JP2003293344A - River water level predicting apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a river water level predicting apparatus which predicts the water level of a river with accuracy. <P>SOLUTION: The river water level predicting apparatus 11 comprises a water level measuring section 5 for measuring the water level of the river 4 and storing therein a measured value, rainfall measuring sections 7, 8 for measuring rainfall quantities of a river 2 and the river 4 and storing therein the quantities, and a water level predicting model 20 for predicting the water level of the river, based on the measured values. The water level predicting model 20 has an autoregressive portion and an FIR model portion, and the model 20 is identified by a model identifying section 21. The model identifying section separately calculates a parameter for the autoregressive section and a parameter for the FIR model section, based on the measured values. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a water level predicting apparatus for predicting a river water level, and more particularly to a river water level predicting apparatus capable of accurately predicting a river water level.

[0002]

2. Description of the Related Art Generally, a drainage pump station is installed at a point where a diversion channel joins a mainstream river. In addition, at the drainage pump station,
There is a drainage pump station and a diversion channel that joins the mainstream river. Normally, the water in the diversion channel merges into the main river under natural flow, but if the water level on the main river side increases due to heavy rain,
There is a risk of backflow from the mainstream river to the side of the diversion channel, resulting in flooding of the river in the diversion channel. Therefore, during heavy rain,
After closing the sluice provided near the confluence of the diversion channel and the main river, preventing backflow from the main river to the diversion channel,
In addition, the drainage pump discharges the water from the diversion channel to the main river. At this time, the timing of starting the drainage pumps and the increase / decrease of the number of pumps to start are important, but in order to grasp the appropriate timing, the water level of the river is accurately predicted, and the drainage pumps are predicted based on the river water level prediction results. It is desirable to provide driving assistance.

In the river water level prediction for the purpose of supporting the operation of the drainage pump and controlling the drainage pump, it is more important to predict the water level in a short period within one hour than to predict the water level in the long term. In particular, it is important to achieve highly accurate water level prediction in situations where water level changes occur rapidly. The reason for this importance is that it generally takes about 10 minutes to operate the water gate (fully open ⇔ fully closed) and to activate the drainage pump, and therefore a series of operations requires several tens of minutes. In the time of about several tens of minutes, the required requirements (the number of operating drainage pumps, the gate opening, etc.) change according to the state of water level change. Therefore, deciding the proper timing for the operation of these drainage pumps and sluices is the most important point in the operation of drainage pump stations, and in order to realize this, highly accurate water level prediction within one hour is a key point. Become.

In order to predict the river water level, it is necessary to construct a model for the prediction. As one of the means for constructing this prediction model, the time series data such as the water level and the rainfall amount actually collected are used. There are ways to build a model based on. This method is optimal when making a prediction with high accuracy in a short time.

Generally, as a method of modeling based on time series data, multivariate analysis (multiple regression analysis etc.), neural network and system identification method are well known and widely used in industrial fields. Although these techniques are interrelated and not independent,
Of these, the system identification method is extremely effective as a model construction method for the purpose of (temporal) prediction and control.
The greatest feature of the system identification method is
This is the point that model structure and parameter identification algorithms are being considered with a strong awareness of the temporal (or spatial) dynamics of the process.

Further, various analyzes (model construction based on this) are possible from the viewpoint of a dynamics system. For example, taking into account the dynamic characteristics (usually expressed as a filter) of the disturbance noise that gets into the process,
Through process characteristic analysis in the frequency domain and analysis of poles and zeros, process characteristics can be grasped and the process model can be reduced in dimension. By performing analysis (and model building based on it) as a dynamics system, it is possible to significantly improve the prediction accuracy.

In particular, like river water level fluctuations, (1) various natural phenomena (and artificial operations) interfere with each other, and water level fluctuations occur as a result of such complex factors, and (2)
Many of the factors that cause fluctuations in river water level are natural phenomena and cannot be manipulated artificially, that is, when the open system water level prediction with the characteristics of (1) and (2) above is performed. , It is very important to construct a water level prediction model by considering the disturbance characteristics of the process in an appropriate manner in order to improve the prediction accuracy.

The reason why it is important to properly consider the process disturbance characteristics in predicting the water level in an open system is as follows. In the open system, many of the factors that are input to the prediction model cannot be manipulated artificially, so the number and quality of time series data obtained when building the model are limited. When building a predictive model from these limited data,
It is known that it is better to adopt some mutually independent factors that have a large influence on prediction as inputs, than adopt all possible factors as inputs to the prediction model. However, when constructing a prediction model using the factors that have a great influence on these forecasts, the remaining factors that are not taken into consideration affect the water level forecast as disturbances. For this reason, it is very important to properly consider the disturbance characteristics in constructing the prediction model in order to improve the prediction accuracy.

[0009] In river water level prediction, rainfall data from rainfall gauge data installed at a plurality of locations and discharge rate at an upstream pump station have an effect on water level fluctuations, but these data have a strong correlation with each other. (In the field of multivariate analysis, it is called "multicollinearity").

It is natural that the rainfall data have a strong correlation with each other as a natural phenomenon, and the reason that the rainfall data and the pump discharge amount have a correlation is that if the rainfall amount is large, the discharge amount also increases accordingly. Therefore, it is difficult to distinguish from the time-series data how much rainfall data or pump discharge data affected the water level fluctuation. Therefore, from these rainfall and discharge data, we will select some of the possible major factors and use the selected major factors as inputs to build a predictive model. , It is also necessary from the viewpoint of numerical stability when identifying the parameters of the prediction model).

