JP2007092517A - Tunnel ventilation control method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that a parameter of a prediction model becomes improper by an influence of a time change in a control object. <P>SOLUTION: This device has an adaptation control means 104 for estimating and compensating for a value of an error included in a model operation result of predicting a state in a tunnel and a prediction model learning means 105 for updating a time change parameter (for example, a smoke and soot discharge quantity of a vehicle is reduced in response to an improvement in fuel economy) of a model by inversely operating from a result, and asynchronously executes adaptation control and leaning control in parallel, and can avoid deterioration in control performance by automatically acquiring the transition of a long-term physical model, and can compensate for a short-term fluctuation in a physical phenomenon in the tunnel each time by appropriate control. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a control method and apparatus for a control device for ventilating a tunnel. The present invention also relates to a parameter adjustment service method for a tunnel ventilation control apparatus that monitors control results and updates parameters of a control model as necessary.

  Patent Document 1 discloses a technique for performing tunnel ventilation control using a physical model that describes ventilation behavior in a tunnel. Here, the state of future contamination inside the tunnel is predicted using a physical model, and the operation state of the exhaust fan and the jet fan is determined from the result of evaluating the prediction result by fuzzy inference. Further, Patent Document 2 describes a technique for improving control accuracy by maintaining model accuracy against aged change in tunnel process characteristics by learning using a neural network.

Japanese Unexamined Patent Publication No. 5-321598 (paragraph 0012, FIG. 1) Japanese Patent Laid-Open No. 5-141200 (paragraphs 0025 and 0026, FIG. 1)

  In the method described in Patent Document 1, accurate control can be performed when the physical model corresponds to the physical behavior in the actual tunnel. Therefore, there is a problem that the control accuracy is lowered. Since there are various uncertain factors such as the influence of natural winds that cannot be measured and variations in vehicle pollutant displacement, good control cannot often be continued with model-dependent control. In addition, when the characteristics of the target change, such as the improvement of the fuel consumption of a car over a long period of time, the target and the physical model are universally deviated, and the control accuracy is constantly reduced. In this case, control accuracy is not recovered unless the physical model is readjusted, but a great deal of labor is required.

  In the method described in Patent Document 2 in which performance is improved by learning a model, the model error can be reduced by combining the model and the control target by learning. However, the above-described variation effect may be learned in the same manner. In this case, changing the characteristics of the model by learning conversely deteriorates the control accuracy over a long period of time. In addition, since a neural network is used, there is a problem that physical knowledge accumulated about the tunnel cannot be used. Moreover, since the neural network is not readable, the only way to check the validity of the learning result is to actually control it. Therefore, there has been a problem in that a decrease in control accuracy cannot be avoided when variations in control results due to disturbance are learned.

  An object of the present invention is to provide a tunnel ventilation control method and apparatus capable of maintaining high accuracy over a long period of time by overcoming the above-described problems of the prior art and updating control parameters as necessary. is there. Another object is to provide a service method for updating parameters.

  In the present invention, in order to solve the above problems, a prediction model describing behavior of smoke concentration and carbon monoxide concentration in a tunnel, and a prediction model for predicting future smoke concentration and carbon monoxide concentration using the prediction model In a tunnel ventilation control apparatus comprising a calculation means and an operation method determination means for determining a desired jet fan or exhaust fan operation method using the prediction result of the prediction model calculation means, the smoke concentration and carbon monoxide detected from the tunnel Using the actual concentration value and the actual operation results of the jet fan or exhaust fan, calculate backward by calculating at least one of the soot emission, monoxide concentration emission, and equivalent resistance area of the vehicle that the prediction model has. Prediction model learning means for updating parameters of the prediction model according to the result is provided.

  An adaptive control unit that corrects the prediction result of the prediction model calculation unit each time is provided, and the learning control of the prediction model learning unit and the adaptive control of the adaptive control unit are performed in parallel.

  The adaptive control means generates a time series of model errors by paying attention only to the actual value corresponding to the latest prediction result, and has for the prediction result of wind direction / wind force, smoke concentration, CO value according to this Estimate the error value. Then, by subtracting this value from the predicted value, the accuracy of the output of the predicted model calculation means is improved without correcting the parameters of the predicted model. As a result, since the operation plan can be evaluated and the operation method can be determined using the highly accurate prediction result, an appropriate operation plan can be selected. In addition, since variations in actual values and noise reduce the autocorrelation of the model error time series, the influence of variations and noise can be minimized by limiting the output of the adaptive control means in accordance with the magnitude of the autocorrelation. Further, since the estimated error value can be compensated promptly by the next control, the response of compensation can be improved.

