JP2847468B2 - Waste Property Estimation Method and Estimation Device for Waste Incinerator - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for estimating refuse properties of a refuse incinerator. More specifically, the present invention relates to a method and an apparatus for estimating the waste property of a waste incinerator, which can estimate the waste property during the incineration of the waste.

[0002]

2. Description of the Related Art In order to accurately control combustion in a refuse incinerator, it is necessary to accurately grasp the properties of the refuse such as the amount of refuse supplied, the amount of heat generated, and the amount of stagnation during combustion. In order to achieve such an object, various proposals have conventionally been made. For example, Japanese Unexamined Patent Publication No. 5-272730 discloses that a unit waste amount is accurately set and incinerated by deriving a unit waste amount as an average value per unit time of the measured waste amount immediately after the set number of units of the waste injecting unit is operated. For the purpose of incinerating the amount, the combustion control means stores the charging time, the weight of the refuse at that time, and the value of the refuse sensor after the refuse in the storage means as needed, and stores the time-average refuse at approximately 40-60 minute intervals. Using the derivation of the quantity as a guide, the number of injections of four times is set as the set number of injections, and then the time from the next injection of the previous time average value derivation example to the time when the number of injections of the fourth time is counted is derived. Therefore, a time-averaged waste input amount is derived from the total waste amount input during that period and the time in between, so that the amount of waste input to the hopper per unit time can be accurately obtained. Incinerator has been proposed.

[0003] However, there is a time delay between the hopper input amount and the furnace input amount, and some of the input refuse remains in the hopper as it is. I can't say. Therefore, it is difficult to say that garbage can be properly incinerated depending on the incinerator related to the proposal.

[0004] Further, in a refuse incinerator, various miscellaneous refuse are randomly injected and incinerated, so that the properties of the refuse to be incinerated are not constant. Therefore, there is a problem that the combustion control cannot be accurately performed in the refuse incinerator only by estimating the refuse input amount as in the above proposal.

[0005]

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems of the prior art, and is intended for a waste incinerator into which various miscellaneous wastes are randomly charged, in which the amount of waste supplied, the amount of heat generated, and the amount of accumulated A main object of the present invention is to provide a method and an apparatus for estimating a waste property in a waste incinerator capable of accurately estimating the property of the waste such as an amount, and further, a combustion control method and a control apparatus for the waste incinerator using the estimation method and the apparatus. It is also intended to provide.

[0006]

According to the present invention, there is provided a method for estimating a waste property of a waste incinerator, comprising: a teaching data generating means; a dynamic characteristic simulating means having a dynamic characteristic model part and a compensation part comprising a neural network; A method for estimating waste properties of a waste incinerator using a waste property estimating means, wherein a procedure for adjusting parameters of a dynamic characteristic model, a procedure for causing a neural network of a compensator to learn, and a neural network for the waste property estimating means are provided. A procedure for performing learning, a procedure for estimating the waste property using data other than the data used for the learning, and a procedure for inputting the estimated waste property to the dynamic characteristic simulation means to obtain a compensated process state quantity. And a procedure for verifying the obtained process state quantity with actual machine data.

Here, the process amount is, for example, a furnace outlet gas temperature, a main steam flow rate or an O ₂ concentration.

In the method for estimating the refuse property of a refuse incinerator according to the present invention, for example, learning of the neural network of the compensator is performed using actual machine data, and learning of the neural network of the refuse property estimating means is generation of teaching data. This is done using the M-sequence signal from the means.

Here, the neural network includes:
An input layer, at least one intermediate layer, and an output layer, wherein the input layer, the intermediate layer, and the output layer include a plurality of processing units. It is connected by a link having a determined weight value and has a threshold function having a threshold value determined by learning.

On the other hand, a waste property estimating device for a waste incinerator according to the present invention comprises a teaching data generating means, a dynamic characteristic simulating means having a dynamic characteristic model part and a compensating part comprising a neural network, and a waste characteristic estimating means comprising a neural network. It is characterized by comprising means.

In the waste property estimation device for a waste incinerator according to the present invention, it is preferable that the teaching data generating means is an M-sequence generator.

Here, the neural network includes:
An input layer, at least one intermediate layer, and an output layer, wherein the input layer, the intermediate layer, and the output layer include a plurality of processing units. It is connected by a link having a determined weight value and has a threshold function having a threshold value determined by learning.

