A soft measurement method for temperature in a high-temperature gasifier of multi-component organic waste

Abstract

The present invention relates to a soft measurement method for the temperature in a high-temperature gasification furnace of multi-component organic waste. The method selects several process variables measured during the normal operation of the gasification furnace, performs data preprocessing, selects a BP neural network with three layers, uses the levenberg-marquardt algorithm to train the BP neural network, selects the neural network with the smallest result error among the neural networks with different numbers of hidden layer nodes as the target neural network, and uses the obtained neural network to realize the soft measurement of the temperature in the high-temperature gasification furnace of multi-component organic waste. The present invention effectively overcomes the shortcomings of traditional temperature physical measurement and the uncertainty and instability of the composition properties of the raw materials entering the furnace, can accurately predict the furnace temperature of the high-temperature gasification furnace of multi-component organic waste, and has important application value for temperature control and stable operation in the waste treatment process.

Description

Technical Field

[0001] The invention belongs to the fields of energy industry and chemical engineering, and in particular is a soft measurement method for the temperature in a high-temperature gasification furnace during the treatment of multi-component organic waste. Background Art

[0002] Industrial organic waste mainly includes solid waste and liquid waste such as distillation residue, waste activated carbon, waste catalyst, industrial sludge, waste tires, waste printed circuit boards, waste organic solvents, etc. Multi-component organic waste treatment is to use a gasifier to dispose of various organic wastes in coordination with other raw materials or fuels, while meeting the normal production requirements of the enterprise, ensuring product quality and environmental safety, and realizing the harmless disposal and resource utilization of organic wastes. At the same time, it reduces the consumption of fossil fuels and achieves carbon emission reduction. In the gasifier, the mixed slurry composed of multi-component organic wastes undergoes pyrolysis and gasification under high temperature conditions to generate synthesis gas mainly composed of CO2, CO, and H2, and the inorganic matter will melt and solidify into glassy slag. However, in the actual treatment and utilization process, organic waste and coal are configured into a multi-component mixed slurry. Since it contains a large amount of low-boiling, volatile, and unstable organic matter, it will inevitably lose its organic matter quality and composition after being dried at a certain temperature, so it is impossible to perform conventional full element analysis and industrial analysis.

[0003] At present, the temperature measurement in the high-temperature gasifier for multi-component organic waste treatment mainly adopts thermocouple real-time monitoring, and transmits the signal to the central control room to provide the operator with the basis for operation. However, on the one hand, since the temperature in the gasifier is as high as 1200-1400℃, the thermocouple is easy to fail after running at this high temperature for a period of time. On the other hand, the thermocouple hole will gradually be covered by the ash in the gasifier, blocking the thermocouple from correctly measuring the temperature in the furnace. Therefore, the furnace temperature value measured by the thermocouple will gradually lose credibility and cannot meet the production needs of long-term operation. Under normal circumstances, the service life of the thermocouple installed before the system is started is about one week. After the gasification system is running normally, the thermocouple cannot be replaced on site. Therefore, the thermocouple has lost its important role in monitoring the furnace temperature and guiding the production operation, and can only be used to observe the temperature changes during the start-up and early operation of the system. Therefore, taking the temperature in the gasifier as the research object, establishing a soft measurement model has important application value for temperature control and stable operation in the process of multi-component organic waste treatment.

[0004] In the paper "Non-contact Temperature Measurement Technology of Coal Gasifier Wall" (Wang Jin, Men Changgui, Chemical Automation and Instrumentation, 2014.41: 349-351), the authors mentioned that the temperature of the inner wall of the gasifier can be measured by infrared thermal imaging. However, the disadvantage of this method is that the installation conditions are very strict, requiring no obstruction between the object to be measured and the infrared camera. During the gasification process of multi-component organic waste, a large amount of ash will be generated in the furnace, which will seriously affect the true results of infrared thermal imaging. In addition, infrared thermal imagers are relatively expensive and costly. Taking all factors into consideration, infrared imaging or other optical imaging methods are not suitable for solving the problem of measuring the temperature in the high-temperature gasification furnace of multi-component organic waste described in the present invention.