  On the other hand, the predictive model learning means learns the parameters of the physical model, such as the amount of smoke emitted from the vehicle, the concentration of carbon monoxide, the wind resistance of the vehicle, etc. Is used to calculate the parameter value according to the current tunnel state. If the parameters of the physical model used for the current control do not reflect the current tunnel state, the calculated parameter is used as a new parameter to learn the physical model and perform learning control.

  The execution timing of the predictive model learning unit is managed by the learning starting unit, and is performed asynchronously with the control execution. In general, when adaptive control and learning control are executed at the same time, there is a case where the control competes and the control performance is impaired. In the present invention, the execution timing of the prediction model learning means is asynchronous with the adaptive control, and a large interval corresponding to the transition of the parameters of the physical model is set, thereby avoiding control competition such as double compensation. Therefore, it is possible to automatically acquire a long-term physical model transition to avoid a decrease in control performance, and to compensate for short-term variations in the physical phenomenon in the tunnel described above by adaptive control each time.

  In the present invention, the tunnel ventilation control device performs back calculation on the adaptive control means for correcting the prediction result of the prediction model calculation means each time and the predetermined parameter of the prediction model from the data detected from the tunnel, and updates the parameter with the reverse calculation result. Predictive model learning means was provided. According to this, even when the physical characteristics of the controlled object change over a long period of time, there is an effect that the control accuracy can be maintained by automatically changing the parameters in the model.

  Further, since the adaptive control and the learning control are implemented in parallel and asynchronously, there is an effect that it is possible to avoid control competition such as double compensation.

  Furthermore, when model learning is performed using a conventional method, there are problems of erroneous learning due to the influence of noise included in the detection data and control response delay due to learning delay. However, in the present invention, high-response compensation is performed by adaptive control that focuses only on the most recent data, and learning control that excludes the influence of noise is performed by model learning using a large amount of accumulated data. There is an effect that can be obtained.

  In addition, since the parameter adjustment service method for fetching the actual data of the control device and updating the parameters of the control model when necessary is provided, a service for maintaining the control performance for a long period of time becomes possible.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram of a tunnel ventilation control device showing a first embodiment of the present invention. The control device 100 has an operation plan generation means 101 that determines and outputs the next operation plan (whether or not the jet fan 154 and the exhaust fan 155 are activated, the air volume, etc.). When the operation plan output by the operation plan generation means 101 is adopted, the wind direction / wind speed, smoke concentration, CO concentration, energy consumption, and the number of start / stop times of the jet fan 154 and the exhaust fan 155 are predicted. Prediction model calculation means 102 for calculating using the model 103 is included. Further, it has adaptive control means 104 that calculates and compensates for a model error from the driving plan output from the driving method determination means 107 and the detected value detected from the control object 150.

  In addition, a prediction model learning unit 105 is provided that calculates a model parameter determined in advance by back calculation of the prediction model 103 from the operation plan output from the driving method determination unit 107 and the detected value detected from the control target 150, and updates the prediction model. . Further, an operation plan evaluation unit 106 that evaluates the operation plan according to the result of compensation of the result of the prediction model calculation unit 102 by the adaptive control unit 104, and an operation mode determination unit 107 that determines the next operation mode according to the evaluation result of the operation plan. Have Furthermore, it comprises learning activation means 108 that manages and activates the activation timing of the prediction model learning means 105.

  The purpose of the control is to appropriately ventilate the air inside the tunnel 151. In this embodiment, a system for creating a ventilation by creating a flow of air in the longitudinal direction of the tunnel, which is called a longitudinal flow type, for a one-way tunnel. Explained as an example.

  A jet fan 154 and an exhaust fan 155 are attached to create an air flow. In many cases, multiple units can be mounted. The operation of taking out the contaminated gas out of the tunnel 151 is mainly performed by the exhaust fan 155. That is, the wind exhauster 155 sends wind upward and discharges the air in the tunnel 151 to the outside of the tunnel through the vertical pile 153. On the other hand, the jet fan 154 minimizes the amount of leakage of contaminated air from the pile port 152 by sending wind in the direction opposite to the traveling direction of the vehicle.

  In the present embodiment, the following detectors are installed in the tunnel. The traffic counter 162 detects in advance the number of vehicles entering the tunnel, the speed, and the large vehicle mixture ratio. The wind direction and wind force are detected by AV meters 156 and 157, the smoke concentration is detected by VI meters 158 and 159, and the CO concentration is detected by CO meters 160 and 161. Hereinafter, the value of wind direction and wind force is referred to as AV value. Such a detector is installed in a general tunnel. The control device 100 predicts the future state using the prediction model 103 while detecting the current state in the tunnel from the signal from the detector, and determines the appropriate operation mode of the exhaust fan 155 and the jet fan 154. .

