A method and system for controlling steam-gas ratio of sludge pyrolysis gasification furnace based on artificial intelligence

Abstract

The present invention relates to the field of sludge treatment, specifically to a method and system for regulating the steam-gas ratio of a sludge pyrolysis gasifier based on artificial intelligence. The amount of gas generated at the next moment of a nonlinear dynamic model is calculated according to current process parameters, and then the amount of gas generated at the same moment as the next moment in the gas generation sequence of a neural network model is determined. If the deviation between the two is less than a preset value, the prediction interval is determined based on the deviation. Otherwise, the trained neural network model is re-run according to the latest process parameters, and the nonlinear dynamic model is re-generated. The amount of gas generated at the next moment is calculated according to the current process parameters and the new nonlinear dynamic model; the optimization objective function of the nonlinear dynamic model is solved to obtain a steam inlet control value, and the steam inlet control value is used to adjust the steam inlet valve. The present invention effectively overcomes the problem of large errors in steam-gas ratio regulation using a model predictive control method.

Description

Technical Field

[0001] The present invention relates to the field of sludge treatment, and specifically to a method and system for regulating the steam-gas ratio of a sludge pyrolysis gasifier based on artificial intelligence. Background Art

[0002] As an emerging treatment method, sludge pyrolysis gasification technology converts sludge into reusable energy and by-products through high-temperature pyrolysis and gasification processes, achieving the reduction, resource utilization, and harmlessness of sludge. The pyrolysis gasifier plays a key role in the sludge treatment process. During the pyrolysis process, organic substances in the sludge are decomposed into combustible gases, liquids, and solid products under anaerobic or anoxic conditions, and during the gasification process, the generated solids and liquids are further converted into gases. Through this series of chemical reactions, not only can the sludge volume be effectively reduced, but it can also be converted into high-value-added energy products such as syngas, carbon, and tar, realizing the efficient utilization of resources.

[0003] During the pyrolysis gasification process, the regulation of the steam-gas ratio is a key factor affecting the system performance and product quality. The steam-gas ratio refers to the ratio of water vapor to sludge during the reaction process, which directly affects the reaction temperature, reaction rate, and the composition and quality of the products. Appropriate regulation of the steam-gas ratio can optimize the pyrolysis gasification process, improve the yield and quality of gas products, reduce the content of harmful substances in by-products, and thus enhance the efficiency and economy of the entire system. However, traditional methods for regulating the steam-gas ratio have problems of lag and inaccuracy and are difficult to respond to complex and changing reaction conditions in real time. Summary of the Invention

[0004] To solve the above problems, the present invention provides a method for regulating the steam-gas ratio of a sludge pyrolysis gasifier based on artificial intelligence, and the method includes the following steps:

[0005] Collect the process parameters of the sludge pyrolysis gasifier, construct the training samples of the neural network model, the label of the training samples is the generated gas volume sequence, and the generated gas volume sequence corresponds to multiple future time points; determine the input and output of the nonlinear dynamic model of the pyrolysis gasifier, construct the nonlinear dynamic model of the pyrolysis gasifier, the independent variables of the nonlinear dynamic model are the steam intake at the current moment and the generated gas volume at the current moment, and the dependent variable is the generated gas volume at the next moment.

[0006] Calculate the gas generation amount at the next moment of the non - linear dynamic model according to the current process parameters, and then determine the gas generation amount at the same moment as the next moment in the gas generation amount sequence of the neural network model. If the deviation between the two is less than the preset value, determine the prediction interval based on the deviation; otherwise, re - run the trained neural network model according to the latest process parameters, and re - generate the non - linear dynamic model, and calculate the gas generation amount at the next moment according to the current process parameters and the new non - linear dynamic model.

[0007] Solve the optimization objective function of the non - linear dynamic model to obtain the steam inlet amount control value, and use the steam inlet amount control value to adjust the steam inlet valve.

[0008] Preferably, the process parameters at least include the steam inlet amount, the feeding speed, the temperature, the pressure, the gas generation amount, and the content of the preset gas components in the generated gas. The preset gas at least includes carbon monoxide, hydrogen, and methane.

