JP2005222444A - Method and device for calculating statistical prediction value - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a prediction value calculation device for sewerage performing a prediction calculation without being influenced even if input data of the prediction model are abnormal in calculating a prediction value using a prediction model. <P>SOLUTION: This statistical prediction value calculation method calculates the predicted value of sewerage using the prediction model created by a statistical method based on past data. By determining whether there is an abnormal value in the data from the present to the past in the input of the prediction model based on the autocorrelation function of the input, if there are data determined to be the abnormal value, the data at the time are not used for anticipation and excluded from anticipation operation, interpolation is performed for the data, and the prediction calculation is performed using the data made suitable by interpolation. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a statistical prediction technique, and more particularly to a prediction control method and apparatus for operating a sewer.

  In process control such as rainwater drainage control in sewage, control based on the prediction result of process values has been performed in recent years. For example, in the rainwater drainage control, if the information “when and how much rainwater inflows” can be accurately grasped at the pump station, the start / stop timing of the rainwater pump can be accurately determined.

  As various information to grasp this, there are measurement values obtained from radar rain gauge, ground rain gauge, pipe water level gauge, etc., runoff analysis and inflow prediction are performed based on these measurement values, Pump automatic control utilizing this outflow analysis and inflow prediction is performed.

  As a method for predicting the amount of rainwater inflow, a white box approach (tracking the temporal change of rainfall runoff using physical models and conceptual models) (hydraulic models such as extended RRL method and general runoff analysis software “MOUSE”) , Based on hydrological model) and black box approach (system identification method, multiple regression analysis, etc.) that builds a prediction model only from past rainfall and inflow data.

  In recent years, the latter black box approach has been attracting attention because of the ease of model construction. This black-box approach uses time series data of process values such as actually collected water level and rainfall, and builds a model from that data. This method makes accurate predictions in a short time. Promising when it is intended to do.

  For modeling from time series data, multivariate analysis (multiple regression analysis, etc.), neural networks, system identification methods, and the like are well known, and these methods are widely used in the industrial field. These methods are related to each other and are not independent, but among them, the system identification method is very effective as a model construction method for the purpose of (temporal) prediction and control.

  The biggest feature of the system identification method is that the model structure and parameter identification algorithm are considered with a strong consciousness of the temporal and spatial dynamics of the process during modeling. This makes it possible to build a model based on.

  For example, considering dynamic characteristics expressed as normal filters for disturbance / noise mixed in the process, process characteristics analysis in the frequency domain, process characteristics analysis through pole and zero analysis, and process model reduction It can be carried out.

By performing analysis as a dynamic system and building a model based on the analysis, it is possible to expect a significant improvement in prediction accuracy. In addition, the rainwater inflow prediction using the system identification method is disclosed in Patent Document 1, for example.
JP 2000-257140 A

However, even if a prediction model can be constructed with high accuracy, if an abnormal value is included in the input of the prediction model, the output of the prediction model as a prediction result may be unreliable.

Therefore, when an abnormal value is included in the input of the prediction model, it is required to determine the abnormality of the prediction model input by the abnormal value determination method, exclude the data, and interpolate with other data. Yes. As a method for processing an abnormal value of an input of a prediction model, a conventional method using a low-pass filter or a moving average process is used.

However, since these methods cause a delay in the signal, there is a case where the prediction accuracy is deteriorated for a process that requires a highly accurate prediction result in a short time. In addition, since the above-described filter and moving average processing are always performed on all input signals, changes in signals that should not be processed are also rounded by processing, and input of a prediction model is performed. May cause an effect (information) to be lost on the output of the prediction model.

  The present invention has been made in consideration of the above-mentioned points, and when calculating a prediction value using a prediction model, even if there is an abnormality in the input data of the prediction model, the prediction calculation is not affected. It aims at providing the predicted value calculation apparatus of the sewer to perform.