FIG. 2 shows an algorithm executed by the operation plan generation unit 101. First, the current driving method is fetched from the driving method determination means 106 in S2-1. Based on this, a plurality of possible next operation plans are generated. For example, several proposals such as “one exhaust fan operation, air volume 200 m ³ / min, high speed operation with two jet fans” are generated. Normally, it is sufficient to generate a driving plan in the vicinity of the current driving method as a driving plan. However, when the smoke concentration changes greatly, it is necessary to generate a large number of driving plans in a wide range and expand the selection range. There is also.

FIG. 3 shows an algorithm executed by the prediction model calculation means 102. In S <b> 3-1, current results are acquired from each sensor of the control target 150. The next operation plan is taken in from the operation plan generation means 101. Usually, a plurality of operation plans are generated. In this case, the following processing is repeated for each operation plan. In S3-2, the wind speed of each part in the tunnel is calculated. The calculation method is detailed in, for example, “Road Tunnel Technical Standard (Ventilation) / Description” (Japan Road Border Edition, December 1985). By dividing the inside of the tunnel into several meshes and using one set describing the dynamics of the gas flow in the tunnel, it can be solved numerically.
(∂u / ∂t) = f (u) / M (1)
Where u: wind speed in the roadway, M: total mass of air in the tunnel, f (u): total external force, t: time.

In S3-3, the smoke concentration (VI value) and CO concentration (CO value) inside the tunnel are calculated. Each concentration is known to follow two convective diffusion equations.
(∂c / ∂t) = - u (∂c / ∂X) + D (∂ 2 c / ∂X 2) + q ... (2)
Here, D: diffusion coefficient, c: smoke or carbon monoxide concentration, q: discharge amount of pollutant, X: position in the tunnel axis direction.

  Similarly, after dividing the tunnel into several meshes, the wind speed obtained in S3-2 is applied to u, and the VI and CO values of the pile mouth 152 are set to 0 as boundary conditions. It can be understood that the VI and CO concentrations can be obtained.

Furthermore, the electric power consumption required for operating the jet fan 154 and the exhaust fan 155 is calculated with respect to the operation plan taken in S3-4. Although there is a method of calculating the power consumption amount U with higher accuracy, as a simple example, for example, in the case of the exhaust fan 155, the power consumption amount U can be expressed by a simple mathematical expression using the air amount as shown in Equation (3). A jet fan can be expressed by a similar mathematical expression.
U = Ust * (W / Wst) / η (3)
Here, Ust: rated power consumption, W: current air volume, Wst: rated air volume, η: efficiency.

  Further, for the operation plan taken in, it is examined whether or not the number of operating jet fans 154 and exhaust fans 155 changes, and the number of times of starting and stopping is calculated. When it is necessary to start or stop the number of operating jet fans 154 relative to the current number of operating units, the number of starting operations can be easily correlated by setting the starting number to 1, for example.

  As described above, for the operation plan presented by the operation plan generation unit 101, the predicted value of the control result and the energy consumption amount when the operation plan is adopted are calculated. A plurality of operation plans are usually presented. In this case, it is necessary to repeat the same processing for each operation plan and calculate the predicted value of the corresponding control result, the energy consumption amount, and the like.

  FIG. 4 shows the configuration of the adaptive control means realized in the present invention. The adaptive control unit 104 includes an observer 401, an estimated trend generation unit 402, an actual trend generation unit 404, an error series calculation unit 406, and a prediction error estimation unit 407.

  In this embodiment, a case where the predicted value of the smoke density (VI) is compensated will be described as an example. The observer 401 calculates the VI estimated value by performing the same calculation as the prediction model calculation means 102. That is, when the wind speed taken from the control object 150 and the actual value of VI are set as initial conditions, and the operation method of the exhaust fan 155 and the jet fan 154 taken from the operation plan generation unit 101 is realized, the change in the VI value is The wind speed is obtained according to Equation 1, and is estimated by calculation according to Equation 2.

  The estimated trend generation means 402 edits the output of the observer 401 in time series, and generates an estimated trend 403 that is an estimated value trend. Similarly, the result trend generation means 404 takes in the result of the controlled variable and edits the result trend 405. As shown in the figure, the actual trend 405 and the estimated trend 403 store the values (−2, −3,...) Retrospectively in the past with the current time value as the latest value as the trend.

Further, an error series calculation unit 406 that calculates a difference between the actual trend 405 and the estimated trend 403 to calculate an error trend, and the next VI predicted value predicted by the prediction model calculation unit 102 is expected to be included. Prediction error estimation means 407 for calculating the error value. The prediction error estimation unit 407 performs, for example, a linear calculation of Formula 5 on the error series expressed by Formula 4, and estimates a VI estimation error value VIerr.
Error series Δ5, Δ4, Δ3, Δ2, Δ1, Δ0, (4)
Here, Δi = (VIact) i− (Vest) i, (VIact) i: VI value of the actual trend, (Vest) i: VI value of the estimated trend.
VIerr = (a0 · Δ0 + a1 · Δ1 + a2 · Δ2 + a3 · Δ3 + a4 · Δ4 + a5 · Δ5) / (a0 + a1 + a2 + a3 + a4 + a5) (5)
Here, a0, a1, a2, a3,...: Constants corresponding to the weight of each error.