[0011] Here, for example, when rainfall is input, the discharge rate of the pump station on the upstream side fluctuates for some reason so as not to correlate with rainfall, and affects the river water level to be predicted. In this case, this discharge amount acts on the river water level as a disturbance. That is, since the accuracy of prediction deteriorates unless the effect of disturbance is properly incorporated into the prediction model, it is necessary to construct the prediction model by appropriately considering this disturbance characteristic.

By the way, in the system identification method, there are various methods for describing the disturbance characteristic in the model to be applied. The method for describing such disturbance characteristics is roughly divided into ARX (Autogressive mo
del with exogenous input
An error model represented by the ARX model and FI
R (Finite Impulse Responses)
e) There is an output error model represented by a model (strictly speaking, the FIR model is both an output error model and an expression error model, but it is better to consider it as an output error model from the viewpoint of prediction). The difference in the description method of the disturbance characteristic appears as a difference in the prediction method, and the difference in the prediction method affects the prediction accuracy.

In the ARX model (formula error model), the prediction of the past water level change which is an indirect factor of the river water level change by the autoregressive part, and the external factors such as rainfall and pump discharge which are the direct factors of the river water level change The river water level is predicted by the input side and the prediction side. On the other hand, FIR
In the model (output error model), the river water level is predicted only by the prediction by external input which is a direct factor. Here, the past river water level is taken as an indirect factor because the past water level changes as a result of changes in rainfall and upstream pump discharge, which are direct factors, and the past water level changes the future water level. Because it is. That is, the past water level is not the "real factor" for changing the future water level.

The prediction methods based on the above two prediction models have the following features.

(1) When the prediction model is the ARX model,
For example, even when there is a water level fluctuation due to a disturbance such as the above-described upstream pump discharge amount, the prediction model incorporates the actually measured water level in the prediction in the autoregressive part. Therefore, the prediction model can deal with the disturbance to some extent robustly and can correct the prediction. On the other hand, in the prediction model, since the identification can be performed only by the limited data as described above, the direct factors such as rainfall cannot be identified correctly. As a result, the parameters of the prediction model are often identified such that the autoregressive part, which is an indirect factor, has a large influence on the prediction. Furthermore, even if external inputs such as rainfall and pump discharge, which are direct factors, change rapidly, the prediction model considers past river water levels, which are indirect factors, and then predicts this rapid change in external inputs. Reflect on. That is, the prediction is behind, and the prediction accuracy deteriorates especially when the water level changes abruptly.

(2) When the prediction model is the FIR model,
The predictive model has only external input terms that are direct factors. As a result, the number and quality of data that can be used for parameter identification are limited, so that the indirect factor does not influence the parameter identification result. Therefore, there is no problem that the prediction is behind when there is a sudden change in water level, and the prediction model can relatively easily determine which factor is the main cause when the prediction accuracy is poor. On the other hand, however, the prediction model cannot deal with the case where the water level fluctuates due to some factor that is not adopted as an input, and cannot correct the prediction when an unexpected water level change occurs.

[0017]

As described above, if an ARX model or an FIR model is simply applied to the prediction of an open system such as a river water level prediction,
(3) Deterioration of prediction accuracy (especially when the water level suddenly changes) (following prediction) (4) Deterioration of water level prediction accuracy due to the influence of factors not taken into consideration
It is difficult to solve both of them at the same time.

The present invention has been made in consideration of such a point, and it is possible to construct a prediction model capable of coping with both the above (3) and (4) and accurately predict the water level. The object is to provide a water level prediction device for a river that can be used.

[0019]

[Means for Solving the Problems] The present invention predicts the river water level based on the information from the river water level fluctuation factor measuring unit for measuring the river water level fluctuation factor and the river water level fluctuation factor measuring unit. Water level prediction that has a first prediction model part that predicts the water level based on the water level from the present to the present, and a second prediction model part that predicts the water level based on the information from the river water level fluctuation factor measurement unit from the past to the present A model and a model identification unit that identifies the water level prediction model are provided. The model identification unit measures at least one of the parameters of the first prediction model part and the parameters of the second prediction model part of the water level prediction model for measuring river water level fluctuation factors. This is a river water level prediction device characterized by being obtained based on information from the department.

According to the present invention, the water level prediction model has a noise part obtained by filtering the noise part in advance, the parameters of the first prediction model part are specified by designing the filter, and the model identification part is the second part. This is a river water level prediction device characterized by obtaining only parameters of a prediction model part.

In the present invention, a plurality of river variation factors are set,
The model identification unit obtains and obtains the parameters of the first prediction model part of the water level prediction model and the parameters of the second prediction model part corresponding to the river water level fluctuation factor based on the specific river water level fluctuation factor among the river fluctuation factors. A parameter of the second prediction model part corresponding to another river water level fluctuation factor is obtained based on the parameter of the first prediction model part and the parameter of the second prediction model part corresponding to the river water level fluctuation factor. This is a river water level prediction device.

According to the present invention, the model identification section independently obtains the parameters of the first prediction model part and the parameters of the second prediction model part of the water level prediction model, based on the information from the river water level fluctuation factor measurement section, respectively. A water level prediction device for a river, characterized by performing a preset conversion so as to match the parameters of the first prediction model part and the parameters of the second prediction model part as parameters of the water level prediction model.

According to the present invention, the noise part of the water level prediction model is
A water level predicting apparatus for a river characterized by being weighted in a desired frequency band.

The present invention is based on the parameters of the first prediction model part and the parameters of the second prediction model part corresponding to a specific river water level fluctuation factor, and the parameters of the second prediction model part corresponding to other river water level fluctuation factors. The water level predicting apparatus for a river is characterized by using a calculation formula adapted to the frequency characteristic of the first predictive model portion when calculating.