In order to achieve the above object, in the present invention,
In the statistical prediction value calculation method of calculating a prediction value using a prediction model created by a statistical method based on past data, the input whether or not there is an abnormal value in the past data from the present in the input of the prediction model If there is data determined to be an abnormal value based on the autocorrelation function of, the data at that time is excluded from the prediction calculation without being used for prediction, and the data is interpolated, Statistical prediction value calculation method characterized by performing prediction calculation using data optimized by interpolation, and statistical calculation to calculate prediction value using prediction model created by statistical method based on past data In the predicted value calculation device, it is determined based on the autocorrelation function of the input whether there is an abnormal value in the past data from the present in the input of the predicted model. If there is data that is determined to be an abnormal value and data that is determined to be abnormal, the data at that time is not used for prediction and is excluded from the prediction calculation, and the data is interpolated and data is optimized. A statistical prediction value calculation device comprising: processing means; and performing prediction calculation using data optimized by the processing means;
Is to provide.

  According to the present invention, when an abnormality is included in the data, the abnormal portion is excluded, the normalized data is formed by interpolation, and the prediction calculation is performed based on the normalized data. It is possible to perform a prediction calculation that is not affected by any abnormality. Thereby, the prediction control using the prediction calculation result and the improvement of driving assistance accuracy can be expected.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Here, the rainwater drainage process at the rainwater drainage pumping station will be described as an example.

(Configuration of Example 1)
FIG. 1 shows a rainwater drainage treatment system using a statistical prediction value calculation apparatus according to the present invention. As shown in FIG. 1, in the statistical prediction value calculation apparatus 11 according to the present invention, rainwater 10 flows into a pump well 3 via an inflow trunk line 1 and an inflow trough 2, and this rainwater 10 is supplied to a plurality of rainwater pumps. This is applied to a process represented by the rainwater pump 5 among the processes discharged to the river by 5.

The statistical predicted value calculation unit 11 receives the signals of the rainfall signal 12 from the ground rain gauge 7, the operation status of the rainwater pump 5, the inflow water level 13, the pump well level 14, and the operation status of the inflow gate 4. The predicted value is calculated. Then, the prediction calculation result is sent to the monitoring control device 17 via the transmission line 20.

  The rainwater drainage control relating to the operation of the rainwater pump 5 is performed by the control device 11A. Here, when this rainwater drainage control system is a hybrid system and hybrid control is performed, a variable conversion unit that converts a discrete time system into a continuous time system may be provided in the control device 11A.

  In order to improve the reliability of the rainwater drainage control device, the control device 11A is configured by a plurality of control devices, and at the same time, a control performance determination unit and an optimal control evaluation unit (not shown) are provided.

  FIG. 2 is a block diagram showing a detailed configuration of the statistical prediction value calculation apparatus 11 shown in FIG. The statistical prediction value calculation device 11 includes a prediction model input abnormal value determination unit 101, a prediction model input interpolation unit 102, a prediction model parameter calculation unit 103, and a prediction model 104.

  Usually, a prediction model input is input to the prediction model parameter calculation unit 103, a parameter calculated based on the information is given to the prediction model 104, and a prediction value is calculated from the input of the prediction model. When there is an abnormal value in the input of the prediction model, the data determined to be abnormal by the prediction model input abnormal value determination unit 101 is excluded, and the excluded part is interpolated by the prediction model input interpolation unit 102. Based on the prediction model, the predicted value is calculated.

(Operation of Example 1)
Hereinafter, the operation of the first embodiment will be described. In addition, the effect | action described here is an effect | action when the rainwater pump 5 is drive | operated by 11 A of control apparatuses based on the result of the statistical prediction value calculating apparatus 11 shown in FIG. Here, the inflow amount to the pump station is predicted as a constituent element of the statistical prediction value calculation device 11 by a system identification method that is a black box approach, and the number of operating rainwater pumps is controlled based on the prediction result. A rainwater drainage control system having a control device will be described. As a black box approach, a model using multiple regression analysis or a neural network may be adopted.