[0013]

In the present invention, the parameters of the dynamic characteristic model are adjusted first, and then the neural network of the compensator is learned to compensate for the deficiencies of the dynamic characteristic model. Thereafter, learning of the neural network of the waste property estimating means is performed using the output of the neural network after the learning. On the other hand, the waste property is estimated using the actual machine data. Next, the process state quantity is calculated by the dynamic characteristic simulation means using the obtained estimated value of the dust property. The calculated process state quantity is compared with the actual machine data, and the estimated value from the waste property estimation means is verified. As a result, when it is determined that the operation is appropriate, the operation is started.

Here, since the learning of the neural network of the waste property estimating means is performed using the data from the dynamic characteristic simulation means, the number of times of learning can be drastically increased. An estimate can be obtained.

[0015]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described based on embodiments with reference to the accompanying drawings, but the present invention is not limited to only such embodiments.

Here, in order to facilitate understanding of the following description, first, a neural network will be briefly described.

FIG. 1 shows the basic structure of a neural network, which is composed of an input layer, an intermediate layer, and an output layer. Each layer is composed of a plurality of processing units, and each processing unit is connected between the next layer and the previous layer with a weight coefficient determined by learning.
Each processing unit has an adder for summing inputs from the previous layer and a threshold function having a threshold determined by learning. In learning of this neural network, a pair of an input pattern (teacher input) and a target output pattern (teacher output) is presented. Immediately after this presentation, the weights and thresholds are adjusted so that the difference between the output of the network and the target output is reduced. At the time of learning, a learning set which is a set of pairs of an input pattern and a target pattern is used, and this is repeatedly presented to the network. When the learning is completed, a network operation test is performed.

In this learning, there are a forward propagation step and a backward propagation step executed thereafter. Both the forward propagation step and the back propagation step are executed each time a pattern is presented during learning. The forward propagation step begins with the presentation of the input pattern to the input layer of the network and continues as the activity level calculation propagates forward through the hidden layer. All the processing units of each layer (indicated by a mark in FIG. 1) calculate the sum of the inputs and calculate the output by a threshold function. The output layer of the unit then outputs the network.

A comparison is made between the output pattern of the network and the target output pattern, and when there is a difference, a backpropagation step is started. In the backpropagation step, the change of the threshold value and the weight of the unit of each layer is calculated, and the calculation is started from the output layer and sequentially traced in the reverse direction to the intermediate layer. In this backpropagation step, the network is adjusted for weights and thresholds so that the observed difference is reduced.

Since such learning is performed, the output pattern from the network basically basically matches the target output pattern.

Note that, specifically, this neural network is configured by storing a program corresponding to the arithmetic processing in a computer.

FIG. 2 shows a block diagram of an estimating apparatus having such a neural network and used in the waste property estimating method of the present invention. This estimating device includes, as main components, a dynamic characteristic simulating means 2, a dust property estimating means 3, and a teaching data generating means 1. When the pusher speed, the air flow rate, the stoker speed, and the load steam valve opening are input to the dynamic characteristic simulating means 2, for example, the waste property estimating means 3 outputs, for example, a waste supply amount, a waste heat amount, and a waste retention amount. An estimated value is output.

The dynamic characteristic simulating means 2, as shown in FIG.
The main part is a dynamic characteristic model unit 21 and a compensating unit 22 configured by a neural network.

The dynamic characteristic model unit 21 is based on energy balance, momentum balance, mass balance, combustion reaction formula, chemical equilibrium formula, etc., based on energy balance, momentum balance, chemical balance formula, etc., and also takes into account heat storage capacity and side capacity. A mathematical model consisting of a group of nonlinear simultaneous differential equations, that is, a model having a function of analyzing a dynamic characteristic model and outputting plant characteristics, and specifically, for example, is configured by storing a program corresponding to a computer. Is done. In this embodiment, when the pusher speed, the load steam valve opening, the air flow rate and the stoker speed are input, the process state quantity,
For example, it is configured to output the incinerator outlet gas temperature, the main steam flow rate, and the O ₂ concentration.

