in the display example of FIG. 2, the first inflow peak value and the cumulative rainfall peak value are substantially the same, but the second and third inflow peak values and the cumulative rainfall peak value are the same. It does not match the value. These second and third inflow peak values should be consistent with the cumulative rainfall peak value, in other words,Reference form 1Then, when the inflow is predicted based on the accumulated rainfall, a prediction error occurs. ThereforeReference form 2Then, the flow rate prediction with no error is performed by using the accumulated rainfall (cumulative rainfall change).
[0024]
  The flow rate prediction formula is d = current flow rate + flow rate change = current flow rate + A · Δc + B, but here, the flow rate change amount is treated as a flow rate change predicted value.
  Also,Reference form 2However, if the flow time is 1 hour, for example, if the flow time is 1 hour, data that is one hour older is used as the cumulative rainfall change Δc.
  FIG.Reference form 2It shows the predicted value and the actual value of the inflow volume. According to this figure, the peak values of the predicted value and the actual value are almost the same.
[0025]
  next,Reference form 3Will be explained.
  In general, the parameters of many prediction models require the flow time as the time until rainfall in the upper reaches is reflected in the flow rate of the dam, etc. Therefore, river flow time analysis is indispensable for building a prediction model . However, the method for obtaining the difference between the peak time of rainfall and the peak time of the flow rate is difficult to automate, and since only the peak time is focused, the calculation accuracy of the flow time is not high. Therefore, in the present invention, the flow time is automatically calculated from the correlation coefficient between the rainfall pattern and the flow rate pattern.
[0026]
  Reference form 3The principle of this is shown in FIG. As shown in the figure, the flow rate changes behind the rainfall in the upstream region. However, if this time delay is excluded, the shape of the flow rate pattern is not extremely different from the shape of the rainfall pattern.
  Therefore, as shown in FIG. 4, the rainfall pattern is moved little by little in the direction of the arrow, that is, in the direction of delaying the time, until the point where it best matches the flow rate pattern, and the amount of movement (movement time) in the time axis direction is obtained to flow down. Time.
[0027]
At this time, a correlation coefficient is used as an index representing the degree of coincidence between the rainfall pattern and the flow pattern. The correlation coefficient is an index of linearity that changes in the range of −1 to 1, and is 1 if the rainfall pattern and the flow pattern at the time of movement are similar, and 0 if they do not match at all. An appropriate flow-down time can be calculated from the amount of movement in the time axis direction of the rainfall pattern when the correlation coefficient shows the maximum value within the above range. These processes are automatically executed by the computer.
[0028]
FIG. 5 shows a correlation coefficient between the movement time of the rainfall pattern and the flow rate pattern for calculating the flow time of one river in a certain flood case.
In this example, since the correlation coefficient is approximately maximum when the moving time is approximately 4 hours, the flow-down time is calculated to be 4 hours. In other words, in this river, it is considered that the flow time from the rainfall in the upstream region to the increase in the dam inflow amount is 4 hours, so this time may be used for the flow rate prediction.
[0029]
  Then the claim1An embodiment of the invention will be described. In this embodiment, assuming that three rivers having different flow times are flowing into the dam, the basin average rainfall of these three rivers is calculated. When there are multiple rivers flowing into the dam, the water discharge characteristics (characteristics due to the terrain, geology, etc.) of each river are different. For example, in some rivers, the amount of water entering the dam increases immediately after rainfall, while in some other rivers it takes a considerable amount of time to increase the amount of inflow to the dam. These time differences can also be regarded as flow time differences.
[0030]
  In the embodiment of the present invention, first, as described above,Reference form 3To calculate the flow time of each river. At this time, since the flow time changes for each water discharge case, the average value of the flow time of many cases is taken for each river. Table 1 shows the cases of flooding and the flow time for each river. The average flow time is 4.2 hours for river 1, 5.6 hours for river 2, and 3.4 hours for river 3. .
[0031]
[Table 1]