[0005] In CN105930585A, the inventor completed the simulation of the flow field and temperature field of the Shell gasifier based on the CFD method, and provided complete furnace information for the design and industrial production of the Shell gasifier. However, CFD calculations need to fully consider the chemical reactions occurring in the gasifier, and the multi-component organic waste treatment process is complicated due to the complex sources of organic waste, including antibiotic slag, distillation residue, waste activated carbon, high-salt wastewater, waste organic solvents, etc., resulting in complex and changeable chemical reactions in the gasifier. In addition to the coal gasification series reactions in which activated carbon participates, there are also high-temperature cracking reactions of various organic substances, HCl and H2S generation reactions caused by the introduction of Cl elements and S elements in high-salt wastewater, etc., and it is impossible to determine the specific situation of the reaction in the gasifier, so CFD calculations are not suitable for the problems to be solved in the present invention. Compared with CFD calculations, the advantage of neural networks is that there is no need to have too much understanding of the entire reaction process. As long as enough data is provided for training, accurate calculation results can be obtained. Therefore, the present invention selects neural networks as a method for realizing soft measurement of temperature in a multi-component organic waste high-temperature gasifier.

[0006] In CN102175345A, the inventor selected dry coal, oxygen flow rate into the center of the furnace, oxygen-coal ratio, hydrogen H2, carbon monoxide CO, and methane CH4 content of the outlet synthesis gas as input variables of the soft measurement model, and realized the soft measurement of the furnace temperature of the multi-nozzle opposed water-coal slurry gasifier based on the BP artificial neural network, and the relative error of the result was within 5%, and the prediction accuracy was high. However, unlike the water-coal slurry gasification process of the above patent, the difficulty of the multi-component organic waste treatment process targeted by the present invention is:

[0007] 1) In the process of water-coal slurry gasification, the main components of the raw water-coal slurry are only water and coal powder. By determining the amount of coal entering the furnace, the situation of the main reactants in the gasification reaction can be determined. Therefore, the inventors in the above patents selected dry basis coal as one of the input variables; however, in the process of coordinated treatment of multi-component organic waste, in addition to coal powder, there are also organic wastes such as distillation residues, antibiotic residues, organic solvents, etc. in the slurry. Although the proportions when they are mixed into the slurry are known, the composition of these organic wastes themselves is also unstable. Therefore, it is impossible to select an input variable to determine the situation of the reactants in the process of multi-component organic waste treatment.

[0008] 2) In the process of water-coal slurry gasification, the oxygen-coal ratio is a very important parameter, because the fundamental purpose of controlling the oxygen-coal ratio is to control the oxygen-carbon ratio. The level of the oxygen-carbon ratio will directly affect the components of the gas production and the amount of effective gas; in the process of coordinated treatment of multi-component organic waste, due to the complex and unstable composition of various organic wastes, the oxygen-carbon ratios of materials obtained from different batches will be different even if the oxygen-to-carbon ratio is controlled to be the same. Therefore, the oxygen-to-carbon ratio is not suitable as an input variable for soft measurement of the temperature of the high-temperature gasifier in the coordinated treatment of multi-component organic waste.

[0009] 3) Currently, most elemental analyzers require samples to be dry, otherwise the instrument will be damaged or the measurement will be inaccurate. In the process of multi-component organic waste treatment, many organic wastes are in a state of solid-liquid mixture, such as antibiotic residues, distillation residues, waste organic solvents, etc. After drying, the liquid organic matter will be lost. Therefore, it is impossible to perform full elemental analysis on the slurry, and the oxygen-carbon ratio or elemental composition of the slurry cannot be used as an input variable for the soft measurement of the temperature of the high-temperature gasifier in the treatment of multi-component organic waste. Summary of the invention

[0010] In view of the above research background, the present invention establishes a soft measurement method for the temperature in a high-temperature gasification furnace of multi-component organic waste, comprising the following steps:

[0011] Step 1, collecting data obtained during the operation of an actual multi-component organic waste high-temperature gasification furnace, and selecting characteristic variables and target variables; the characteristic variables include the gasifier cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber syngas temperature; the target variable is the gasifier internal temperature;

[0012] Step 2, preprocessing the characteristic variable data collected in step 1 and its corresponding target variable data, using the box plot method to remove abnormal data points caused by measurement errors and abnormal working conditions, and then normalizing the data to remove the influence of the order of magnitude of the data itself;

[0013] Step 3, establishing a BP neural network model for measuring the temperature in the gasifier;

[0014] Step 4, use the data processed in step 2 (i.e., the processed characteristic variable data and its corresponding target variable data) to train the neural network model established in step 3, so as to determine the optimal number of hidden layer nodes and network weights of the BP neural network model, and obtain a soft measurement model that can accurately measure the temperature in the high-temperature gasification furnace of multi-component organic waste; use the soft measurement model to measure the temperature in the high-temperature gasification furnace of multi-component organic waste.