According to the present invention, when the parameters of the first prediction model part and the parameters of the second prediction model part are converted so as to be adapted as the parameters of the water level prediction model,
This is a river water level prediction device characterized by conversion using an observable canonical form.

According to the present invention, it is possible to accurately predict the water level by preventing the deterioration of the water level prediction accuracy when the water level changes abruptly and the deterioration of the water level prediction accuracy due to the influence of factors not taken into consideration.

[0027]

BEST MODE FOR CARRYING OUT THE INVENTION First Embodiment A first embodiment of a river water level prediction device according to the present invention will be described below with reference to FIGS. 1, 2 and 4.

In the first embodiment, when identifying the water level prediction model used in the river water level prediction device, the parameters of the autoregressive part of the water level prediction model are specified at the time of filter design, and only the parameters of the FIR model part are specified. It is obtained by the model fixing unit.

As shown in FIG. 2, the diversion channel 4 is the main river 2.
A drainage pump station 1 is installed at a point where the main stream river 2 that joins the diversion channel 4 is connected to the sea 10. In the vicinity where the watershed 4 and the main river 2 merge, a sluice 3 for preventing backflow from the mainstream river 2 to the watershed 4 and a sluice 3 for preventing backflow from the mainstream river 2 to the watershed 4 are provided. A pump (drainage pump) 9 for discharging the water in the waterway 4 to the mainstream river 2 is provided.

FIG. 1 is a diagram showing a first embodiment of a water level predicting apparatus for a river according to the present invention.

As shown in FIG. 1, a river water level predicting apparatus 11 according to the present invention is provided in a water diversion channel 4, and a water level measuring unit 5 for measuring the water level of the water diversion channel (river) 4 from the past to the present.
And the rainfall measuring units 7 and 8 provided in the main river 2 and the diversion channel 4, respectively, for measuring rainfall in the rivers 2 and 4 from the past to the present.

The sea 10 is the sea 1 from the past to the present.
A tide level measuring unit 6 for measuring a tide level of 0 is installed. Further, the pump 9 is provided with a discharge amount measuring unit 16 for measuring the discharge amount (pump operating condition) of the pump 9 from the past to the present, and the floodgate 3 has an opening / closing degree of the floodgate 3 from the past to the present ( An opening / closing degree measuring unit 15 for measuring the sluice operation status) is installed. The water level measuring unit 5, the tide level measuring unit 6, the rainfall amount measuring units 7 and 8, the discharge amount measuring unit 16, and the opening / closing degree measuring unit 15 are identified by the model identifying unit 21 via the input / output data storage unit 30. Water level prediction model 20
It is connected to the.

The model identification unit 21 includes an autoregressive parameter 22, a discharge amount parameter 23, and an opening / closing degree parameter 2.
4, the rainfall parameter 25, and the water level parameter 26
And a water level prediction model 20 having a tide level parameter 27
Is to be identified. That is, the model identification unit 21
Is the discharge amount from the discharge amount measuring unit 16 and the opening / closing degree measuring unit 15
The degree of each parameter and the coefficient of each parameter are calculated based on the degree of opening and closing from the water level, the water level from the water level measuring unit 5, the rainfall amount from each of the rainfall amount measuring units 7 and 8, and the tide level from the tide level measuring unit 6. The water level prediction model 20 that has been determined and confirmed is constructed.

Next, the operation of the embodiment having such a configuration will be described.

The tide level measuring unit 6 measures the tide level of the sea 10,
The rainfall measuring units 7 and 8 measure the rainfall in the two main river basins and the four watersheds. Further, the water level measuring unit 5 measures the water level of the diversion channel 4, the discharge amount measuring unit 16 measures the discharge amount of the pump 9, and the opening / closing degree measuring unit 15 measures the opening degree of the sluice 3.

Of these, the tide level measuring unit 6, the rainfall amount measuring unit 7,
The opening / closing degree measuring unit 15 and the discharge amount measuring unit 16 serve as a changing factor measuring unit for measuring the tide level, the rainfall amount, and the opening / closing degree, which are the factors of water level variation in the river.

The tide level measuring unit 6, the rainfall amount measuring units 7 and 8, the water level measuring unit 5, the discharge amount measuring unit 16 and the opening / closing degree measuring unit 15
Sends the measured tide level, rainfall amount, water level, discharge amount and opening / closing degree to the input / output data storage means 30 for storage. The input / output data storage means 30 stores the measured values from the past to the present in time series, and then sends them to the model identification unit 21 and the water level prediction model 20. The model identification unit 21 identifies the water level prediction model 20 based on the transmitted tide level, rainfall amount, water level, discharge rate, and opening / closing degree, and determines the order of each parameter of the water level prediction model and the coefficient of each parameter. , And establishes the fixed water level prediction model 20.

Next, by the determined water level prediction model 20, the tide level, the rainfall amount, the discharge amount and the opening / closing amount sent from the tide level measuring unit 6, the rainfall amount measuring units 7 and 8, the discharge amount measuring unit 16 and the opening / closing degree measuring unit 15. Based on the degree, the water level of the river 2 is predicted.

When the water level of the river 2 is predicted by the water level prediction model 20, the predicted water level is displayed on the display device 31 and sent to the driving support means 32.

Next, a method of identifying the above water level prediction model and determining the order of each parameter and the coefficient of each parameter of the water level prediction model will be described in detail with reference to FIG.

The water level prediction model 20 includes an ARX model shown in the following (1 expression).

[0042]

[Equation 1] Here, each parameter of the above (equation 1) shows the following.