[0032]
  Here, the basin average rainfall of each of the rivers 1 to 3 is calculated according to the obtained downflow time. In the present embodiment, when the rainfall information is data in units of one hour, if the decimal part is ignored, the river 1 is about 1 hour longer than the river 3 and the river 2 is about 2 hours longer than the river 3 The average rainfall of the basin is calculated by the following formula.
  Average basin rainfall = (River 1 basin area x River 1 basin rainfall + River 2 basin area x River 2 basin rainfall 2 hours + River 3 basin area x River 3 basin Current rainfall) / Basin area of all rivers
  That is, the basin average rainfall is obtained by a weighted average with the basin area as a weight. By using the average basin rainfall thus obtained, it is possible to predict the amount of water that flows from the rivers 1 to 3 to the dam almost simultaneously, and it is possible to predict the flow without time delay or dredging.
  The basin average rainfall calculated in this way is, for example,Reference Form 1 and Reference Form 2This method can be applied to the case where the accumulated rainfall and the accumulated rainfall change are obtained.
[0033]
  Next, the claim2An embodiment of the invention will be described.
  The present invention is for calculating the accumulated time and thus the accumulated rainfall having a high correlation with the dam flow rate. If the accumulated time is too long or too short, the correlation with the flow rate is low and cannot be used for prediction of the flow rate. In the present invention, an appropriate accumulated time is calculated by using a correlation coefficient between accumulated rainfall and flow rate.
[0034]
In the following, ~ time cumulative rainfall is not the cumulative rainfall from the start of rainfall, but the cumulative rainfall every hour from n hours before m hours before the river basin. For example, the 1-hour cumulative rainfall is the cumulative rainfall from n hours ago to (n-1) hours ago, and the 2-hour cumulative rainfall is the cumulative rainfall from n hours ago to (n-2) hours ago. Therefore, the 2-hour cumulative rainfall based on n hours ago includes the 1-hour cumulative rainfall.
[0035]
FIGS. 6A and 6B are diagrams showing temporal changes in accumulated rainfall and flow for calculating the accumulated rainfall time. In general, the relationship between accumulated rainfall and flow varies depending on the rainfall conditions. That is, when the ground is dry, the flow rate is highly correlated with the long-term (long accumulation time) accumulated rainfall, and when the ground is wet, the flow rate is short-term (short accumulation time) accumulated rainfall. The correlation tends to increase.
It should be noted that the 1-hour cumulative rainfall, the 2-hour cumulative rainfall, and the flow rate in FIGS. 6A and 6B are actual values, and are the past one hour (FIG. 6A) or two hours (FIG. 6). It shows that the change in the flow rate appears behind the change in the cumulative rainfall (b)).
[0036]
Calculation of the accumulated time in the present embodiment is performed according to the following procedure.
(1) The correlation coefficient of both patterns is obtained while moving the hourly cumulative rainfall pattern shown in FIG. 6A along the time axis in the direction of the flow rate pattern for one hour. That is, the procedure is the same as in FIG. 4, and the correlation coefficient is also obtained as an index of linearity that changes in the range of −1 to 1. In this case, the movement time of the 1-hour cumulative rainfall pattern corresponds to the above-described flow-down time.
[0037]
(2) Similarly, the 2-hour cumulative rainfall in FIG. 6B, the 3-hour cumulative rainfall not shown, the 4-hour cumulative rainfall,.kEach pattern of the time accumulation rainfall is moved along the time axis by one hour, and a correlation coefficient with the flow rate pattern is obtained. How many types of cumulative rainfall patterns are used per cumulative time (as mentioned abovekThe number of hours is determined as appropriate based on experience.
(3) Next, a maximum value is selected from the obtained correlation coefficients, and an accumulated time corresponding to the maximum value is searched. In other words, the “~ time” of the time-accumulated rainfall with the maximum correlation coefficient becomes the accumulated time as it is.
[0038]
For example, when the correlation coefficient when moving 1 hour to 12 hours of accumulated rainfall in a river basin by 1 hour is shown in Table 2, the correlation coefficient when the movement time is 1 hour in 10 hours of accumulated rainfall 0.934 "is the largest among all correlation coefficients. Therefore, for this river, the accumulated time for calculating the accumulated rainfall is 10 hours. Thereafter, the accumulated rainfall for 10 hours is used, and the flow rate is predicted in consideration of the flow time.
[0039]
[Table 2]