[0015] As a preferred embodiment of the present invention, the operation of the box plot method in step 2 is:

[0016] Arrange a set of data from small to large and divide it into 4 equal parts. The value of the first division point is called the lower quartile, recorded as Q1, the second division point is called the median, recorded as Q2, and the third division point is called the upper quartile, recorded as Q3; the difference between the lower quartile Q1 and the upper quartile Q3 is called the interquartile range IQR, that is, the formula for calculating IQR is:

[0017] IQR=Q3-Q1

[0018] The minimum value of normal data, that is, the lower edge is:

[0019] Lower edge = Q1-1.5QIR

[0020] The maximum value of normal data, that is, the upper edge is:

[0021] Upper edge = Q3 + 1.5QIR

[0022] Data that is greater than the upper edge or less than the lower edge is considered as an abnormal data point and is deleted.

[0023] As a preferred embodiment of the present invention, the data normalization operation in step 2 is:

[0024] The measured data values ​​are converted to the interval [-1,1], and the normalization function used is:

[0025]

[0026] Among them, X i is the input data, X min is the minimum value of the input data, X max is the maximum value of the input data, X new is the normalized data.

[0027] As a preferred embodiment of the present invention, the specific steps of step 3 are:

[0028] Step 3.1, the number of neural network layers is set to three;

[0029] Step 3.2, the first layer is the input layer, the number of its nodes is equal to the number of feature variables, which is 6; the second layer is the hidden layer, the number of its nodes is determined by the following empirical formula:

[0030]

[0031] Where h is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a is a regulation constant between 1 and 10, which is determined by the subsequent training results. Here, h = 3 + a; the third layer is the output layer, with a node number of 1, which is the target of soft measurement: the temperature inside the gasifier;

[0032] Step 3.3, the transfer function from the input layer to the hidden layer is the hyperbolic tangent S-type function, and its formula is

[0033]

[0034] The transfer function from the hidden layer to the output layer is a linear transfer function, and its formula is

[0035] f(n)=n

[0036] Step 3.4, the training algorithm uses the Levenberg-Marquardt algorithm;

[0037] Step 3.5, let the BP neural network input layer node be x i , the hidden layer neuron node is p j , the output layer neuron node is y, and the weight value between the input layer neuron and the hidden layer neuron is w ij , the weight value between the hidden layer neurons and the output layer neurons is v j ; When the expected value of the output data is z, the calculation formula of the BP neural network is as follows:

[0038] Output of hidden layer neurons:

[0039]

[0040] The output of the output layer neurons:

[0041]

[0042] The error value of the output result is:

[0043]

[0044] Step 3.6: In order to reduce the error E of the output result, it is necessary to adjust the weight value w between neurons. ij and v j; Since the negative gradient is the direction in which the function value decreases fastest, a learning rate η is set. The weight is adjusted η units each time along the negative gradient direction. The learning rate η is set to 0.01, and the adjustment each time is:

[0045]

[0046] The adjustment order of BP neural network is:

[0047] 1) First adjust the weight v from the hidden layer to the output layer j , the iterative formula for weight adjustment from hidden layer to output layer is:

[0048] v j (t+1)=v j (t)+ηy(1-y)p j

[0049] 2) Adjust the weight w from the input layer to the hidden layer ij , the iterative formula for adjusting the weights from the input layer to the hidden layer is:

[0050] w ij (t+1)=w ij (t)+ηδ j x i

[0051] where δ j =(zy)f′(u j )v j ,u=∑ i w ij x i ;

[0052] By continuously adjusting the weight values ​​between neurons, the error value E is gradually reduced until the convergence requirement is met.