U ₁ ^* (k): tide level (input) from the tide level measurement unit u ₂ ^* (k): rainfall amount (input) from the rainfall amount measurement unit u ₃ ^* (k): discharge amount from the discharge amount measurement unit Discharge rate (input) y ^* (k): River water level (output) e (k): Noise part A (z): Autoregressive parameter (autoregressive part) 22 B ₁ (z): Tide level parameter (tide level (input ) 27 B ₂ (z): Rainfall parameter (FIR model part for rainfall (input)) 25 B ₃ (z): Discharge amount parameter (FIR model part for discharge (input)) 23 z : Shift operator Among the above parameters, the auto-regression part 22 becomes the first prediction model part that predicts the water level based on the water level of the river from the past to the present, and the FIR model parts 23, 24, 25,
Reference numeral 27 is a second prediction model part for predicting the water level based on the information from the river water level fluctuation factor measurement units 6, 7, 8, 15, 16 from the past to the present.

First, the model identifying unit 21 uses the following (equation 2) in which only the tide level u ₁ (k) is considered as an input, and the B ₁ (z) (tide level parameter) 27 of the tide level u ₁ ^* (k) is calculated. A method for performing identification will be described.

From the above (1), the model identifying unit 21
Considering only tide level u ₁ ^* (k) as input, the following (2
Formula).

[0046]

[Equation 2] The model identification unit 21 uses the noise unit including the filter to
Convert the water level prediction model to the water level prediction model.
That is, the model identifying unit 21 transforms the above-described (2) so that the noise part of the ARX model has a noise model that has passed through the filter L (z) in advance, and obtains the following (3).

[0047]

[Equation 3] That is, in the above equations (4) and (5), the difference is calculated in advance from the input data and the output data.
This means that an offset that cannot be dealt with by the conventional system identification method can be effectively dealt with, and the identification accuracy can be improved. Generally, the data obtained by taking the difference as in the above equations (4) and (5) has an effect of removing an offset, and a high pass filter having a gain close to 0 in a low frequency band also has a similar function. . That is, it is possible to remove the offset by filtering the data in advance with the high-pass filter. Further, since this high-pass filter is not an identification target but a design target, the high-pass filter can be designed according to the purpose.

The model identification unit 21 substitutes the autoregressive parameter A (z) = 1 in the above (Equation 3) to obtain the following (Equation 6). That model identification section 21 performs the tide level parameter (FIR model _portion) B 1 _(z) 27 only for the identification, to determine the order and coefficients of tidal parameters (FIR model _portion) B 1 _(z) 27.

[0049] _{Here, B 1 (z) = 1} b 0 + 1 b 1 z
And ^{_{-1 + 1 b 2 z -2 +}} ··· + 1 b m z -m.

[0050]

[Equation 4] As described above, the model identification unit 21 is
Based on the water level of the
Using the multiplication method, the tide level parameter B in (Equation 7) above ₁(Z)
27 are identified and the tide level parameter (FIR part) B₁
(Z) Determine the order and coefficient of 27. Where the model
The parameter identification unit 21 identifies parameters using the least squares method.
However, identification may be performed using other methods.
(Hereinafter, the same shall apply to identification of parameters).

The model identification unit 21 uses B ₂ (z) (rainfall amount parameter) of the other input amount of rainfall u ₂ ^* (k).
For 25, the order and coefficient of B ₂ (z) (rainfall amount parameter) 25 are determined similarly to B ₁ (z), and B ₃ (z) (discharge amount parameter) of discharge amount u ₃ ^* (k) is determined. Regarding 23 as well, similarly to B ₁ (z), the order and coefficient of B ₃ (z) (ejection amount parameter) 23 are determined.

In the above equation (Equation 7), L (z) is a setting
Since it is a total parameter, the model identification unit 21
7) FIR model part (B₁(Z)) Only for 28
Identify. However, the model identification unit 21
7) FIR model part (B ₁(Z)) Only for 28
Auto-regressive part in (Equation 7)
Since (1-L (z)) 22 is included, the predicted water
In the process of finding the rank, the autoregressive part (1-L
(Z)) 22 will be considered.

As described above, according to the present embodiment, the autoregressive part (1-
By considering L (z) 22, the past of the water level y (k) which is an output can be considered, and a robust prediction result can be expected even for an unknown disturbance. On the other hand, the model identifying unit 21 uses the FIR model portion (B ₁ (z)) 2
Since only 8 are identified, the autoregressive part 2
It is possible to reduce the effect of 2 and avoid the phenomenon that the prediction is behind the sudden water level change.

As L (z) (design parameter), the above-described differential filter (high-pass filter) is designed in order to improve the prediction accuracy of the external input that changes very rapidly. When it is desired to regard the change of as a disturbance, a weighted bandpass filter may be designed so as to have a constant bandwidth (generally on the high frequency side). That is, the designer can select the frequency band for which the prediction accuracy is desired to be increased in L (z) (design parameter).

For example, when it is necessary to predict the water level from 10 minutes to 30 minutes ahead, the frequency band from 0.007 rad / s to 0.01 rad / s corresponding to the prediction time is selected.

The model identification unit 21 sets L (z) = 1 in (Equation 6) and sets the autoregressive portion (1-L (z)).
It is possible to construct the water level prediction model by only identifying the FIR model portion 28 without considering 22. In the case where the influence of the disturbance can be ignored, the method of identifying only the FIR model portion 28 can perform the identification while considering the influence of the input on the output, so that a sufficient number and sufficient quality of the input can be obtained. If obtained, it will be an effective means for identification against natural phenomena.

Next, a modification of the present invention will be described.