The calculated VI estimation error is multiplied by a gain 408 (Equation 6) and output from the adaptive control means 104 as the final VI estimated value correction amount VIcomp.
VIcomp = G1 * VIerr (6)
Finally, as shown in FIG. 1, a value obtained by subtracting the output of the adaptive control means 104 from the output of the prediction model calculation means 102 is used for the prediction control.

  Although two VI detectors are provided in FIG. 1, this case can be dealt with by performing the same calculation for each. The operation of the observer 401 can be shared. In this embodiment, the case where the VI prediction error is compensated has been described as an example. However, the case where the CO value prediction error is compensated can also be performed in the same manner. The compensation value for the AV value prediction error can be obtained by omitting the calculation based on the two equations.

  FIG. 5 shows processing performed by the operation plan evaluation means. The operation plan evaluation unit 106 evaluates the appropriateness of the control amount (AV value, VI value, CO value), energy consumption, etc., realized for each of the plurality of operation plans generated by the operation plan generation unit 101, Generate criteria for selecting an operation plan.

  In the present embodiment, a case is shown in which an operation plan is evaluated using predictive fuzzy reasoning to determine an operation method. The foreseeing fuzzy is composed of a combination of a rule and a membership function, and the rule has a form unique to foreseeing fuzzy, such as “The VI value is satisfied by the IF operation plan A and the THEn operation plan A is adopted”.

  First, in S5-1, the predicted values of each control amount and energy consumption are taken in according to the flow of FIG. Next, in S5-2, the fitness of the predicted value is calculated using the membership function. It shows that the desired control result is realized as the degree of conformity is larger.

FIG. 6 shows an example in which the fitness for the predicted value of VI is calculated using the membership function. Assuming that the predicted VI value is 37% and the shape of the figure is assumed as the membership function (satisfaction function), the fitness is 0.4 by the operation shown in the figure. By the same operation, it is possible to obtain the degree of adaptation such as VI value, AV value, energy consumption and the like. Finally, in S5-3, the overall satisfaction degree Wj of each operation plan j is calculated. The total satisfaction level Wj is calculated by, for example, Equation 7. .beta.1, .beta.2, .beta.3, .beta.4,... are weights by which the fitness of each evaluation factor is multiplied, and correspond to the importance of each evaluation factor. For example, when importance is attached to the AV value and the energy consumption, β1, β2, β6, and β7 may be relatively increased. Alternatively, only a factor having high importance may be selectively used as a target for evaluating the overall satisfaction.
Wj = β1AVI1 + β2AVI2 + β3ACO1 + β4ACO2 + β5AAV1 + β6AEJ + β7AEH + (7)
Here, AVI1: VI1 fitness, AVI2: VI2 fitness, ACO1: CO1 fitness, ACO2: CO2 fitness, AAV1: AV1 fitness, AEJ: Jet fan energy consumption fitness, AH : Conformity of exhaust fan energy consumption.

  In this way, the overall satisfaction level Wj corresponding to the driving plan can be calculated. In the same manner, the overall satisfaction level of other driving plans is calculated.

  FIG. 7 shows the processing executed by the driving method determination means. The driving method determination means 107 selects the most preferable driving plan from the result of calculating the total satisfaction level for each driving plan in S7-1. In S7-2, the operation amount according to the selected operation method is output to each device (jet fan 154, exhaust fan 155). In the present embodiment, the most desirable operation plan is selected in S7-1. However, it is also possible to perform apportioning processing on some desirable operation plans, generate new operation plans, and output them as the operation method.

  In this embodiment, the operation plan generation unit 101 uses the output of the operation method determination unit 107 to capture the current operation method. However, the current operation method may be recognized by directly capturing the outputs of the jet fan 154 and the exhaust fan 155 of the control target 150. Further, although the case of the longitudinal flow type as the tunnel ventilation method has been described as an example, the same method can be applied to other methods such as a cross flow type and a semi-cross flow type.

FIG. 8 shows the configuration of the prediction model learning means. In the present embodiment, the prediction model learning means 104 calculates the equivalent resistance area A of the vehicle used for calculating the soot amount q _VI , the carbon monoxide amount q _CO , and the fu of the equation 1 emitted by the vehicle corresponding to the equation q. An example of learning as a target will be shown.