[0040]
  Here, FIG.andClaim2In order to explain the effectiveness of this invention, it is the figure which showed the average rainfall which carried out the simple average of the measured value of the several rain gauge installed in the dam upstream area, and the dam inflow actual value overlaid. The actual amount of inflow and the amount of rainfall as the basis of the display in FIG. 7 are the same as in the example in FIG.
  Since rain falls, stops, and varies widely, simply averaging the measured values of the rain gauge does not eliminate the variation. As can be seen from FIG. 7, the average rainfall may decrease even when the inflow is increasing, and it is difficult to grasp the increase / decrease trend of the inflow from the average rainfall.
[0041]
  In contrast,Reference form 3Automatically calculates the flow time for each river basin, and claims1According to the invention, the basin average rainfall is calculated from a weighted average weighted by the basin area, and further,2If converted to cumulative rainfall over an appropriate cumulative time according to the invention, the correlation between the change in cumulative rainfall and the increase / decrease trend of the inflow becomes clear, and the increase / decrease trend of the inflow can be easily grasped from the cumulative rainfall Can be used for flow prediction.
[0042]
  Finally, the claim2The prediction result when the embodiment of the present invention is applied to the inflow amount prediction model will be examined. Here, a neural network was used as an inflow prediction model. When applying a neural network to inflow prediction, rainfall information for a plurality of hours is required. In addition, as the method for determining the input time for how many hours of rainfall data to use, the operator decides empirically (1993 IEEJ Power and Energy Division Conference No. 185, Ichiyanagi et al.) There are a method of determining by the flow-down time (Japanese Patent Application No. 10-187108, etc.) and a method of determining by the accumulated time of the accumulated rainfall having a high correlation with the flow rate (the invention of claim 3).
[0043]
Here, two examination cases 1 and 2 are examined.
In Study Example 1, the prediction model shown in FIG. 8 is used, and the basin average rainfall difference from the current time to 6 hours before is input to the input layer I every hour. Note that II is an intermediate layer, and III is an output layer from which an inflow amount change after one hour from the current time (corresponding to A · Δc + B, which is the aforementioned change in flow rate) is output as a predicted value. The period up to 6 hours ago is based on the flow time of the target basin and the experience value of the operator.
[0044]
  In Study Example 2, the prediction model shown in FIG. 9 is used, and the basin average rainfall difference from the current time to 20 hours before every other hour is input. A value of 20 hours is claimed2It is the accumulated time calculated by the invention. Although the accumulated time varies depending on the flooding case, the longest time among the past cases was selected here. In the average case, the value is about 10 hours as shown in Table 2 above.
  The input / output factors of the two prediction models shown in FIGS. 8 and 9 are summarized as shown in Table 3. Based on these input / output factors, a plurality of past cases are sufficiently learned, and two flooding cases are predicted. I let you.
[0045]
[Table 3]

[0046]
In Table 3, the basin average rainfall when obtaining the basin average rainfall difference is a value that is not adjusted by the downflow time when there are a plurality of rivers. In other words, it is a value obtained by adding all the rainfalls of each river at the same time multiplied by the basin area of each river and dividing the result by the total basin area.
Further, for example, the basin average rainfall difference up to one hour ago is a value obtained by subtracting the basin average rainfall one hour ago from the current basin average rainfall.
[0047]
FIG. 10 (a) shows the predicted inflow amount and actual value for the flood case 1 according to the prediction model of FIG. 8, and FIG. 10 (b) shows the predicted inflow amount and actual value for the flood case 2 as well. Yes. Similarly, FIGS. 11 (a) and 11 (b) show the inflow amount predicted values and actual values for the water discharge case 1 and the water discharge case 2 based on the prediction model of FIG.
The specific method for predicting the inflow is to add the change in inflow after 1 hour (predicted value) obtained by inputting the basin average rainfall difference to the prediction model to the inflow after 1 hour. A method of predicting the quantity is taken, and this is repeated along the time axis direction of FIGS.
[0048]
Tables 4 and 5 show the results of evaluating the prediction error for Study Examples 1 and 2. Table 4 relates to flood case 1 and Table 5 relates to flood case 2.
As an evaluation index of the prediction error, an absolute value average error [%] and a correlation coefficient (correlation coefficient between the prediction value and the actual value) were used. Here, the absolute value average error should be small, and the correlation coefficient should be close to 1.0.
[0049]
[Table 4]

[0050]
[Table 5]