[0053] As a preferred embodiment of the present invention, the step 4 comprises the following specific steps:

[0054] The measured data are used to train neural networks with different numbers of hidden layer neurons, where the number of hidden layer neurons of each neural network is h=3+a, and a is 1 to 10 respectively; the error values ​​E of the output results of these neural networks are compared, and the neural network with the smallest error is used as the soft measurement model of the temperature in the high-temperature gasification furnace of multi-component organic waste.

[0055] Compared with the prior art, the present invention is characterized in that:

[0056] 1) The present invention selects neural network as the implementation method of temperature soft measurement in the high-temperature gasifier of multi-component organic waste. Since the source of hazardous waste in the collaborative treatment process of multi-component organic waste is complex, the chemical reactions in the gasifier are complex and changeable, and the neural network does not need to have too much understanding of the entire reaction process. As long as sufficient data is provided for training, accurate calculation results can be obtained.

[0057] 2) The neural network structure selected by the present invention is a three-layer BP neural network. Generally speaking, a three-layer neural network can approximate any nonlinear function. The input variables of the present invention are 6. The three-layer neural network structure can ensure the rapid convergence of the neural network while achieving high-accuracy soft measurement, and reduce the time cost of debugging the neural network. BP (back propagation) is an error back propagation algorithm. It realizes the automatic adjustment of neuron weights through the two processes of forward propagation of information and back propagation of errors, which can help the calculation results of the neural network achieve higher accuracy.

[0058] 3) The present invention selects some process variables that can reflect the temperature conditions in the gasifier during operation as input variables of the neural network, including cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber syngas temperature, and finally realizes the soft measurement of the temperature in the high-temperature gasifier of multi-component organic waste. The proposed method has high accuracy, and compared with the true value measured by the thermocouple in the production process (accurate data when the thermocouple is not invalid), the error is within the allowable range in industrial production.

[0059] 4) The advantage of the present invention in selecting six input variables, namely cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber synthesis gas temperature, is that these data are all obtained by automatic testing of instruments in the actual production line. Compared with data such as slurry concentration and calorific value that need to be sampled and measured in the laboratory, the input variable data selected by the present invention are easy to obtain and have a high data frequency, while avoiding sampling errors caused by uneven distribution of slurry components. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a schematic diagram of the process of coordinated treatment of multi-component organic waste;

[0061] Figure 2 A flowchart for establishing and training the model of the present invention;

[0062] Figure 3 A structural topological diagram of a neural network used in the present invention;

[0063] Figure 4 Comparison between the prediction results of the soft sensor model and the real data;

[0064] Figure 5 It is the relative percentage error between the predicted result of the soft sensor model and the actual result;

[0065] Figure 6 To compare the predicted results of the comparative soft sensor model with the real data;

[0066] Figure 7 It is the relative percentage error between the predicted results of the comparative soft sensor model and the actual results. DETAILED DESCRIPTION

[0067] The present invention is further described and illustrated below in conjunction with specific embodiments. The technical features of each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0068] like Figure 1 As shown, it is a process schematic diagram of the coordinated treatment process of multi-component organic waste. The present invention selects cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber synthesis gas temperature as input variables. The selected characteristic variable data are all obtained by automatic testing of instruments in the actual production line. Compared with the slurry concentration, calorific value and other data that need to be sampled and measured in the laboratory, the selected input variable data are easy to obtain and have a high data frequency, and at the same time avoids sampling errors caused by uneven distribution of slurry components.

[0069] like Figure 2 As shown, the soft measurement method for the temperature in a high-temperature gasification furnace of multi-component organic waste provided by the present invention comprises the following steps:

[0070] Step 1: Collect data obtained during the operation of the actual multi-component organic waste high-temperature gasifier, summarize the reaction principle and operating experience, and select the characteristic variables and target variables of the soft sensor model. The selected characteristic variables include cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber syngas temperature. The target variable is the temperature inside the gasifier.