The compensating unit 22 is composed of a neural network, and has a function of using a learning function of the neural network to compensate for a point where the expression by the mathematical model is insufficient and a point where the parameter adjustment of the dynamic characteristic model is insufficient. Things. That is, in a refuse incinerator, various types of refuse are randomly put into the incinerator and incinerated.Therefore, there is a part that cannot be mathematically modeled because a strict analysis cannot be performed. As always in modeling, a mathematical model is created by simplifying and approximating the phenomenon. Therefore, when the dynamic characteristics simulation (hereinafter, also referred to as simulation) is performed using such a mathematical model, an error due to the limit of the mathematical model exists in the output. In addition, parameter adjustment is required when operating the dynamic characteristic model, but in a garbage incineration plant that needs to be treated as a distribution system, there are thousands to thousands of parameters, so all these parameters are optimally adjusted. That is virtually impossible. Therefore, an error due to the limit of the parameter adjustment also exists in the output. Therefore, these errors are compensated for by the compensator 22.

Here, the input layer of the neural network is provided with the gas temperature at the incinerator outlet,
The main steam flow rate and the O ₂ concentration are branched and input in chronological order (see FIG. 4), and the pusher speed, the load steam valve opening, the air flow rate, and the stoker speed are also input. Then, from the output layer, the process state quantity from the dynamic characteristic model unit 21, for example, the incinerator outlet gas temperature, the main steam flow rate, and the compensation amount of the O ₂ concentration are output. After the output from the dynamic characteristic model is compensated by this compensation amount, the dynamic characteristic simulating means 2
Is output as a process state quantity. 5 and 6
The results of simulating a waste incinerator (plant) with this configuration are shown for reference. FIG. 5 shows a case where learning is repeated 100 times, and FIG. 6 shows a case where learning is repeated 1000 times. For comparison, FIG. 7 shows a simulation result using only the dynamic characteristic model unit 21 without using the compensation neural network.

The waste property estimating means 3 is composed of a neural network, and the pusher speed and the stalker speed are input to the input layer in time series, and the respective compensation amounts from the compensator 22 are also branched and input in time series. Is done. Then, an estimated value of the waste supply amount, the waste heat generation amount, and the waste retention amount is output from the output layer.

Here, the input of the waste property estimating means 3 is configured as described above for the following reason.

Generally speaking, in order to simplify the configuration of the refuse property estimating apparatus, in the present embodiment, it is preferable that the dynamic characteristic model unit 21 and the compensating unit 22 provided in the preceding stage are not provided. . However, in order for the generalization function (interpolation function) of the neural network after learning to function effectively, it is necessary that a certain correlation exists between its input and output. Under such circumstances, if the device is constituted only by the waste property estimation means 3, the following problem occurs. That is, in the learning of the neural network of the refuse property estimation means 3, the operation amount,
For example, in addition to the pusher speed and the stoker speed, the state quantities of the actual machine, such as the furnace outlet gas temperature, the main steam flow rate, and the O
₂ Concentration is required, and waste property data such as waste supply amount, waste heat generation amount, and waste retention amount are required as output side teaching data. Incidentally, the input-side teaching data can be easily obtained because it is data obtained by actually operating the plant. On the other hand, the output-side teaching data cannot be obtained because it cannot be measured.

Therefore, if the dynamic characteristic simulating means 2 in which the dynamic characteristic model unit 21 and the compensating unit 22 composed of a compensation neural network are combined is used as in this embodiment, after learning, the input-output The correlation can be accurately simulated as shown in FIG. 6, and the uncertainty of the dust property is compensated for because the steam system and the ventilation system constituting the dynamic characteristic model unit 21 can be sufficiently simulated by mathematical expressions. Can be aggregated into a neural network. for that reason,
If the output of the compensation neural network is branched and used as the input-side teaching data of the estimating neural network, it can be correlated with the output-side teaching data. Further, if a dynamic characteristic model is used, input / output data for teaching can be obtained by dynamic simulation without operating the incinerator. Also, a large number can be obtained. This is convenient for learning a neural network used in a refuse incinerator in which miscellaneous garbage is randomly incinerated. Moreover, if the input to the dynamic characteristic model is made at random, the neural network can be made to learn in a state closer to the operating state of the refuse incinerator. Therefore, in this embodiment, an M-sequence generator that outputs teaching data randomly is used as the teaching data generating means 1.

Here, just in case, the configuration and function of this M-sequence generator will be briefly described.