The present invention is intended for the water level prediction process of the river 4, but not limited to this, the rainwater inflow amount prediction process to the rainwater pumping station, the sewer trunk trunk flow rate and water level prediction process, and the water treatment process. It is also possible to target an open process whose target is a natural phenomenon such as.

Second Embodiment Next, the identification of the second embodiment of the present invention will be described with reference to FIGS. 1 and 5.

FIG. 1 shows a second embodiment of the river water level prediction apparatus according to the present invention.

In the second embodiment, in identifying the water level prediction model, first, the parameters of the auto-regression part and the parameters of the FIR model part are obtained based on the river water level fluctuation factors in fine weather, and then in rainy weather and heavy rain. The parameters of other FIR model parts are sequentially obtained on the basis of the other river water level fluctuation factors in.

The second embodiment is substantially the same as the first embodiment except for the method of determining the order of each parameter and the coefficient of each parameter of the water level prediction model. In the second embodiment, detailed description of the same parts as those in the first embodiment will be omitted.

The model identifying unit 21 determines the parameter related to either the tide level parameter 27 or the rainfall amount parameter 25 based on the water level from the water level measuring unit 5 and the rainfall amount from the rainfall amount measuring units 7 and 8. Determining the order and parameter coefficients and the parameter order and parameter coefficients for the autoregressive parameter 22 (ARX
Model identification), one of the determined tide level parameter 27 or rainfall parameter 25, and the autoregressive parameter 2
2, the water level from the water level measurement unit 5, the tide level from the tide level measurement unit 6, and the rainfall amount from the rainfall amount measurement units 7 and 8, based on the tide level parameter 27 or the rainfall amount parameter 25 Determine the order and the coefficient of the parameter (identification of the FIR model part).

That is, the model identifying unit 21 constructs a water level prediction model determined by the following three (Step). Here, the model identification unit 21 uses the autoregressive parameter A (z) of Equation (1), the tide level parameter B ₁ (z), the rainfall amount parameter B ₂ (z), and the discharge amount parameter B.
The order and coefficient of ₃ (z) are determined, and a fixed water level prediction model is constructed.

(Step 1) As shown in FIG. 5, the model identifying section 21 uses the equation (1) based on the tide level from the tide level measuring section 6 and the water level from the water level measuring section 5 in fine weather. (8) is identified using the least squares method, and (8
The coefficient and order of the autoregressive parameter A (z) and the order and coefficient of the tide level parameter (FIR part) B ₁ (z) are determined.

[0066] _{Here, B 1 (z) = 1} b 0 + 1 b 1 z
^{_{-1 + 1 b 2 z -2 +}} ··· + 1 b m z -m, A (z)
^{_{= A 0 + a 1 z -1}} + a 2 z -2 + ··· + a m z -m
And

[0067]

[Equation 5] This is the identification at the time of fine weather, and by this, the order and coefficient of the autoregressive parameter A (z) and the tide level parameter B ₁
The order and coefficients of (z) are determined.

The model identification unit 21 uses the equation (8).
The autoregressive parameter A (z) and the tide level parameter (F
IR model part) B₁Both (z) are identified. Profit
Since most of the data that is collected is in clear weather,
The autoregressive parameter A (z) largely depends on the data in fine weather.
Since it is considered that the model identification unit 21
Is an autoregressive parameter A that is common to all input parts.
(Z) is the tide level parameter B ₁Identified with (z)
(Identification of ARX model).

(Step 2) Next, in rainy weather, the pump 9 is used.
Consider when the car is not driving (light rain).

The model identifying section 21 calculates the water level output y _c (k) in fine weather by the identified (Equation 8) based on the tide level from the tide level measuring section 6. Further, the model identifying unit 21 performs an operation of (y _r (k) = y (k) −y _c (k)) based on y (k) and the calculated y _c (k), The water level output y _r (k) when the pump 9 is not operating in rainy weather is calculated.

The model identifying section 21 also identifies the identified A
(Z) and the rainfall amount (u
₂ (k)), based on (u _2r (k) = (1 / A
(Z)) u ₂ (k)) is calculated to calculate u _2r (k).

Further, the model identification unit 21 calculates y _r (k) calculated above, u _2r (k), and the identified A (z).
Based on and, the following (formula 9) is identified using the least squares method, and the order and coefficient of the rainfall parameter B ₂ (z) of (formula 9) are determined.

[0073] _{Here, B 2 (z) = 2} b 0 + 2 b 1 z
And ^{_{-1 + 2 b 2 z -2 +}} ··· + 2 b n z -n.

[0074]

[Equation 6] It should be noted that the model identification unit 21 uses the FI
The R model is identified (identified only for B ₂ (z)).

In addition, when 1 / A (z) has a low-pass characteristic (equation (9)) (when A (z) has a high-pass characteristic), the model identifying unit 21 determines that A (z ) Y
_{Based on} the calculation result of _r (k) and the rainfall amount from the rainfall amount measuring units 7 and 8, the following (formula 10) is identified, and the order and coefficient of the rainfall parameter B ₂ (z) are determined. May be.

[0076]

[Equation 7] By the above identification, the identification signal has the effect of satisfying the continuous excitation condition due to the nature of the high-pass characteristic, and improvement in identification accuracy can be expected.

In the above equation (9), when the autoregressive parameter A (z) has both high-pass characteristics and low-pass characteristics, the portion having the high-pass characteristics is A _h (z) and the portion having the low-pass characteristics is It can be defined as A ₁ (z). When defined in this way, the autoregressive parameter A
(Z) can be expressed as A (z) = A _h (z) A _l (z).