[0071] Step 2: pre-process the characteristic variable data collected in step 1 and its corresponding target variable data, use the box plot method to remove abnormal data points caused by measurement errors and abnormal working conditions, and then normalize the data to remove the influence of the order of magnitude of the data itself on the model. The specific operation of the box plot method is:

[0072] Arrange a set of data from small to large and divide it into 4 equal parts. The value of the first dividing point is called the lower quartile, recorded as Q1, the second dividing point is called the median, recorded as Q2, and the third dividing point is called the upper quartile, recorded as Q3. The difference between the lower quartile Q1 and the upper quartile Q3 is called the interquartile range IQR, that is, the formula for calculating IQR is:

[0073] IQR=Q3-Q1

[0074] The minimum value (lower edge) of normal data is:

[0075] Lower edge = Q1-1.5QIR

[0076] The maximum value (upper edge) of normal data is:

[0077] Upper edge = Q3 + 1.5QIR

[0078] Data that is greater than the upper edge or less than the lower edge is considered an abnormal data point and needs to be deleted.

[0079] The specific operation of normalization is:

[0080] The measured data values ​​are converted to the interval [-1,1], and the normalization function used is:

[0081]

[0082] Among them, X i is the input data, X min is the minimum value of the input data, X max is the maximum value of the input data, X new is the normalized data.

[0083] Step 3: randomly divide the normalized data in step 2 into three groups of 70%:15%:15% as training set, test set and validation set respectively.

[0084] Step 4: Establish a BP neural network model for measuring the temperature inside the gasifier. Figure 3 The structural topology diagram of the BP neural network of the present invention is shown; the specific steps of establishing the BP neural network model for measuring the temperature in the gasifier are:

[0085] In step 4.1, the number of neural network layers is set to three.

[0086] Step 4.2, the first layer is the input layer, the number of its nodes is equal to the number of feature variables, which is 6; the second layer is the hidden layer, the number of its nodes is determined by the following empirical formula:

[0087]

[0088] Among them, h is the number of hidden layer nodes, m is the number of input layer nodes, n is the number of output layer nodes, and a is a regulation constant between 1 and 10, which is determined by the subsequent training results. Here, h = 3 + a. The third layer is the output layer, with a node number of 1, which is the target of soft measurement: the temperature inside the gasifier.

[0089] Step 4.3, the transfer function from the input layer to the hidden layer is the hyperbolic tangent S-type function, and its formula is

[0090]

[0091] The transfer function from the hidden layer to the output layer is a linear transfer function, and its formula is

[0092] f(n)=n

[0093] Step 4.4: Use the Levenberg-Marquardt algorithm to train the BP network with a target error of 1×10 -7 ,The maximum number of iterations is 10000 and the learning rate is 0.01.

[0094] Step 4.5, let the BP neural network input layer node be x i , the hidden layer neuron node is p j , the output layer neuron node is y, and the weight value between the input layer neuron and the hidden layer neuron is w ij , the threshold is θ, and the weight between the hidden layer neurons and the output layer neurons is v j When the expected value of the output data is t, the calculation formula of the BP neural network is as follows:

[0095] Output of hidden layer neurons:

[0096]

[0097] The output of the output layer neurons:

[0098]

[0099] The error value of the output result is:

[0100]

[0101] Step 4.6: In order to reduce the error E of the output result, it is necessary to adjust the weight value w between neurons. ij and v j Since the negative gradient is the direction in which the function value decreases fastest, a learning rate η is set, and the weight is adjusted by η units each time along the negative gradient direction. The learning rate η of the present invention is set to 0.01, and the adjustment each time is:

[0102]

[0103] The adjustment order of BP neural network is:

[0104] 1) First adjust the weight v from the hidden layer to the output layer j , the iterative formula for weight adjustment from hidden layer to output layer is:

[0105] v j (t+1)=v j (t)+ηy(1-y)p j

[0106] 2) Adjust the weight w from the input layer to the hidden layer ij , the iterative formula for adjusting the weights from the input layer to the hidden layer is:

[0107] w ij (t+1)=w ij (t)+ηδ j x i

[0108] where δ j =(zy)f′(u j )v j ,u=∑ i w ij x i ;

[0109] This adjustment method is called gradient descent. By continuously adjusting the weight values ​​between neurons, the error value E is gradually reduced until the convergence requirement is reached.