  The prediction model learning unit 104 is activated by an activation signal from the learning activation unit 108. The learning activation unit 108 activates the prediction model learning unit 104 at a timing according to a time constant at which the learning target parameter changes. In the case of the amount of smoke and carbon monoxide emitted by a car, model learning reflecting this is performed according to the improvement of the fuel consumption of the car and the penetration rate of electric cars and hybrid cars. It is enough to start up about once. In addition, since the equivalent resistance area A mainly depends on the shape of the vehicle body, an activation frequency of about once a year can be considered because of the interval between vehicle model changes. Therefore, the activation signal can be easily generated by providing the learning activation means 108 with a timer for measuring about one year.

  The prediction model learning unit 104 includes a control result accumulation unit 807 that accumulates performance data such as the operating state of the exhaust fan and the jet fan, the amount of traffic of the car, and the corresponding VI value, CO concentration, wind direction wind speed (AV value), and the like. . At this time, changes in driving state and traffic volume are not instantly reflected as the corresponding VI value, CO concentration, and AV value. Therefore, a VI time delay adjusting unit 804, a CO time delay adjusting unit 805, and an AV delay time adjusting unit 806 for appropriately compensating for the delay time corresponding to each of them are provided. Each delay time t1, t2, t3 is a tunnel-specific parameter depending on the positional relationship between the operation end (exhaust fan, jet fan) and the detection end (VI meter, CO meter, AV meter, traffic counter). Since several approaches are known, such as a method for experimentally obtaining a time delay that maximizes the cross-correlation between signals, it may be determined in advance according to these approaches.

  The learning calculation means 808 takes in the data of the control result storage means 807 in accordance with the activation signal from the learning activation means 108, performs the following calculation, determines the necessity, and updates each parameter of the prediction model 103.

  FIG. 9 shows the configuration of the control result storage unit 807. The date and time is the timing of data capture, and is used for data retrieval when the data used for learning is limited to the date and time (for example, the latest three months). Corresponding to each date and time, smoke concentration (VI1), carbon monoxide concentration (CO1), wind speed (AV1) and the like corresponding to traffic volume and exhaust air volume are accumulated.

FIG. 10 shows an algorithm executed by the learning calculation means 808. Here, a case where the smoke amount q _VI is learned will be described as an example. The learning calculation means 808 starts execution when the prediction model learning means 104 receives the activation signal from the learning activation means 108. In S10-1, one set of data is fetched from the control result storage unit 807. Assume an initial value _{q VI} in S10-2. As an initial value, for example in the predictive model 103 to use a q _VI using currently contemplated. The wind speed is calculated in S10-3. That is, the wind speed u in the roadway is calculated by solving one equation. In S10-4, a VI value corresponding to the VI meter mounting position is calculated. VI is obtained by solving two differential equations for c. At this time, it goes without saying that the calculation is performed by substituting the assumed q _VI into q in the equation (2).

As shown in FIG. 1, since the present embodiment includes two VI meters, the estimated values VI1 * and VI2 * of the calculation result VI can be calculated. Differences ΔVI1 and ΔVI2 between VI1 and VI2, which are the actual values of VI taken out from the control result accumulating means 807 in S10-5, and the estimated values VI1 * and VI2 * of the corresponding VI are calculated. ΔE is calculated according to
ΔE = | ΔVI1 | + | ΔVI2 | (8)
In S10-6, it is determined whether or not ΔE is smaller than ΔEth. ΔEth is a reference value for determining whether or not the assumed q _VI is a sufficiently appropriate value. When q _VI is appropriate, ΔVI1 and ΔVI2 are both small values, and ΔE becomes small. At this time, if it is not smaller than ΔE _th, it is determined that q _VI is not a sufficiently appropriate value, and q _VI is updated according to Equation 9 in S10-7. a: Constant.
_{(Q VI -a) ⇒q VI ...} (9)
Then, S10-3 to S10-6 are repeated again. If ΔE is smaller than ΔE _{th, the} calculation for one set of data extracted from the control result storage unit 807 is terminated, and the process proceeds to S10-8.

In S10-8, it is determined whether or not the calculations in S10-2 to S10-7 have been completed for all data sets used for calculation from the control result storage unit 807. As a data set used for the calculation, it is conceivable that the data is selected according to the date and time, and a predetermined period (the latest six months, etc.) is selected as the extraction target. If the calculation has not been completed for all the target data sets, the process returns to S10-1. In S10-9 if finished, averaging the calculated _{q VI} for each data set, and updates the _{q VI} predictive model 103 with the average value. When the difference between the current q _VI and the current calculated q _VI is less than or equal to a predetermined value, it is conceivable to omit the update.