[0051]
  As is clear from Tables 4 and 5, the basin average rainfall difference is claimed as2In Study Example 2 using difference data corresponding to the accumulated time in the invention of FIG. 4, the absolute value average error is smaller than in Study Example 1 in both Tables 4 and 5, and the correlation coefficient is close to 1.0. Yes. In other words, it was confirmed that better results were obtained when the accumulated time was selected as the rain input time.
[0053]
【The invention's effect】
  As stated above, the claims1, 2The described invention is a technique that can be used when constructing a prediction model. In building a prediction model, it is necessary to understand the characteristics of the basin.The
[0054]
  Claim1The described invention relates to a method for calculating a basin average rainfall. There are known methods for predicting the flow rate for each river when multiple rivers merge, but the prediction of individual rivers is not highly accurate due to variations in individual prediction values, and the flow prediction value of the target dam is not high. There are limits to improving accuracy. On the other hand, in the method of predicting a plurality of rivers as a whole, variation in data is reduced, but accuracy is not high when there is a difference in characteristics between rivers.
  According to the present invention, it is possible to calculate the basin average rainfall in consideration of the downflow time for each river, and it is possible to predict with high accuracy by using this basin average rainfall.
[0055]
  Claim2According to the invention, the accumulated time for calculating the accumulated rainfall is calculated. In the reference forms 1 and 2 described above, accumulated rainfall having a high correlation with the flow rate is required as data. This claim2According to the invention, since the accumulated rainfall with high correlation can be obtained, for example, when applied to the reference forms 1 and 2, it is possible to perform the prediction with relatively high accuracy. Also, other forecast models often use several hours of rainfall as input information.2According to this invention, it becomes possible to determine the input range of the optimal rainfall (how many hours of rainfall are used for prediction).
  As detailed above, the claims1, 2This invention is a general-purpose and easy-to-automate method that can be applied to rivers of any shape.
[Brief description of the drawings]
FIG. 1 is a diagram showing a display example of a screen according to a reference form 1;
FIG. 2 is a diagram showing a display example of a screen according to a reference form 1;
FIG. 3 is a diagram illustrating a display example of a screen according to a reference mode 2;
[Fig. 4]Reference form 3It is explanatory drawing of the principle of.
[Figure 5]Reference form 3It is a figure which shows the relationship between the flow down time and correlation coefficient in.
FIG. 6 Claim2In embodiment of this invention, it is a figure which shows the time change of the accumulation rainfall and flow volume for calculating accumulation time.
FIG. 7 is a diagram in which average rainfall obtained by simply averaging measured values of a plurality of rain gauges and a dam inflow actual value are superimposed.
FIG. 8 Claim2It is a conceptual diagram of the prediction model with which embodiment of this invention is applied.
FIG. 9 Claim2It is a conceptual diagram of the prediction model with which embodiment of this invention is applied.
FIG. 10 is a diagram showing a prediction result for each water discharge case.
FIG. 11 is a diagram showing a prediction result for each water discharge case.
FIG. 12 is an explanatory diagram of a conventional dam flow rate prediction method.
FIG. 13 is an explanatory diagram of a conventional dam flow rate prediction method.

Claims (2)

In a flow prediction method that uses a prediction model built by a computer and predicts the flow of a dam or river while taking into account the flow down time, which is the time until the rainfall in multiple upstream areas is reflected in the flow,
In one flood case, the correlation coefficient of both patterns is calculated while moving the time change pattern of rainfall in each upstream area along the time axis in the direction of the time change pattern of the subsequent actual flow rate value. The time corresponding to the amount of movement of the time change pattern of rainfall is defined as the flow time of each upstream area, and the flow time of each upstream area is determined for multiple flooding cases, and the average value is calculated for the average time of flow of each upstream area. age,
That calculates the basin average rainfall, to predict the flow rate by using the basin average rainfall by using the average rainfall and the area of each of the upper basin time duration only previously each of the upper basin based on falling average time of the upstream region A method for predicting the flow rate of a characteristic dam or river. In a flow prediction method that uses a prediction model built by a computer to predict the flow of a dam or river from the accumulated rainfall over a predetermined period in the upstream area,
Calculate the correlation coefficient of both patterns while moving the time change pattern of accumulated rainfall for multiple types of accumulated time along the time axis in the direction of the time change pattern of the actual flow rate value thereafter, and the correlation coefficient is the maximum. A method for predicting a flow rate in a dam or a river, wherein a cumulative time corresponding to a cumulative amount of rainfall at the time of the rain is searched for and the accumulated rainfall during the cumulative time is used for the flow rate prediction.

Images