[0110] Step 5, using the processed training data (each training data in the present invention includes a set of characteristic variable data and its corresponding target variable data) to analyze the neural network model established in step 4, thereby determining the optimal number of hidden layer nodes and network weights of the BP neural network, and obtaining a soft measurement model that can accurately measure the temperature in the high-temperature gasification furnace of multi-component organic waste. The specific method is: using the measurement data to train each neural network with different numbers of hidden layer neurons h=3+a (a=1-10) (i.e., neural networks with 4, 5, 6, 7, 8, 9, 10, 11, 12, and 13 hidden layer neurons respectively), and comparing the error values ​​E of the output results of these neural networks, the neural network with the smallest error is the target model of the present invention.

[0111] The following experiments illustrate that the soft measurement method of the temperature in a high-temperature gasification furnace of multi-component organic waste of the present invention is effective and feasible:

[0112] The data of the Texaco high-temperature gasifier for the coordinated treatment of multi-component organic waste in a certain unit collected from April 1 to 6, 2021 were used as the original data. After normalization, they were randomly divided into three groups of 70%: 15%: 15%, respectively, as the training set, test set and validation set. The characteristic variables selected were cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber synthesis gas temperature. After training the neural network model with the data of the training set and the test set, the data of the validation set were input into the neural network, and the obtained simulation results were compared with the true values ​​measured by the thermocouples in the production process. Figure 4 As shown in the figure, the dashed line data is the simulation result of the neural network, and the solid line data is the real value. Compared with the real furnace temperature measured by the thermocouple, the average relative percentage error of the simulation results of the soft sensor model is 0.71%, the average absolute error is 9.04℃, and the root mean square error is 12.62. Figure 5 As shown, the maximum relative percentage error is 4.25%, which is within the allowable error range in industrial production.

[0113] For comparison, the data of these five days are still used as the original data, and the same normalization and grouping processing are performed. The characteristic variables are selected as slurry flow, central oxygen flow, oxygen-to-material ratio, and H2, CH4, and CO contents of the syngas out of the quench chamber. The gasifier temperature simulation results obtained after training the network are compared with the actual values ​​measured by the thermocouple. Figure 6 As shown in the figure, the dashed line data is the simulation result of the neural network, and the solid line data is the real value. Compared with the real furnace temperature measured by the thermocouple, the average relative percentage error of the simulation results of the soft sensor model is 0.957%, the average absolute error is 12.25℃, and the root mean square error is 17.25. Figure 7 As shown, the maximum relative percentage error is 5.26%.

[0114] It can be seen that the present invention can achieve a high degree of accuracy in the soft measurement of the temperature inside the high-temperature gasification furnace of multi-component organic waste. It solves the problem of limited selection of characteristic variables caused by the complex and non-fixed composition of the reaction raw materials in the coordinated treatment process of multi-component organic waste and the inability to perform elemental analysis. It successfully uses the process variables of the gasifier operation including the circulating cooling water related data and the temperature of the synthesis gas as the input variables of the neural network, and finally realizes the soft measurement of the temperature inside the high-temperature gasification furnace of multi-component organic waste.

[0115] In summary, the soft measurement method of the temperature in a high-temperature gasification furnace of multi-component organic waste of the present invention can accurately and effectively simulate and estimate the temperature in a high-temperature gasification furnace of multi-component organic waste, which is helpful for the smooth operation of the coordinated treatment process of multi-component organic waste.

Claims

1. A soft measurement method for the temperature in a high-temperature gasification furnace of multi-component organic waste, characterized in that: The following steps are involved: Step 1, collecting data obtained during the operation of the actual multi-component organic waste high-temperature gasification furnace, and selecting characteristic variables and target variables; The characteristic variables include the gasifier cooling water inlet flow, cooling water outlet flow, cooling water inlet temperature, cooling water outlet temperature, cooling water pressure, and quench chamber syngas temperature, and the target variable is the gasifier internal temperature; Step 2, preprocessing the characteristic variable data collected in step 1 and its corresponding target variable data, using the box plot method to remove abnormal data points caused by measurement errors and abnormal working conditions, and then normalizing the data to remove the influence of the order of magnitude of the data itself; Step 3, establishing a BP neural network model for measuring the temperature in the gasifier; Step 4, use the data processed in step 2 to train the neural network model established in step 3, so as to determine the optimal number of hidden layer nodes and network weights of the BP neural network model, and obtain a soft measurement model that can accurately measure the temperature in the high-temperature gasification furnace of multi-component organic waste; use the soft measurement model to measure the temperature in the high-temperature gasification furnace of multi-component organic waste.