In this embodiment, q _VI is a single constant. However, when q _VI is stratified according to the type of vehicle or the output of the exhaust fan, etc., if q _VI is calculated for each stratification, the same procedure is used. it can. In this embodiment, the case where q _VI is learned has been described as an example. However, even in the case of q _CO , the same procedure can be used.

Further, the equivalent resistance area A of the car is a value in the formula 12 to the formula 12 in which f (u) in the formula 1 is described in detail. Can be calculated in the same way.
f (u) = Pr + Pb1 (10)
However, Pr: Roadway resistance, Pb: Ventilation force by an exhaust fan and a jet fan.
Pr = Pin + Pt + PM (11)
However, Pin: entrance loss, Pt: traffic ventilation, PM: natural ventilation.
Pt = (A / Ar) · (ρ / 2) {n · (Vt−Vr) ² } (12)
However, A: Automobile equivalent resistance area, Ar: Tunnel cross-sectional area, ρ: Air density, n: Number of vehicles in tunnel, Vt: Vehicle speed, Vr: Wind speed in tunnel.

In FIG. 10, q _VI is calculated by repeated calculation of S10-3 to S10-7. However, various calculation methods for obtaining the result are conceivable, such as a method of directly solving for q by transforming Formula 2 into Formula 13.
q = (∂c / ∂t) = - u (∂c / ∂x) + D (∂ 2 c / ∂x 2) ... (13)
Further, although the prediction model learning unit 104 is automatically activated from the learning activation unit 108, a method in which an administrator of the control device 100 manually activates the method is also conceivable.

  FIG. 11 qualitatively shows the energy saving effect when this embodiment is realized, taking as an example the case of learning the amount of smoke emitted by the car. The horizontal axis shows the year, and the vertical axis shows the average fuel consumption and power consumption of the car. The fuel economy improves year by year due to an increase in the ratio of vehicles with good fuel consumption, the spread of electric vehicles, etc. in addition to the improvement in the fuel economy of the vehicle itself (1 in FIG. 11). On the other hand, at the time of trial operation adjustment, when the exhaust fan 155 and the jet fan 154 are operated using the smoke amount discharged by the vehicle at that time as a model and the operation according to the same standard is continued, the power consumption is maintained ( 2) in FIG. On the other hand, when the fuel efficiency is improved, the amount of smoke discharged from the vehicle is reduced, so that the power consumption necessary for obtaining the same ventilation performance is reduced (3 in FIG. 11). The shaded area shown in the figure is the power consumption that is excessively consumed. When this embodiment is applied, the power consumption can be expected to decrease as the vehicle fuel efficiency improves, as indicated by 3 in FIG.

  In the present embodiment, the tunnel ventilation control has been described as an example. As a method for improving and maintaining control accuracy by mixing adaptive control and learning control, the same concept can be applied to other control devices as long as predictive control using a control model is effective.

FIG. 12 shows an example of a parameter adjustment service method for a tunnel ventilation control apparatus that monitors the control performance of the control apparatus 100 and adjusts parameters (q _VI , q _CO , A in this embodiment) of the prediction model 103 as necessary. Indicates.

In this embodiment, an example in which the quality of control is remotely monitored from the service center 1200 using the Internet or the like, and q _VI , q _CO , and A are similarly remotely adjusted using the Internet is shown. Various types of services are conceivable, such as a method of loading parameters directly to the control device 100.

  In this embodiment, the control device 100 includes a result collection unit 1210, a result storage unit 1211 that stores the collected results, and a result transmission unit 1212 that transmits the contents of the result storage unit 1211 to the service center 1200 via the public line network 1220. ing. In addition, the service center 1200 is provided with an information database 1204 that stores the control status of other tunnel ventilation control, and trends in the fuel consumption of vehicles and the diffusion rate of low fuel vehicles announced by automobile companies and the like. Furthermore, a performance receiving means 1201 for receiving the results transmitted from the information database control apparatus 100, a performance evaluation means 1202 for evaluating whether the control state is good and whether the parameter adjustment from the service center 1200 is working well is provided. . A parameter for determining the necessity of parameter adjustment from the evaluation result of the result evaluation unit 1202 and the result of referring to the information database 1204, and, if necessary, fetching the result necessary for adjustment from the result reception unit 1201 and calculating the parameter value Adjustment means 1203 is provided. Furthermore, parameter transmission means 1205 for transmitting parameters to the control apparatus 100 via the public line network 1220 and parameter adjustment frequency count means 1206 for counting the number of times the parameter transmission means 1205 has transmitted the parameters are provided.

  First, the operation of the control device 100 will be described. The performance collecting means 1210 collects the control performance made up of the combination of the operation method (exhaust air flow rate, jet fan operation number, etc.) and the corresponding control performance (VI value, CO concentration, etc.) and accumulates it in the performance storage means 1211. In addition to this, the amount of smoke and carbon monoxide leaking out from the mouth can also be considered as control results. The actual result transmission unit 1212 transmits these to the service center 1200, but the transmission may be performed at a timing when a new control result is accumulated in the actual result accumulation unit 1211 or after a certain amount is accumulated.