First, a prediction model input is input to the prediction model parameter calculation unit 103, a parameter calculated based on the information is given to the prediction model 104, and a prediction value is calculated from the prediction model input. If there is an abnormal value in the prediction model input, the prediction model input abnormal value determination unit 101 determines that there is an abnormality, and the data is excluded.

As a means for determining data abnormality, an autocorrelation function of a prediction model input is used. As a criterion for abnormality determination, a predetermined threshold value of the autocorrelation function is used, and data exceeding this threshold value is determined as abnormal. The prediction model input abnormal value determination unit 101 can exclude an abnormal value that causes deterioration in prediction accuracy, and can prevent deterioration in prediction accuracy.

Subsequently, the excluded data is interpolated by the prediction model input interpolation unit 102. As a method for interpolating data, spline interpolation or the like is used. A cubic spline curve or higher is used for smoother interpolation. As a result of the abnormal value determination means described above, the interpolation process can smoothly interpolate the excluded prediction model input data, and input to the prediction model without impairing the essence of the process value fluctuation. . Based on the interpolated data, a predicted value is calculated using a prediction model.

The abnormal value determination means and the interpolation processing described above may be performed offline, but a mechanism for performing online processing at a predetermined cycle may be provided. By using it online, this statistical prediction value calculation device can be used in a form directly connected to online control, and control performance in various controls can be enhanced.

The inflow prediction model is offline based on the signals such as the rainfall signal 12 from the ground rain gauge 7, the operation status of the rainwater pump 5, the inflow water level 13, the pump well level 14, and the operation status of the inflow gate 4. Or a mechanism for sequentially updating the prediction model online may be provided.

In this way, by sequentially updating the prediction model online, the model is updated following changes in the basin, and deterioration in prediction accuracy can be suppressed. As a typical example of a method for updating a model sequentially online, a sequential least square method in a system identification method is known. This sequential least squares method has a small calculation load and can be implemented on a relatively small computer. At the same time, it has the advantage that model parameters can be updated following process fluctuations. There is no.

However, it is very effective to provide a model updating means for systems where process fluctuations appear as a result of complex factors of natural phenomena, such as rainwater drainage processes, and fluctuations are likely to occur. It can be said that a mechanism that sequentially updates the prediction model online is effective.

  In addition, when a rainwater drainage control is performed by a certain control device, even if a certain control device does not operate normally, it automatically switches to another control device, without stopping the rainwater drainage automatic control, In order to continuously and effectively carry out rainwater drainage control, it is also effective to construct a composite rainwater drainage control device having a plurality of control devices.

  Specifically, the control performance determination unit determines the control performance of a plurality of control devices that are a combination of the prediction model and the control model of the statistical prediction value calculation device. Regarding the determination of a plurality of different control outputs, the performance of the plurality of control devices is determined by an optimal control using an operation amount, pump operation time, pump operation cost, etc. as an index at a predetermined cycle. The period is, for example, a control period, and can be reset according to operation.

  Then, a control model having optimal control performance is selected for each predetermined period, that is, for each period in accordance with operation, and the output of the control device is input to the process control unit to perform optimal rainwater drainage control. When it is determined that a certain control device does not operate normally due to an abnormality diagnosis result such as a sensor value abnormality, a mechanism is automatically switched to the other control device immediately and the control device operates. Provide.

  By solving the above-mentioned optimal control problem, it is possible to quantitatively evaluate the control problem of the stormwater drainage process, so that a stormwater drainage control method and apparatus capable of leveling the pump operation time and reducing the number of operation switching are provided. can do. Performing the above-described control performance determination, that is, determination based on abnormality diagnosis and determination based on the evaluation function value leads to high reliability of rainwater drainage control, and stable automatic control operation can be performed.

  In addition, when performing control performance determination offline, it takes time to separate the system from the system and incorporate it into the system again. This may take time, and automatic control may be continued when the machine is unmanned. There is a problem that it is impossible to deal with when it is necessary. On the other hand, if the control performance determination is performed online as in the present invention, the above problem can be solved. Conventionally, there is no example dealing with the optimization problem including such cost in the rainwater drainage control problem, and it can be said that it is an effective method.