2. The soft measurement method for the temperature in a high-temperature gasification furnace of multi-component organic waste according to claim 1 is characterized in that: The operation of the box plot method described in step 2 is: Arrange a set of data from small to large and divide it into 4 equal parts. The value of the first division point is called the lower quartile, recorded as Q1, the second division point is called the median, recorded as Q2, and the third division point is called the upper quartile, recorded as Q3; the difference between the lower quartile Q1 and the upper quartile Q3 is called the interquartile range IQR, that is, the formula for calculating IQR is: IQR=Q3-Q1 The minimum value of normal data, that is, the lower edge is: Lower edge = Q1-1.5QIR The maximum value of normal data, that is, the upper edge is: Upper edge = Q3 + 1.5QIR Data that is greater than the upper edge or less than the lower edge is considered as an abnormal data point and is deleted.

3. The soft measurement method for temperature in a high-temperature gasification furnace of multi-component organic waste according to claim 1 is characterized in that: The data normalization operation in step 2 is: The measured data values ​​are converted to the interval [-1,1], and the normalization function used is: Among them, X i is the input data, X min is the minimum value of the input data, X max is the maximum value of the input data, X new is the normalized data.

4. The soft measurement method for temperature in a high-temperature gasification furnace of multi-component organic waste according to claim 1 is characterized in that: The specific steps of step 3 are: Step 3.1, the number of neural network layers is set to three; Step 3.2, the first layer is the input layer, the number of its nodes is equal to the number of feature variables, which is 6; the second layer is the hidden layer, the number of its nodes is determined by the following empirical formula: Where h is the number of nodes in the hidden layer, m is the number of nodes in the input layer, n is the number of nodes in the output layer, and a is a regulation constant between 1 and 10, which is determined by the subsequent training results. Here, h = 3 + a; the third layer is the output layer, with a node number of 1, which is the target of soft measurement: the temperature inside the gasifier; Step 3.3, the transfer function from the input layer to the hidden layer is the hyperbolic tangent S-type function, and its formula is The transfer function from the hidden layer to the output layer is a linear transfer function, and its formula is f(n) = n Step 3.4, the training algorithm uses the Levenberg-Marquardt algorithm; Step 3.5, let the BP neural network input layer node be x i , the hidden layer neuron node is p j , the output layer neuron node is y, and the weight value between the input layer neuron and the hidden layer neuron is w ij , the weight value between the hidden layer neurons and the output layer neurons is v j ; When the expected value of the output data is z, the calculation formula of the BP neural network is as follows: Output of hidden layer neurons: The output of the output layer neurons: The error value of the output result is: Step 3.6: In order to reduce the error E of the output result, it is necessary to adjust the weight value w between neurons. ij and v j ; Since the negative gradient is the direction in which the function value decreases fastest, a learning rate η is set. The weight is adjusted η units each time along the negative gradient direction. The learning rate η is set to 0.01, and the adjustment each time is: The adjustment order of BP neural network is: 1) First adjust the weight v from the hidden layer to the output layer j , the iterative formula for weight adjustment from hidden layer to output layer is: v j (t+1)=v j (t)+ηy(1-y)p j 2) Adjust the weight w from the input layer to the hidden layer ij , the iterative formula for adjusting the weights from the input layer to the hidden layer is: In ij (t+1)=in ij (t)+ηδ j x i among themδ j =(zy)f′(u j )v j ,u=∑ i w ij x i ; By continuously adjusting the weight values ​​between neurons, the error value E is gradually reduced until the convergence requirement is met.

5. The soft measurement method for temperature in a high-temperature gasification furnace of multi-component organic waste according to claim 1 is characterized in that: The specific steps of step 4 are as follows: The measured data are used to train neural networks with different numbers of hidden layer neurons, where the number of hidden layer neurons of each neural network is h=3+a, and a is 1 to 10 respectively; the error values ​​E of the output results of these neural networks are compared, and the neural network with the smallest error is used as the soft measurement model of the temperature in the high-temperature gasification furnace of multi-component organic waste.

Images