The M-sequence is a kind of pseudorandom number generated by an algorithm using a computer, and uses a higher-order recurrence formula. For example, in recurrence formula 3 paragraphs, X _n = X _np plusX _nq the (p> q) used. Here, plus represents exclusive OR (addition without carry in binary integer representation). For example, n = 1
In the example of 1, p = 9, and q = 11, the shift register circuit shown in FIG. 8 can generate an M-sequence signal (time series of 0, 1) having a period N = 2 ¹¹ -1 = 2047. Furthermore, a sequence obtained by sampling the above-described M sequence with an interval prime to N is also an M sequence with a period N, so that a multiplexed M sequence group can be obtained.

As described above, the teaching data generator 1 is, for example, an M-sequence generator that generates teaching data randomly and multiplexed. Therefore, from the teaching data generating means 1, the pusher speed, the load steam valve opening, the air flow rate, and the stoker speed can be randomly input to the dynamic characteristic model unit 21 as basic data for creating input-side teaching data. Further, the amount of waste and the amount of heat generated can be randomly input to the input / output side as input / output side teaching data. It should be noted that the dust retention amount calculated based on the input is used as the dust retention amount of the output-side teaching data.

Next, a description will be given of a procedure for estimating the waste property by the waste property estimating apparatus having the above-described configuration.

Step 1: The parameters of the dynamic characteristic model are adjusted. This may be a trial and error method as in the past, or may be performed using a genetic algorithm related to the previous proposal of the present inventors (T. IEEE Japan V.
ol. 113-D, no. 12, '93, 1410-1
(Page 415)

Step 2: A learning is performed by adding a compensation neural network to the dynamic characteristic model whose parameters have been adjusted in step 1. In this case, actual machine data is used as the teaching data.

Step 3: Learning is performed by adding an estimation neural network to the dynamic characteristic model and the compensation neural network learned in step 2. In this case, the data from the teaching data generating means (M-sequence generator) 1 is used as the teaching data, and the data used in steps 1 and 2 is used as the dynamic characteristic model input data.

Step 4: Only the dynamic characteristic model of which learning has been completed in steps 1, 2 and 3 and the dynamic characteristic model input data of the actual machine data other than the teaching data are input to the group of neural networks for compensation, and the dust property is estimated. .

Step 5: The estimated refuse property is input to the dynamic characteristic model to obtain a process state quantity, which is compensated by a compensation neural network and output as a process state quantity.

Step 6: The output process state quantity is verified with the actual machine data.

Step 7: If it is determined in the verification that the estimated value of the waste property is not appropriate, the process returns to step 1 and repeats the above procedure. On the other hand, when it is determined that the estimated value of the waste property is appropriate, the adjustment operation ends.

After the adjustment has been made in this way, the teaching data generating means (M-sequence generator) 1 is disconnected to start the practical operation. In this practical operation, for example, automatic combustion control of the incinerator is performed using the estimated value.

As described above, according to the waste property estimation method and the waste estimation apparatus, it is possible to accurately estimate the amount of waste supplied, the amount of generated heat, and the amount of retained waste.

Hereinafter, automatic combustion control of a refuse incinerator using the estimated value will be described.

FIG. 9 shows a block diagram of an automatic combustion control device used for automatic combustion control of this refuse incinerator. This automatic combustion control device has, as main components, an automatic combustion control device main body (hereinafter, simply referred to as a main body), a waste property estimating device, a stay amount calculating means, a compensation signal generating means, and a compensating means.

The configuration of the main body is conventionally known in that when a set value of a refuse supply amount and a set value of a refuse heat amount are input, manipulated variables such as a pusher speed, a stalker speed, and an air flow rate are output. .

The staying amount calculating means calculates the staying amount of waste by a conventional formula based on the set value of the amount of waste supplied. More specifically, the computer stores a program corresponding to the above formula in a computer. Be composed.

The compensation signal generating means performs an arithmetic operation on the branch signal of the waste supply set value signal input to the main body and the waste supply estimated value from the waste property estimating device to compensate the pusher speed set value. Compensates the air flow rate set value by processing the pusher speed set value compensation signal, the branch signal of the waste heat value set value signal input to the main unit, and the waste heat value estimation value from the waste property estimation device. The accumulated amount calculation value calculated by the accumulated amount calculation means based on the air flow set value compensation signal for performing the operation and the branch signal from the waste supply amount set value is calculated with the estimated dust accumulation amount value from the dust property estimation device. The processing is performed to output a stalker speed set value compensation signal for compensating the stalker speed set value to the compensation circuit. Specifically, it is configured by storing a program corresponding to the arithmetic processing in a computer.