Next, the operation of the service center 1200 will be described. The performance receiving means 1201 receives and accumulates the transmitted results, and the performance evaluation means 1202 monitors the control status based on these results and determines whether the control results are acceptable. The quality of the control result can be easily determined by calculating a value W that combines the satisfaction of the control by performing the same calculation as shown in Equation 7, and evaluating the time series transition of W. Specifically, the equation 14 may be calculated and evaluated by the fact that this size has not decreased since the initial adjustment.
W = β1AVI1 + β2AVI2 + β3ACO1 + β4ACO2 + β5AAV1 + β6AEJ + β7AEH + (14)
FIG. 13 shows processing executed by the parameter adjusting means. The parameter adjustment unit 1203 determines the necessity of parameter adjustment in S13-1. Necessity is the presence or absence of a decrease in the overall satisfaction level W evaluated by the performance evaluation means 1202, the relative performance comparison result with other ventilation control devices obtained from the information database 1204, the vehicle fuel consumption and the ratio of low fuel vehicles. It depends on whether or not it has changed most recently. The necessity of control parameter adjustment currently used in the control apparatus 100 is determined from the results of various determinations. If the determination result is “adjustment required” in S13-2, parameter calculation is performed in S13-3. The model parameters may be calculated by the method shown in the first embodiment. The parameter calculated in S13-4 is transmitted to the control device 100. The parameter adjustment count counting means 1206 counts and stores the parameter adjustment count.

  The control device manager is charged for the work of the service center 1200 for at least one of the monitoring performance of control results, the number of parameter adjustments counted by the parameter adjustment number counting means 1206, and the control performance evaluated by the performance evaluation means 1202. Done.

The block diagram by one Example of the tunnel ventilation control apparatus of this invention. The flowchart which shows the process of a driving plan production | generation means. The flowchart which shows the process of a prediction model calculating means. The block diagram of the adaptive control means by one Example. The flowchart which shows the process of a driving plan evaluation means. Explanatory drawing of the method of evaluating a fitness using a membership function. The flowchart which shows the process of a driving | operation system determination means. The block diagram of the prediction model learning means by one Example. The data block diagram of a control result storage means. The flowchart which shows the process of a learning calculating means. The schematic diagram which shows the effect of this invention. The block diagram which shows the parameter adjustment system of the tunnel ventilation control apparatus by this invention. The flowchart which shows the process of a parameter adjustment means.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 100 ... Control apparatus, 101 ... Driving plan production | generation means, 102 ... Prediction model calculation means, 103 ... Prediction model, 104 ... Adaptive control means, 105 ... Prediction model learning means, 106 ... Driving plan evaluation means, 107 ... Driving method determination means , 150 ... control target, 401 ... observer, 402 ... estimated trend generation means, 404 ... performance trend generation means, 406 ... error series calculation means, 407 ... prediction error estimation means, 807 ... control result storage means, 808 ... learning calculation means DESCRIPTION OF SYMBOLS 1200 ... Service center, 1202 ... Performance evaluation means, 1203 ... Parameter adjustment means, 1204 ... Information database, 1206 ... Parameter adjustment frequency count means, 1220 ... Public line network.

Claims (10)