[0009] Collect the values of each process parameter to form a sequence of collected values of the process parameters, align the sequences of collected values of all process parameters according to the collection time, and the later the collection time in the sequence of collected values, the later the collection time.

[0010] Starting from the k - th of all the aligned sequences of collected values, take the k collected values closest to the (k + 1)-th collected value and before the (k + 1)-th collected value in each sequence of collected values as a subsequence in the training sample. All subsequences form a sample. The (k + 1)-th to (k + m)-th values of the sequence of collected values corresponding to the gas generation amount form the gas generation amount sequence; increase k by 1 to get the next sample until k = M - m; where k and m are positive integers greater than 3, and m < k < M, and M is the length of the aligned sequence of collected values.

[0011] Preferably, the re - generation of the non - linear dynamic model is specifically:

[0012] Re - generate the non - linear dynamic model using multiple steam inlet amounts and multiple gas generation amounts closest to the current moment.

[0013] Input the data composed of the latest process parameters into the neural network model to obtain the gas generation amount at the next moment, and input the steam inlet amount and the gas generation amount at the current moment into the new non - linear dynamic model to obtain the gas generation amount at the next moment, and determine the prediction interval based on the deviation.

[0014] Preferably, the determination of the prediction interval based on the deviation is specifically:

[0015] Calculate the ratio of the deviation to the preset value, and determine the size of the prediction interval based on the ratio.

[0016] Preferably, the control value of the steam inlet quantity is obtained by solving the optimization objective function of the non-linear dynamic model, specifically as follows:

[0017] Define the optimization objective function as:

[0018]

[0019] where N is the prediction interval, y i is the amount of generated gas at the i-th moment predicted by the neural network model, r is the desired amount of generated gas, and y i ′ is the amount of generated gas at the i-th moment predicted by the non-linear dynamic model, u i is the steam inlet quantity at the i-th moment, and Q and R are weight matrices.

[0020] Use a quadratic programming solver to solve the steam inlet quantity sequence to minimize the optimization objective function.

[0021] Retain the steam inlet quantity closest to the current one in the steam inlet quantity sequence as the control value of the steam inlet quantity at the next moment.

[0022] On the other hand, the present invention provides a steam-gas ratio regulation system for a sludge pyrolysis gasification furnace based on artificial intelligence. The system includes the following modules:

[0023] A model construction module, which is used to collect the process parameters of the sludge pyrolysis gasification furnace, construct a training sample of the neural network model, the label of the training sample is the sequence of the amount of generated gas, and the sequence of the amount of generated gas corresponds to multiple future time points; determine the input and output of the non-linear dynamic model of the pyrolysis gasification furnace, and construct the non-linear dynamic model of the pyrolysis gasification furnace. The independent variables of the non-linear dynamic model are the steam inlet quantity and the amount of generated gas at the current moment, and the dependent variable is the amount of generated gas at the next moment;

[0024] A parameter prediction module, which is used to calculate the amount of generated gas at the next moment of the non-linear dynamic model according to the current process parameters, and then determine the amount of generated gas at the same moment as the next moment in the sequence of the amount of generated gas of the neural network model. If the deviation between the two is less than the preset value, determine the prediction interval based on the deviation. Otherwise, re-run the trained neural network model according to the latest process parameters, re-generate the non-linear dynamic model, and calculate the amount of generated gas at the next moment according to the current process parameters and the new non-linear dynamic model.

[0025] A regulation module, which is used to solve the optimization objective function of the non-linear dynamic model to obtain the control value of the steam inlet quantity, and use the control value of the steam inlet quantity to adjust the steam inlet valve.

[0026] Preferably, the process parameters at least include steam inlet volume, feed rate, temperature, pressure, generated gas volume, and content of a preset gas component in the generated gas, and the preset gas at least includes carbon monoxide, hydrogen, and methane.

[0027] Collect the values of each process parameter to form a sequence of collected values of the process parameters. Align the sequences of collected values of all process parameters according to the collection time, and the later the collection time in the sequence of collected values, the later it is.