When the autoregressive parameter A (z) has both high-pass characteristics and low-pass characteristics, the model identification unit 2
1 was obtained by substituting A (z) = A _h (z) A _l (z) into (Equation 9) and obtaining the following (Equation 11).
Formula) may be identified.

[0079]

[Equation 8] The model identifying unit 21 considers the continuous excitation condition in both the output water level y _r (k) and the input rainfall amount u ₂ (t) by identifying (Equation 11). You can

(Step 3) Next, when the pump 9 is operating in rainy weather, consider heavy rain, that is, (water protection).

The model identifying section 21 includes the rainfall measuring sections 7 and 8
Based on the amount of rainfall from
The output (y _rr (k)) at the time of flood control is calculated. Next, the model identifying unit 21 _calculates (y _rp (k) = y (k) −y) based on y (k) and the obtained y _rr (k).
_rr (k)) is calculated to calculate the output y _rp (k) when the pump 9 is not operating in rainy weather. Thereafter, the model identifying unit 21 determines u _3r (k) = (based on the autoregressive parameter A (z) identified in (Equation 9) and the discharge amount u ₃ (k) from the discharge amount measuring unit 16. 1 / A
(Z)) U ₃ (k) is calculated to calculate u _3r (k).

The model identification unit 21 calculates the calculated y
_{Based on rp} (k), u _3r (k), and the identified autoregressive parameter A (z), the following (Equation 12) is identified using the least square method, and the discharge amount parameter B ₃ (z ) Determines the order and coefficient.

[0083] _{Here, B 3 (z) = 3} b 0 + 3 b 1 z
And ^{_{-1 + 3 b 2 z -2 +}} ··· + 3 b l z -l.

[0084]

[Equation 9] Similarly to the above, the model identifying unit 21 determines the order and coefficient of the opening / closing degree parameter B ₄ (z).

[0085] _{Here, B 4 (z) = 4} b 0 + 4 b 1 z
And ^{_{-1 + 4 b 2 z -2 +}} ··· + 4 b k z -k.

The identification of (Equation 12) is the identification of the FIR model part (identification is made only for B ₃ (z)).

The model identification unit 21 is the same as (Step 2).
Consider the high-pass characteristics of autoregressive parameter A (z)
And then A (z) y_rpCalculation result of (k) and discharge amount measurement
Based on the discharge amount from the part 16,
(Formula 13) is identified, and B _Three(Z) degree and coefficient
May be determined.

[0088]

[Equation 10] In addition, the autoregressive parameter A as in (Step 2)
(Z) is high-pass characteristic A _h(Z) and low-pass characteristic A
_lIf both characteristics of (z) are present, the model identification unit
21 is A (z) = A_h(Z) A_l(Z) below (14
(Equation 14) to identify this (Equation 14)
It

[0089]

[Equation 11] The model identifying unit 21 considers the continuous excitation condition in both the water level y _rp (k) that is the output and the pump discharge amount u ₃ (t) that is the input by identifying (Equation 14). be able to.

Generally, when identifying an FIR model, the number of parameters is large, and if sufficient number and sufficient quality of data are not obtained, the identification accuracy is deteriorated as compared with the ARX model.

As described above, according to the present embodiment, since the FIR model part is identified after the ARX model is identified, the number of parameters can be suppressed and the identification accuracy can be improved. You can

Further, since the ARX model is usually applied to a model in fine weather in which a sufficient number of data can be prepared, the reliability of the model in fine weather can be maintained. In addition, since the FIR model part is taken into consideration and the FIR model part is identified with regard to the rainfall amount and the pump discharge amount, which are the direct factors at the time of rain when it is considered that the rapid change of the water level occurs at the same time, the prediction becomes a follow-up. You can avoid that.

Third Embodiment Next, a third embodiment of the present invention will be described with reference to FIGS. 1 and 6.

In the third embodiment, in identifying the water level prediction model, the parameters of the auto-regression part and the parameters of the FIR model part are independently obtained based on the information from the river water level fluctuation factor measurement part, and further the auto-regression is performed. The conversion is performed in advance so that the parameters of the part and the parameters of the FIR part are adapted as water level prediction model parameters.

FIG. 1 shows a third embodiment of the river water level predicting apparatus according to the present invention.

The third embodiment is substantially the same as the first embodiment except that the order of each parameter of the water level prediction model and the method of determining the coefficient of each parameter are different. In the third embodiment, detailed description of the same parts as those in the first embodiment will be omitted.

As shown in FIG. 6, the model identifying unit 21 determines the order of the parameter and the coefficient of the parameter regarding the autoregressive parameter 22 based on the water level from the water level measuring unit 5, and independently of this determination, the water level. The water level from the measuring unit 5, the tide level from the tide measuring unit 6, and the rainfall measuring units 7 and 8
And the coefficient of the parameter relating to the rainfall parameter 25 and the tide level parameter (FIR model part) 27 are determined based on the rainfall amount and the tide level parameter 2
7 and 7 are subjected to preset conversion.

Specifically, the model identifying section 21 uses the least squares method to calculate A based on the water level from the water level measuring section 5.
(Z) y (k) = e (k) (autoregressive part) is identified, and the order and coefficient of the autoregressive parameter A (z) are determined. In addition, the model identifying unit 21 uses the tide level u ₁ (k) from the tide level measuring unit 6 and the water level y from the water level measuring unit 5.
(K) and y (k) = B using the method of least squares
₁₀ (z) u ₁ (k) + e (k) (FIR model part)
To identify the order and coefficient of the tide level parameter B ₁₀ (z).

The model identification unit 21 identifies the water level prediction model (ARX model) of the following (Equation 15) and constructs a confirmed water level prediction model.