Prediction model describing behavior of smoke concentration and carbon monoxide concentration in tunnel, prediction model calculation means for predicting future smoke concentration and carbon monoxide concentration using the prediction model, prediction of the prediction model calculation means In the tunnel ventilation control device equipped with the operation method determining means for determining the desired operation method of the jet fan and the exhaust fan using the result,
Using the actual smoke and carbon monoxide concentrations detected from the tunnel and the operation results of jet fans and exhaust fans, the prediction model has the vehicle smoke emissions, monoxide concentration emissions, and equivalent vehicle resistance. A tunnel ventilation control apparatus comprising: a prediction model learning unit that reversely calculates at least one of the areas and updates the parameters of the prediction model according to a reverse calculation result. Prediction model describing behavior of smoke concentration and carbon monoxide concentration in tunnel, prediction model calculation means for predicting future smoke concentration and carbon monoxide concentration using the prediction model, prediction of the prediction model calculation means In the tunnel ventilation control device equipped with the operation method determining means for determining the desired operation method of the jet fan and the exhaust fan using the result,
Calculate the size of the error included in the prediction result of the prediction model calculation means that performs the calculation using the prediction model from the actual value of the smoke concentration and carbon monoxide concentration detected from the tunnel and the operation result of the jet fan or the exhaust fan And adaptive control means used for control by compensating the output of the prediction model calculation means using the calculated prediction error;
Using the actual value and the driving performance, the prediction model has at least one of the smoke emission, monoxide concentration emission, and the equivalent resistance area of the vehicle that the prediction model has, and the prediction model according to the reverse calculation result A tunnel ventilation control device comprising a predictive model learning means for updating the parameters of the above. In claim 2,
A tunnel ventilation control device, wherein the learning control of the prediction model learning means and the adaptive control of the adaptive control means are performed in parallel. In claim 2 or 3,
A timing for determining a driving method of the jet fan or the exhaust fan, comprising learning starting means for managing activation of the prediction model learning means, executing the prediction model learning means in response to an activation command of the learning activation means; The tunnel ventilation control device is characterized in that the prediction model learning means and the adaptive control means are executed asynchronously. Predict the future smoke concentration and carbon monoxide concentration using a prediction model that describes the behavior of the smoke concentration and carbon monoxide concentration in the tunnel, and use these prediction results to determine the desired jet fan and exhaust fan operation method In a tunnel ventilation control method for performing adaptive control,
Using the actual smoke and carbon monoxide concentrations detected from the tunnel and the operation results of jet fans and exhaust fans, the prediction model has the vehicle smoke emissions, monoxide concentration emissions, and equivalent vehicle resistance. A tunnel ventilation control method, characterized in that learning control is performed in parallel with the adaptive control, in which at least one of the areas is reversely calculated and the parameters of the prediction model are updated according to the reverse calculation result. In claim 5,
The adaptive control calculates an error of the prediction result from the actual value of the smoke concentration and the carbon monoxide concentration detected from the tunnel and the operation result of the jet fan or the exhaust fan, and a desired jet according to the result of compensating the prediction value by the error. A tunnel ventilation control method characterized by determining a method of operating a fan or a fan. In claim 5 or 6,
The tunnel ventilation control method characterized in that the adaptive control is executed at a timing for determining an operation method of the jet fan or the exhaust fan, and the learning control is executed at a timing independent of the adaptive control for learning of the prediction model. . Predict the future smoke concentration and carbon monoxide concentration using a prediction model that describes the behavior of the smoke concentration and carbon monoxide concentration in the tunnel, and use these prediction results to determine the desired jet fan and exhaust fan operation method In the tunnel ventilation control device parameter adjustment service method
The service center receives the actual value of smoke concentration and carbon monoxide concentration detected from the tunnel and the operation result of the jet fan and the exhaust fan, and the prediction model has the vehicle smoke emission, monoxide concentration emission, Back-calculate at least one of the equivalent resistance area of the car, update the parameter of the prediction model using this back-calculation result, and send it to the tunnel ventilation control device,
Thereby, the parameter adjustment service method of the tunnel ventilation control device, wherein the parameter of the prediction model of the tunnel ventilation control device is updated. Predict the future smoke concentration and carbon monoxide concentration using a prediction model that describes the behavior of the smoke concentration and carbon monoxide concentration in the tunnel, and use these prediction results to determine the desired jet fan and exhaust fan operation method In the tunnel ventilation control device parameter adjustment service method
The service center receives the actual smoke smoke and carbon monoxide concentration values detected from the tunnel and the operation results of the jet fan and the exhaust fan, and the prediction model has the vehicle smoke emission, monoxide concentration emission, vehicle And reversely calculating at least one of the equivalent resistance areas, and updating the parameters of the prediction model using the result of the reverse calculation and transmitting the parameters to the tunnel ventilation control device,
A parameter adjustment service method for a tunnel ventilation control device, characterized in that a predetermined value contracted in advance by a tunnel administrator is obtained in accordance with at least one of the number of parameter updates and the degree of control improvement by the updated parameters. Predict the future smoke concentration and carbon monoxide concentration using a prediction model that describes the behavior of the smoke concentration and carbon monoxide concentration in the tunnel, and use these prediction results to determine the desired jet fan and exhaust fan operation method In the tunnel ventilation control device parameter adjustment service method
The service center receives the control values such as the smoke concentration and carbon monoxide concentration detected from the tunnel, the energy consumption associated with the operation of the jet fan and the exhaust fan, and the amount of smoke leaking from the tunnel entrance. The necessity of updating the parameter of the prediction model is determined from the value and the control state, and if it is determined that the prediction model is necessary, the smoke emission amount of the vehicle, the monoxide concentration emission amount, the vehicle equivalent possessed by the prediction model Back-calculate at least one of the resistance areas, update the parameters of the prediction model using the back-calculation result,
A tunnel ventilation control device that obtains a predetermined value contracted in advance from a tunnel administrator in accordance with at least one of the control status monitoring operation, the number of times of updating of the parameter, and the degree of improvement of control by the updated parameter Parameter adjustment service method.

Images