[0028] Starting from the k-th of all the aligned sequences of collected values, take the k collected values that are before the (k + 1)-th collected value and closest to the (k + 1)-th collected value in each sequence of collected values as a subsequence in the training sample. All the subsequences form a sample. The (k + 1)-th to (k + m)-th values of the sequence of collected values corresponding to the generated gas volume form the generated gas volume sequence; increase k by 1 to obtain the next sample until k = M - m; where k and m are positive integers greater than 3, and m < k < M, and M is the length of the aligned sequence of collected values.

[0029] Preferably, the regenerating the non-linear dynamic model is specifically:

[0030] Regenerate the non-linear dynamic model by using multiple steam inlet volumes and multiple generated gas volumes closest to the current time.

[0031] Input the data composed of the latest process parameters into the neural network model to obtain the generated gas volume at the next moment, and input the steam inlet volume and generated gas volume at the current moment into the new non-linear dynamic model to obtain the generated gas volume at the next moment, and determine the prediction interval based on the deviation.

[0032] Preferably, the determining the prediction interval based on the deviation is specifically:

[0033] Calculate the ratio of the deviation to the preset value, and determine the size of the prediction interval based on the ratio.

[0034] Preferably, the solving the optimization objective function of the non-linear dynamic model to obtain the steam inlet volume control value is specifically:

[0035] Define the optimization objective function as:

[0036]

[0037] where N is the prediction interval, y i is the generated gas volume at the i-th moment predicted by the neural network model, r is the desired generated gas volume, y i ′ is the generated gas volume at the i-th moment predicted by the non-linear dynamic model, u i$Q_i$ is the steam inlet flow rate at the $i$-th moment, and $Q$ and $R$ are weight matrices.

[0038] A quadratic programming solver is used to solve the steam inlet flow rate sequence to minimize the optimization objective function.

[0039] The steam inlet flow rate closest to the current one in the steam inlet flow rate sequence is retained as the steam inlet flow rate control value for the next moment.

[0040] In addition, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned method is implemented.

[0041] Aiming at the problems of poor real-time performance and large error in the regulation of the steam-to-gas ratio of the sludge pyrolysis gasifier, the present invention uses a neural network model to predict the generated gas volume within a period of time, and uses a non-linear dynamic model to generate the generated gas volume at the next moment in real time, avoiding the consumption of computing resources brought by the neural network model, reducing the error of the model predictive control method at the same time, and improving the accuracy and real-time performance of the steam-to-gas ratio regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of Embodiment 1;

[0043] Figure 2 、 3 is a schematic diagram of constructing training samples;

[0044] Figure 4 is a schematic diagram of the optimal objective function value;

[0045] Figure 5 is a structural diagram of Embodiment 2. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0046] In this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0047] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0048] Embodiment 1. The present invention provides a method for regulating the steam-gas ratio of a sludge pyrolysis gasification furnace based on artificial intelligence. As Figure 1 shown, the method includes the following steps:

[0049] S1. Collect the process parameters of the sludge pyrolysis gasification furnace, construct the training samples of the neural network model, where the label of the training samples is the generated gas volume sequence, and the generated gas volume sequence corresponds to multiple future time points; determine the input and output of the nonlinear dynamic model of the pyrolysis gasification furnace, and construct the nonlinear dynamic model of the pyrolysis gasification furnace. The independent variables of the nonlinear dynamic model are the steam intake at the current moment and the generated gas volume at the current moment, and the dependent variable is the generated gas volume at the next moment.

[0050] When pyrolyzing sludge, the water content in the sludge is reduced to within 20% through a drying system, and then granulation is carried out. The sludge particles are sent into the pyrolysis gasification furnace. In the gasification furnace, the substances in the sludge are decomposed by high temperature. During the operation of the gasification furnace, the temperature, pressure, feeding speed, steam intake speed, etc. of the gasification furnace are controlled, all of which will affect the pyrolysis efficiency. In one embodiment, collect the process parameters of the sludge pyrolysis gasification furnace, and the process parameters at least include steam intake, feeding speed, temperature, pressure, generated gas volume, and the content of preset gas components in the generated gas. The preset gas at least includes carbon monoxide, hydrogen, and methane.