  For the configuration of this optimal control system, for example, receiving horizon control known as a basic concept of model predictive control is applied. This receiving horizon control is a control in which online optimization is repeated while shifting an interval for solving the optimization problem for each sampling period, and the constraint condition can be handled in an explicit form.

  That is, there is an advantage that the constraint condition concerning the control input and the process output can be directly reflected in the control algorithm. In this case, the upper and lower limits of the pump well level are considered as a constraint condition. As described above, a given rainwater drainage control problem can be reduced to an optimization problem including a constraint condition.

  As a method for solving this optimal control problem, it is conceivable to apply a branch and bound method or a genetic algorithm. The branch and bound method can eliminate unnecessary search procedures in advance by checking a partial space in the solution space at a time and verifying beforehand whether or not a solution candidate can exist in the space. Has the advantage of being able to search for the optimal solution. Genetic algorithms can only search for suboptimal solutions, but not optimal solutions, but have the advantage that they cannot easily fall into local minima and can expect high-speed computation.

  As described above, the system can be handled systematically by applying model predictive control, that is, receiving horizon control, to the combined stormwater drainage control problem, and stormwater drainage control is always performed with a control device with optimal control performance. be able to. For this reason, it is possible to establish a rainwater drainage automatic control system that can level the pump operation time, reduce the number of operation switching times, and solve problems such as the intervention of the manual operation of the operator.

Note that the above method assumes that the pump station inflow, the pump well inflow, and the pump well water level are predicted using only the information up to the present of the ground rain gauge 7. If only the ground rain gauge 7 is used, it will not be possible to make accurate rain predictions, so predictions of about 10 to 15 minutes ahead are considered the limit.

Provided with the rainwater drainage control, the rainwater drainage control apparatus incorporating the prediction information of 30 minutes to several hours ahead is taken into consideration that the rainfall and the prediction value of the rainfall intensity based on the information of the rainfall radar 18 are taken into consideration. Is also possible. The data observed by the rainfall radar 18 is subjected to various types of data processing by the rainfall radar data processing device 19, and it is possible to predict the rainfall situation up to a certain time ahead.

Various weather data including forecasts of rainfall conditions are distributed from the weather service center. By sending information from the rainfall radar data processing device 19 and the weather information terminal 21 to the driving support device 17 via the transmission line 20, and performing inflow prediction based on the predicted values of the rainfall and rainfall intensity, The inflow amount can be predicted from 30 minutes to several hours ahead, and as a result, the reliability of rainwater drainage control can be improved.

  Here, a control system in which a plurality of different control devices are mounted involves discrete control devices and mode switching. Moreover, the pump rotation speed control unit and the inflow amount prediction unit of the rainwater drainage control device are continuous time systems, and the pump number control unit is a discrete time system.

  Thus, a system in which a continuous time system and a discrete time system coexist is called a hybrid system, and a rainwater drainage control system is a hybrid control system. If a control system is constructed without considering the hybrid system, a discontinuous phenomenon occurs in the mixed part of the continuous time system and the discrete time system, and the pump start / stop phenomenon is repeatedly induced. there is a possibility.

  As a result, the rainwater drainage control system becomes unstable and control is no longer possible, and it becomes necessary for the operator to switch the automatic control mode to the manual operation mode to operate the pump and the like, increasing the burden on the operator. obtain. Therefore, the control system is regarded as a continuous time system by performing appropriate variable conversion.

  Thus, since the rainwater drainage control system is a continuous time system, it is possible to systematically handle this system that has not been easy to systematically construct a control system until now because it is a hybrid system. Then, it can be expected that problems such as repeated stoppage of pump operation, destabilization of the rotation speed control, and manual intervention associated therewith, which may occur due to a mixture of continuous time systems and discrete time systems, can be expected.