[0100]

[Equation 12] Here, the model identification unit 21 performs preset conversion described below on B ₁₀ (z), obtains an FIR model part of the tide level u ₁ (k) obtained as a result, and calculates the FIR
Let the model part be B ₁ (z) in the above (Equation 15). Further, C [zI-A] ^-1 B is the minimum realization of B ₁ (z) / A (z) and is arbitrary realization.
Sets A, B and C so that they are in the controllable canonical form as follows.

[0101]

[Equation 13] The model identification unit 21 substitutes the above (20) into (15). In this way, the parameter of B (z) in (Equation 15) is given as C.

The model identifying unit 21 substitutes the above equation (20) into equation (15), and the tide level u ₁ is calculated from the substituted equation (15).
The FIR model portion of (k) is obtained, and the FIR model portion is defined as B ₁ (z) in (Equation 15).

The model identification unit 21 uses other external parameters (rainfall amount parameter B ₂ (z) 25, discharge amount parameter B ₃ (z) 23, and opening / closing degree parameter B).
₄ (z) 24), the tide level parameter B ₁ (z)
Similarly, determine the order and coefficient of each parameter.

By applying the above conversion, the autoregressive parameter 22 and the parameter of the FIR model part 28 are set to A
It can be used as a parameter of the RX model.

The feature of this method is that the identification can be performed while checking the causal relationship between the input and the output, and the identification can be performed in consideration of the actual phenomenon.

The model identifying unit 21 performs conversion using the controllable canonical form represented by (Equation 14) when the external input is one tide level.

In general, when the river water level is predicted, a plurality of external inputs are often used, but the conversion using the controllable canonical form cannot represent the case of having a plurality of inputs. Therefore,
When the model identification unit 21 has a plurality of inputs, (16
The observable canonical form of (Equation 22) is used instead of (Equation).

Here, the model identifying section 21 determines the following (21
The determined water level prediction model is constructed using the water level prediction model (ARX model) represented by the formula), and similarly, the water level prediction model (ARX model) represented by the above (1 formula) is used, A fixed water level prediction model may be constructed.

[0109]

[Equation 14] In addition, in said (21 type | formula), each symbol has shown the following.

U ₁ (k): tide level (input) from the tide level measurement unit u ₂ (k): rainfall amount (input) from the rainfall amount measurement unit y (k): river water level (output) e (k) : Noise part A (z): Autoregressive parameter B ₁ (z): Tide level parameter (FIR model part for tide level (input)) 27 B ₂ (z): Rainfall parameter (FIR partial model for rainfall amount (input)) 25 The model identification unit 21 separately identifies the tide level parameter B ₁ (z) and the rainfall amount parameter B ₂ (z) of the above (21 expression). Specifically, the model identifying unit 21 uses A (z) based on the water level from the water level measuring unit 5 using the least squares method.
Identify y (k) = 0 (autoregressive model) and determine the order and coefficients of the autoregressive parameter A (z). In addition, the model identifying unit 21 uses the tide level u ₁ (k) from the tide level measuring unit 6.
And y (k) = B ₁₀ (z) u ₁ (k) (FIR model part) are identified using the least squares method based on the water level from the water level measurement unit 5 and the tide level parameter B ₁₀ (z) Determine the order and coefficient. Further, the model identifying unit 21 is configured to measure the rainfall amount u ₂ (t) from the rainfall amount measuring units 7 and 8 and the water level measuring unit 5.
Based on the water level from y (k) =
B ₂₀ (z) u ₂ (k) (FIR model part) is identified, and the order and coefficient of the rainfall parameter B ₂₀ (z) are determined.

Here, the model identifying unit 21 determines B ₁₀ (z)
Is subjected to preset conversion described below, the FIR model portion 28 of u ₁ (k) obtained as a result is obtained, and this FIR model portion 28 is referred to as B ₁ (z). Similarly, the model identification unit 21 performs a preset conversion shown below on B ₂₀ (z), and the resulting u is obtained.
The FIR model portion 28 of ₂ (k) is obtained, and the FIR portion 28 is designated as B ₂ (z).

The model identifying unit 21 is C [zI which is the minimum realization of B ₁ (z) / A (z) as in the above-mentioned conversion.
-A] ^-1 B ₁ is used to convert B ₁₀ (z) into B ₁ (z). Similarly, the model identification unit 21 determines that B ₂ (z)
/ A with minimal realization is _{^{C [zI-A] -1 B}} 2 of _(z), converts _B 20 a (z) _B to 2 (z). Here, the model identifying unit 21 sets A, B ₁ , B ₂ and C so as to be in the observable canonical form as follows.

[0113]

[Equation 15] On the other hand, B₁₀(Z), B₂₀For (z), y
(K) = B₁₀(Z) u₁(K), y (k) = B
₂₀(Z) u_Two(K), B₁₀(Z) = b₁₀₁z^-1
+ B₁₀₂z^-2+ ... + b_10mz^-M, B
₂₀(Z) = b₂₀₁z^-1+ B ₂₀₂z^-2+ ...
+ B_20nz^-NSince,

[Equation 16] The model identifying unit 21 obtains B ₁ from the above (Equation 29), and determines the order and coefficient of B ₁ (z) in (Equation 21). In addition, the model identification unit is
₂ is determined, and the order and coefficient of B ₂ (z) in (21) are determined.

Similarly, the model identifying unit 21 determines that the discharge amount parameter B ₃ (z) and the opening / closing degree parameter B
_4. Determine the orders and coefficients of ₄ (z).