The compensating means calculates the pusher speed set value, the stoker speed set value, and the air flow rate set value which are compensated by arithmetically processing each compensation signal from the compensation signal generating means and each set value of the waste incinerator. It is output to each device to adjust the pusher speed and the like. This is also specifically configured by storing a program corresponding to the arithmetic processing in a computer.

Next, a description will be given of the combustion control by the automatic combustion control device thus configured.

Step 1: A set value of a refuse supply amount and a refuse heat value is input to the main body by input means (not shown).

Step 2: The main body outputs the set value of the pusher speed, the set value of the stalker speed, and the set value of the air flow rate corresponding to the set values of the supplied amount of waste and the generated amount of waste heat.

Step 3: The pusher speed, the air flow rate, the stoker speed, and the load steam valve opening are fed back from the waste incinerator to the waste property estimation device.

Step 4: The refuse property estimating device performs the refuse supply estimated value, the refuse heat value estimated value, and the refuse accumulation amount estimated value based on the fed-back pusher speed, air flow rate, stoker speed, and load steam valve opening. Is output.

Step 5: Each estimated amount output from the waste property estimation device is input to the compensation signal generating means.

Step 6: The compensation signal generating means performs the above processing and outputs a pusher speed set value compensation signal, a stalker speed set value compensation signal and an air flow rate set value compensation signal to the compensation means.

Step 7: The compensating means performs the above processing,
The compensated pusher speed set value, stoker speed set value and air flow rate set value are output to the refuse incinerator.

Step 8: The pusher speed and the like are corrected to the compensated set values, and the incineration of the refuse is performed.

As described above, according to the present automatic combustion control device, the pusher speed set value, the stoker speed, and the waste supply amount estimation value, the waste heat generation amount estimation value, and the waste accumulation amount estimation value during combustion with high accuracy are used. Since the set value and the air flow set value are adjusted, it is possible to maintain optimal combustion with low pollution.

As described above, the present invention has been described based on the simulation examples. However, the present invention is not limited to only such examples, and various modifications are possible. For example, depending on the input / output correlation, a dynamic characteristic model can be configured by connecting the dynamic characteristic model and the compensation neural network in series.

[0061]

As described above in detail, according to the dust property estimation method and the estimation apparatus of the present invention, the neural network learning of the dust property estimation means is performed by using the output from the compensation unit of the dynamic characteristic simulation means. As a result, an excellent effect of accurately estimating the dust property can be obtained.

Further, if the combustion control of the refuse incinerator is performed by using the estimated value, it is possible to perform the combustion control matching the actual operation state, and to obtain an excellent effect that the emission of harmful substances can be suppressed as much as possible.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram of a neural network.

FIG. 2 is a block diagram of a waste property estimation device used in the waste property estimation method for a waste incinerator according to the present invention.

FIG. 3 is a schematic configuration diagram of the garbage property estimation device.

FIG. 4 is an explanatory diagram showing a relationship between an input and an output in a neural network.

FIG. 5 is a graph showing a simulation result of a refuse incinerator after learning 100 times by a neural network of a compensating unit of the dynamic characteristic simulation means.

FIG. 6 is a graph showing a simulation result of a refuse incinerator after a neural network of a compensating unit of the dynamic characteristic simulating means is trained 1000 times.

FIG. 7 is a graph showing a simulation result of a refuse incinerator using only a dynamic characteristic model part of dynamic characteristic simulation means.

FIG. 8 is a block diagram of a shift register used in the M-sequence generator.