  FIG. 3 is a block diagram showing a detailed configuration of the rainwater drainage control device 11 in FIG. The rainwater drainage control device 11 includes a process data input unit 201 that inputs process data from the target device O via an online signal line L, a process data quality determination unit 202 that determines quality of the input process data, and a prediction. TT rainwater drainage control amount calculation units J1, J2,..., JT in which a model 203 and a control model 204 are provided, respectively, and optimum control evaluation means 206 are provided in the inside, and these calculation units J1, J2,. .., JT, and a control selection unit 205 that selects one of JT, and a process control unit 207 that controls the target device O based on the calculation result of the selected calculation unit. The operator can manually operate the target device O using the manual operation means 208. The manual operation means 208 may be dedicated hardware, or may be software operated via the monitoring control device 17.

  The target device O refers to the rainwater pumps P1, P2,..., PM, the discharge valves V1, V2,..., VM, and the inflow gate 4, etc., but is connected to the process data input unit 201 via the signal line L. The signals to be sent include signals related to these target devices such as water level gauges 8 and 9 in addition to signals related to these.

  Next, the operation of the apparatus shown in FIG. 3 will be described. The process data input unit 201 inputs process data from the target device O via the online signal line L, and outputs this to the process data pass / fail determination unit 202. Then, the process data pass / fail determination unit 202 determines pass / fail of the process data.

  For example, when a certain data value indicates an abnormal value of a certain number or more, the process data pass / fail judgment unit 202 displays this abnormality on the monitoring control device 17 (FIG. 1), and the operator sees the abnormality display. The automatic control is immediately stopped and manual operation is performed using the manual operation means 208.

  In the apparatus configuration shown in FIG. 3, T rainwater drainage control amount calculation units J1, J2,..., JT are provided as having the function of calculating the rainwater drainage control amount. When the system is started, one of the T rainwater drainage control amount calculation units J1, J2,..., JT is selected in advance by the control selection unit 205 as the calculation unit used for the actual control. .

  As selection criteria in this case, for example, the operation time is the shortest, the operation cost is the lowest, the operation amount is the smallest (the number of times the number of pumps is switched is the smallest), etc. .

  For example, assuming that the rainwater drainage control amount calculation unit J1 is selected, the prediction model 203 of the calculation unit J1 is based on data input via the process data pass / fail judgment unit 12 (for example, rainwater inflow The control model 204 calculates a predetermined control amount (for example, the number of pumps operated and the number of pump rotations) based on the calculated predicted amount.

  The calculated control amount is output to the process control unit 207 via the control selection unit 15, and the process control unit 207 performs process control on the target device O based on the input control amount.

  When the control selection unit 205 determines that the reliability of the rainwater drainage control amount calculation unit J1 has been lost during the operation control for the system, the control selection unit 205 immediately selects the rainwater drainage control amount calculation unit J1. Try to stop.

  Here, as to whether or not the reliability is impaired, for example, it is determined whether or not the value of the evaluation function for the operation time, the operation cost, or the operation amount calculated by the optimal control evaluation unit 16 is larger than a certain value. Can be done.

  Next, the optimum control evaluation means 206 calculates the evaluation functions of the rainwater drainage control amount calculation units J2,..., JT, and selects the rainwater drainage control amount calculation unit having the smallest evaluation function value among them. To do.

  According to the configuration as described above, since a plurality of types of software having a function of calculating the rainwater drainage control amount are prepared in advance, while selecting one of them and executing the operation control, Even if the reliability of the selected software falls, it is possible to immediately select the optimum software from the other software and continue the operation control without stopping.

  As each prediction model 203 in the rainwater drainage control amount calculation unit J1, J2,..., JT, either software that performs a white box approach or software that performs a black box approach may be employed. it can. Software that performs a white-box approach can include “MOUSE”, a commercial outflow analysis package software that performs pump station inflow prediction, but also includes an extended RRL method, hydraulic model, hydrological model, etc. An outflow analysis model based on the above may be adopted.

  On the other hand, as a software that performs a black box approach, there is a model based on a system identification method that performs pump station inflow prediction, but a model based on multiple regression analysis or a neural network may be adopted.