As described above, according to the present embodiment, the identification can be performed while looking at the direct causal relationship between the external input directly affecting the water level and the river water level which is the output. It becomes possible to perform identification while considering. Therefore, if a satisfactory parameter identification result is not obtained, it is possible to easily make a correction so as to improve the prediction accuracy of the water level prediction model in consideration of the direct causal relationship.

That is, as shown in FIG. 3, according to the present invention, the water level prediction model 20 includes an ARX model having an autoregressive part and an FIR model part, and the parameters of these autoregressive part and the FIR model part are The model identification unit 21 independently determines the different algorithms. Next, the parameters of the auto-regression part and the parameters of the FIR model part obtained by the model identification unit 21 are converted, and the water level prediction model 20 is established.

Next, using the water level prediction model 20 established in this way, the pump discharge amount, which is a river variation factor,
It is possible to accurately predict the water level of a river based on past or present measured or calculated values of rainfall, water level, tide level, and the like.

On the other hand, for example, when the parameters of the auto-regressive part and the parameters of the FIR model part are all simultaneously obtained by using the least square method with limited information, the auto-regressive part is relatively predictive. Parameters are identified to have a large impact. For this reason, when the FIR model portion that is an external input changes abruptly, the fluctuation of the external input cannot be sufficiently dealt with, and the water level prediction of the river may be behind.

[0119]

According to the present invention, the river water level can be predicted with respect to disturbances with high robustness such that the accuracy of river water level prediction does not deteriorate, and with high accuracy even with rapid water level fluctuations. Further, it becomes possible to provide the operator with effective driving support information for the drainage pump station based on the predicted river water level information.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of a river water level prediction device according to the present invention.

FIG. 2 is a diagram showing a river water level prediction device and a drainage pump station.

FIG. 3 is a diagram showing a method of identifying a water level prediction model together with a comparative example.

FIG. 4 is a diagram showing an identification method of a water level prediction model according to the first embodiment of the present invention.

FIG. 5 is a diagram showing an identification method of a water level prediction model according to the second embodiment of the present invention.

FIG. 6 is a diagram showing an identification method of a water level prediction model according to the third embodiment of the present invention.

[Explanation of symbols]

2 rivers 4 waterways 5 Water level measurement unit 6 tide level measurement section 7 Rainfall measurement section 8 Rainfall measurement part 11 River water level prediction device 15 Opening / closing degree measurement unit 16 Discharge rate measurement unit 20 Water level prediction model 21 Model Identification Section 22 Autoregressive part 23 Discharge rate parameter 24 Open / close degree parameter 25 rainfall parameters 26 Water level parameters 27 tide level parameters

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Akihiro Iwa             No. 1 Toshiba-cho, Fuchu-shi, Tokyo Toshiba Corporation             Fuchu Office (72) Inventor Shunsuke Iwasaki             No. 1 Toshiba-cho, Fuchu-shi, Tokyo Toshiba Corporation             Fuchu Office (72) Inventor Masaki Kunimi             1-1 Shibaura, Minato-ku, Tokyo Co., Ltd.             Toshiba headquarters office F-term (reference) 5H004 GA01 GA28 GB08 HA02 HA05                       HB02 HB05 JA01 KA72 KC24                       KC26 KC27                 5H223 AA15 BB05 CC08 DD07 FF05

Claims (7)

[Claims] [Claim 1] A river water level fluctuation factor measuring unit for measuring a river water level fluctuation factor, and a river water level is predicted based on information from the river water level fluctuation factor measuring unit, and based on the past to present water levels. Identify the water level prediction model, which has a first prediction model part that predicts the water level and a second prediction model part that predicts the water level based on information from the river water level fluctuation factor measurement part from the past to the present. And a model identification unit for performing, and the model identification unit obtains at least one of the parameters of the first prediction model portion and the second prediction model portion of the water level prediction model based on the information from the river water level fluctuation factor measurement unit. A water level prediction device for rivers. 2. A water level prediction model has a noise part obtained by filtering a noise part in advance, parameters of a first prediction model part are specified by designing a filter, and a model identification part is a second prediction model. The water level predicting apparatus for a river according to claim 1, wherein only the partial parameters are obtained. 3. A plurality of river fluctuation factors are set, and the model identification unit determines the parameters of the first prediction model part of the water level prediction model and the relevant river water level fluctuation factors based on a specific river water level fluctuation factor among the river fluctuation factors. Corresponding to other river water level fluctuation factors based on the obtained parameters of the first prediction model part and the parameters of the second prediction model part corresponding to the relevant river water level fluctuation factors. The water level predicting apparatus for a river according to claim 1, wherein the parameter of the second predictive model portion is calculated. 4. The model identification section independently obtains the parameters of the first prediction model part and the parameters of the second prediction model part of the water level prediction model based on the information from the river water level fluctuation factor measurement section, and further, the first 2. The river water level prediction device according to claim 1, wherein the parameters of the prediction model part and the parameters of the second prediction model part are subjected to preset conversion so as to be adapted as parameters of the water level prediction model. 5. The river water level prediction apparatus according to claim 2, wherein the noise part of the water level prediction model is weighted in a desired frequency band. 6. Based on the parameters of the first prediction model part and the parameters of the second prediction model part corresponding to specific river water level fluctuation factors, the parameters of the second prediction model part corresponding to other river water level fluctuation factors are set. 4. The water level predicting apparatus for a river according to claim 3, wherein a calculation formula adapted to the frequency characteristic of the first predictive model portion is used when obtaining. 7. The observable canonical form is used for conversion so that the parameters of the first prediction model part and the parameters of the second prediction model part are adapted as parameters of the water level prediction model. The river water level prediction device according to claim 4.