FIG. 9 is a block diagram of an automatic combustion control device used for automatic combustion control of a refuse incinerator.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Teaching data generation means 2 Dynamic characteristic simulation means 21 Dynamic characteristic model part 22 Compensation part 3 Garbage property estimation means

Continued on the front page (72) Inventor Kenji Yuasa 1-3-1 Higashikawasakicho, Chuo-ku, Kobe Kawasaki Heavy Industries, Ltd. Kobe Head Office (72) Inventor Susumu Hashidera 1-1, Kawasakicho, Akashi-shi Kawasaki Heavy Industries, Ltd. Inside the Akashi Factory (72) Inventor Yoshinobu Mori 1-1-1, Kawasaki-cho, Akashi-shi Kawasaki Heavy Industries, Ltd. Inside the Akashi Factory (72) Inventor Satoshi Kurosaka 1-3-1, Higashi-Kawasaki-cho, Chuo-ku, Kobe Kawasaki Heavy Industries, Ltd. Kobemoto In-house (56) References JP-A-6-34118 (JP, A) JP-A-5-279980 (JP, A) JP-A-4-161710 (JP, A) JP-A-4-208306 (JP, A) JP-A-62-255717 (JP, A) JP-A-5-87322 (JP, A) JP-A-6-117619 (JP, A) (58) Fields investigated (Int. Cl. ⁶ , DB name) F23G 5/50 G05B 17/02 G06F 15/18 G05B 13/02

Claims (14)

(57) [Claims] 1. A method for estimating a waste property of a waste incinerator using a teaching data generating means, a dynamic characteristic simulating means having a dynamic characteristic model part and a compensation part comprising a neural network, and a waste characteristic estimating means comprising a neural network. The procedure for adjusting the parameters of the dynamic characteristic model, the procedure for making the neural network of the compensator learn, the procedure for making the neural network of the garbage property estimating means learn, and using data other than those used for the learning. Includes a procedure for estimating the waste property, a procedure for inputting the estimated waste property to the dynamic characteristic simulation means to obtain a compensated process state quantity, and a procedure for verifying the obtained process state quantity with actual machine data. A method for estimating waste properties of a waste incinerator, characterized in that: 2. The method for estimating waste properties of a waste incinerator according to claim 1, wherein said process amount is a furnace outlet gas temperature, a main steam flow rate, an O ₂ concentration or the like. 3. The method for estimating the refuse property of a refuse incinerator according to claim 1, wherein the learning of the neural network by the compensation unit is performed using actual machine data. 4. The learning of the neural network of the waste property estimating means is performed using data created by inputting an M-sequence signal from the teaching data generating means to the dynamic characteristic simulating means. The method for estimating the waste properties of a waste incinerator according to claim 1 or 2. 5. The neural network includes an input layer, at least one intermediate layer, and an output layer, wherein the input layer, the intermediate layer, and the output layer include a plurality of processing units. 3. The method according to claim 1, wherein the first and second layers are connected by a link having a weight value determined by learning, and have a threshold function having a threshold value determined by learning. 3. The method for estimating waste properties of a waste incinerator according to 3 or 4. 6. A combustion control of a refuse incinerator, wherein the combustion control of the refuse incinerator is performed by using an estimated value obtained by the refuse property estimation method according to claim 1, 2, 3, 4, or 5. Method. 7. The combustion control method for a refuse incinerator according to claim 6, wherein the operation amount of the combustion control device is compensated by the estimated value. 8. The method according to claim 1, wherein the operation amount is a pusher speed set value,
8. The combustion control method for a refuse incinerator according to claim 7, wherein the stoker speed setting value, the air flow rate setting value, and the like are used. 9. A refuse incinerator comprising teaching data generating means, dynamic characteristic simulating means having a dynamic characteristic model part and a compensation part comprising a neural network, and refuse property estimating means comprising a neural network. Waste property estimation device. 10. The waste property estimating apparatus for a waste incinerator according to claim 9, wherein said teaching data generating means is an M-sequence generator. 11. The neural network includes an input layer, at least one intermediate layer, and an output layer, wherein the input layer, the intermediate layer, and the output layer include a plurality of processing units, and each of the processing units includes: 10. A threshold function having a threshold value determined by learning and connected by a link having a weight value determined by learning between the preceding layer and the next layer.
A waste property estimation device for a waste incinerator according to item 0. 12. A combustion control apparatus for a refuse incinerator, comprising the refuse property estimation apparatus according to claim 9, 10 or 11. 13. The combustion control device for a refuse incinerator according to claim 12, further comprising compensating means for compensating an operation amount based on an estimated value from the refuse property estimating device. 14. The combustion control device for a refuse incinerator according to claim 12, wherein the manipulated variables are a pusher speed set value, a stalker speed set value, an air flow rate set value, and the like.