  In addition, for various parameters used by the prediction model 203, the control selection unit 205 performs measurement data from the water level gauges 8, 9, the ground rain gauge 7, and the rain radar data processor 19, or the rain water pumps 5, 5,. Based on the data on the operational status of the gate 3 and the like, it can be determined offline in advance.

  On the other hand, according to the configuration of FIGS. 1 to 3, the control selection unit 205 can capture these data online, so that the parameters of the prediction model 203 can be updated sequentially. Thus, by sequentially updating the prediction model 203 online, the model can be updated following changes in the basin, and deterioration of prediction accuracy can be suppressed.

  As a typical method for sequentially updating the prediction model 203 online, a sequential least square method in a system identification method is known. This sequential least squares method has a small calculation load and can be implemented even on a relatively small computer. In addition, the model parameter can be updated following process fluctuations. There are no examples of actual use in the field of control.

  However, as in the case of the stormwater drainage process, for the system in which the process variation appears as a result of the complex factor of the natural phenomenon and the variation is likely to occur, the means for sequentially updating the prediction model 203 online as described above is provided. It can be said that it is very effective.

  The control model 204 in this embodiment performs pump control on the inflow amount predicted by the prediction model 203. As the pump control method, the number of pumps operated may be determined directly based on the inflow prediction value, or a pump well water level prediction model that separately outputs the pump well water level prediction value by inputting the inflow prediction value, The structure which determines the number of pump operation based on a pump well water level prediction value can also be employ | adopted.

  Although the set water level automatic control using the fixed pump well level setting value is a method conventionally performed, the control model 204 in this embodiment corrects the set water level based on the predicted inflow amount from the prediction model 203. The start timing or stop timing of the pump can be appropriately changed based on the set water level after correction.

  FIG. 5 is a flowchart showing an outline of the function of the rainwater drainage control amount calculation unit J1. The inflow amount prediction model 203A receives process data determined to be good from the process data pass / fail determination unit 202, and predicts a rainwater inflow amount Q (continuous time event value) based on the process data (step 1).

  When the rainwater inflow Q of the predicted continuous time event value is input, the variable defining unit 209 defines auxiliary variables δ for converting the continuous time event value into a discrete time event value (step 2). The auxiliary variable δ and the amount of rainwater inflow Q are output to the pump operation number determination control unit 204A. The pump operation number determination control unit 204A calculates the necessary pump operation number N by applying the input auxiliary variable δ and the rainwater inflow amount Q to a predetermined arithmetic expression (step 3).

  In other words, the pump operation number determination control unit 204A uses the auxiliary variable δ, and based on the input of the rainwater inflow amount Q that is a continuous time event value, the pump operation number N can be treated as a continuous time system variable. Convert to time event value.

  Also, when the variable definition unit 209 receives the pump operation number N as a discrete time event value from the pump operation number determination control unit 204A, the variable definition unit 209 defines an auxiliary variable γ for converting the discrete time event value into a continuous time event value. (Step 4), the auxiliary variable γ and the number N of pumps to be operated are output to the pump speed controller 204C. At this time, the deviation between the set value from the pump well level setting unit 204B and the detected value from the water level gauge 9 is output from the subtractor 211 to the pump rotation number control unit 204C.

  The pump rotation speed control unit 204C applies the auxiliary variable γ and the pump operation number N input from the variable definition unit 209 to a predetermined arithmetic expression while considering the deviation from the subtractor 211, and calculates the pump rotation speed R. Calculate (step 5). In other words, the pump rotation speed control unit 204C uses the auxiliary variable γ to treat the pump operation number N, which is a discrete time event value, as a continuous time system variable, and treats this as the pump rotation speed R, which is a continuous time event value. It has been converted.

  Next, a specific example of a method for calculating the number N of operating pumps from the rainwater inflow amount Q which is a continuous time event value using the auxiliary variable δ in the rainwater drainage control amount calculation unit J1 of FIG. 4 will be described. When the rainwater inflow at a certain time t is Q (t) and the rainfall as the process data is u (t), the rainwater inflow Q (t + 1) at the time t + 1 is expressed by the following equation (1). In addition, a system in which the number of pumps N (t) is directly calculated from the rainwater inflow amount Q (t) using the following formulas (2) to (4) will be considered.

  Note that a and b are parameters, and can be determined in advance offline or can be determined while sequentially changing online. Further, although the units are omitted in the formulas (1) to (4), Q (t) and u (t) are, for example, [m3 / s: cubic meters / second], and N (t) is [unit]. .

Q (t + 1) = a * Q (t) + b * u (t) (1)
N (t) = 1 if Q (t) ≦ 3 (2)
N (t) = 2 if 3 <Q (t) ≦ 6 (3)
N (t) = 3 if 6 <Q (t) (4)
Since the relationship between N (t) and Q (t) in the above equations (2) to (4) is a relationship between continuous value input and discrete value output, the stormwater drainage control amount calculation unit J1 in FIG. By introducing an auxiliary variable to, the output is converted into a form that can be treated as a continuous variable. For example, an MLD (Mixed Logical Dynamical) system can be considered as a specific method of introducing the auxiliary variable.

  The MLD system is a dynamic system representation method that includes a logical description. It expresses discrete events included in a process as logical variables, and its formulation is characterized by the expression of logical expressions as linear inequalities. It is.

  By using this concept of the MLD system in the relationship between the predicted inflow amount or the predicted pump well level and the number of pumps operated, the discrete output can be associated as a continuous variable. Now, when a discrete value auxiliary variable δi (i = 1, 2, 3) is introduced in order to associate a discrete value output with an output as a continuous variable, this variable δi is represented by the variable definition unit 209 (5) to (7) below. It is defined as an expression.

δ1 = 1 if Q (t) ≦ 3 (5)
δ2 = 1 if Q (t) ≦ 6 (6)
δ3 = 1 (7)
If the auxiliary variable δi defined in this way (i = 1, 2, 3) is used, the pump operation number determination control unit 204A sets the pump operation number N (t) as an output as shown in the following equation (8). It can be expressed as a continuous time representation of an MLD system.

N (t) = 1 · δ1 (t) + 2 · {δ2 (t) −δ1 (t)} + 3 · {δ3 (t) −δ2 (t)} (8)
When specific numerical values are substituted into the above equations (5) to (8), for example, when Q (t) = 2, δ1, δ2, and δ3 in equations (5) to (7) are δ1 = 1. Since δ2 = 1 and δ3 = 1, N (t) in the equation (8) is N (t) = 1 · 1 + 2 · (1-1) + 3 · (1-1) = 1 + 0 + 0 = 1 [ Stand].

  Similarly, for example, when Q (t) = 4, δ1, δ2, and δ3 in equations (5) to (7) are δ1 = 0, δ2 = 1, and δ3 = 1, respectively. N (t) is N (t) = 1 · 0 + 2 · (1-0) + 3 · (1-1) = 0 + 2 + 0 = 2 [units].

  For example, when Q (t) = 7, δ1, δ2, and δ3 in the equations (5) to (7) are δ1 = 0, δ2 = 0, and δ3 = 1, respectively. N (t) is N (t) = 1 · 0 + 2 · (0−0) + 3 · (1−0) = 0 + 0 + 3 = 3 [units].

  In the example described above, the case where the pump operation number determination control unit 204A outputs the discrete time event value N based on the input of the continuous time event value Q is explained. The same applies to the case where the continuous-time event value R is output based on N inputs.

  Therefore, the rainwater drainage control system is a continuous time system. Therefore, the present invention can systematically handle a system that has not been easy to construct a control system systematically because it is a hybrid system.

  Therefore, problems such as repeated operation and stoppage of pumps, unstable rotation speed control, and increased operator burden due to manual operation interventions caused by the mixture of continuous-time and discrete-time systems. Can be expected to reduce.

Other examples

  In the first embodiment, the rainwater drainage process in the rainwater drainage pump station is targeted as a specific process, but the idea of the first embodiment can be applied to other processes. For example, it is particularly effective for a process using control based on a prediction result, such as a rainwater drainage process to a river using river water level prediction at a drainage station.

Explanatory drawing which shows an example of the rainwater drainage process to which this invention is applied. The block diagram which shows the structure of one Example of this invention. The block diagram which shows the structure of the rainwater drainage control apparatus 11 in FIG. The block diagram which shows the specific structural example of the rainwater drainage control amount calculating part J1 which is the principal part of other embodiment which concerns on this invention. The flowchart which shows the function of the rainwater drainage control amount calculating part J1 in FIG.

Explanation of symbols

1 Inflow trunk 2 Inflow dredge 3 Pump well 4 Inflow gate 5 Rainwater pump 6 Discharge valve 7 Rain gauge 8 Water level gauge (inflow dredge)
9 Water level gauge (pump well)
DESCRIPTION OF SYMBOLS 10 Inflow rainwater 11 Statistical prediction value calculating apparatus 11A Control apparatus 11C Control apparatus 12 Rainfall amount signal 13 Inflow dredging water level inflow dredge signal 14 Pump well water level signal 15 Rainwater pump standby command 16 Discharge valve opening command 17 Operation support device 18 Rainfall radar 19 Rain radar data processor 20 Transmission path 21 Weather information terminal 201 Process data input unit 202 Process data pass / fail judgment unit 203 Prediction model 204 Control model 205 Control selection unit 206 Optimal control evaluation unit 207 Process control unit 208 Manual operation unit 209, 210 Variables Definition part 211 Subtractor

Claims (10)

In the statistical prediction value calculation method that calculates a prediction value using a prediction model created by a statistical method based on past data,
Determine whether there is an abnormal value in the past data from the present in the input of the prediction model based on the autocorrelation function of the input,
If there is data determined to be an abnormal value, the data at that time is not used for prediction, but excluded from the prediction calculation, and the data is interpolated,
A statistical prediction value calculation method, wherein prediction calculation is performed using data that has been optimized by interpolation. In a statistical prediction value calculation device that calculates a prediction value using a prediction model created by a statistical method based on past data,
Determination means for determining whether there is an abnormal value in the past data from the present in the input of the prediction model based on the autocorrelation function of the input;
When there is data determined to be an abnormal value, the data at that time is not used for prediction, but is excluded from the prediction calculation, and the data is interpolated and processing means for optimizing the data is provided. ,
A statistical prediction value calculation apparatus, wherein prediction calculation is performed using data optimized by the processing means. The apparatus of claim 2.
A statistical predictive value calculation apparatus, wherein whether or not there is an abnormal value in the past data of the input of the prediction model is determined based on an autocorrelation function of the input. The apparatus of claim 2.
If there is data determined to be an abnormal value at the input of the prediction model, the data at that time is not used for prediction, is excluded from the prediction calculation, and the data is interpolated by spline interpolation, Statistical prediction value calculation apparatus characterized by performing prediction calculation using simple data. The apparatus of claim 2.
A statistical prediction value calculation device characterized in that an abnormality of a prediction model input is determined online at a predetermined cycle, an abnormality data interpolation process is performed using the result, and a prediction calculation is performed. The apparatus of claim 2.
A statistical predictive value calculation apparatus comprising a function for updating parameters of the predictive model online.   The statistical predictive value calculation device according to claim 2 is used, and a plurality of prediction devices and control devices are provided, and control is performed by switching to a normal and optimal control device based on a predetermined determination method. Control device.   The statistical predictive value calculation device according to claim 2 is used, and by converting continuous time events and discrete time events mixed in the control system, it is regarded as a process of only continuous time events, and the control system is configured. Control device.   A control apparatus using the statistical prediction value calculation apparatus according to claim 2 and applying optimum control using an operation amount and cost as an index as means for constituting a control system.   A control apparatus using the statistical prediction value calculation apparatus according to claim 2 and applying receiving horizon control with the upper and lower limits of a process value as a constraint as means for configuring an optimal control system.

Images