[0051] Put the values corresponding to each process parameter into a sequence, so that multiple sequences will be obtained, such as a temperature sequence, a pressure sequence, a steam intake sequence, etc. Each sequence has multiple values, and each value is the value collected at a certain moment. Collect the values of each process parameter to form a collection value sequence of the process parameters, align the collection value sequences of all process parameters according to the collection time, and the later the collection time in the collection value sequence, the later it is. After alignment, the values with the same serial number in all sequences are the values collected at the same moment. For example, the 5th value in all sequences is the value collected at the same moment.

[0052] To train a neural network model, training samples are obtained using the aligned sequences. Specifically, starting from the k-th of all the aligned acquired value sequences, for each acquired value sequence, the k acquired values before the (k + 1)-th acquired value and closest to the (k + 1)-th acquired value are used as a subsequence in the training sample. All the subsequences form a sample. The (k + 1)-th to (k + m)-th values of the acquired value sequence corresponding to the generated gas volume form the generated gas volume sequence. Increase k by 1 to obtain the next sample until k = M - m, where k and m are positive integers greater than 3, m < k < M, and M is the length of the aligned acquired value sequences.

[0053] For example, if all the aligned sequences have 10 values, then M = 10. Further assume k = 4 and m = 2. Starting from the 5-th value, the 4 acquired values before the 5-th value and closest to the 5-th value are used as a subsequence, and thus temperature subsequences, pressure subsequences, etc. are obtained. A sample includes multiple subsequences, and the label corresponding to this sample is the sequence formed by the 5-th to 6-th values of the acquired value sequence of the generated gas volume. Then increase k by 1. At this time, k = 5, that is, starting from the 6-th value, the 4 acquired values before the 6-th value and closest to the 6-th value are used as a subsequence, and thus temperature subsequences, pressure subsequences, etc. are obtained. These subsequences form another sample, and the label corresponding to this sample is the sequence formed by the 6-th to 7-th values of the acquired value sequence of the generated gas volume. And so on until k = 8, a total of 5 samples are obtained. Figure 2-3 Shows the comparison of samples before and after k + 1. Figure 2-3 In it, k = 4, m = 3, and M = 13.

[0054] After obtaining the training samples, the loss between the output of the samples input into the neural network model and the labels of the samples is calculated to complete the training of the neural network model. The neural network model includes, but is not limited to, convolutional neural networks, recurrent neural networks, etc.

[0055] The training of the neural network model requires a large amount of computing resources, and also requires strong computing power during use, resulting in a relatively long computing time. After training the neural network model, the generated gas volume for a period of time in the future is predicted according to the process parameters before the current moment at preset intervals. However, the parameters in the gasifier are constantly changing, which cannot achieve the effect of real-time control. The present invention combines MPC (Model Predictive Control) to regulate the steam-gas ratio of the gasifier. Specifically, the inputs and outputs of the nonlinear dynamic model of the pyrolysis gasifier are determined, and the nonlinear dynamic model of the pyrolysis gasifier is constructed. The independent variables of the nonlinear dynamic model are the steam input at the current moment and the generated gas volume at the current moment, and the dependent variable is the generated gas volume at the next moment.

[0056] Specifically, historical steam intake and gas production are collected, and a function for the gas production at the next moment is obtained by fitting, with the expression form being y k+i = f(y k , u k ), where y k is the gas production at the k-th moment, and u k is the steam intake at the k-th moment. The process of calculating f preferably uses a non-linear optimization solver.

[0057] S2. Calculate the gas production at the next moment of the non-linear dynamic model according to the current process parameters, and then determine the gas production at the same moment as the next moment in the gas production sequence of the neural network model. If the deviation between the two is less than the preset value, a prediction interval is determined based on the deviation; otherwise, the trained neural network model is re-run according to the latest process parameters, and the non-linear dynamic model is re-generated, and the gas production at the next moment is calculated according to the current process parameters and the new non-linear dynamic model.

[0058] After obtaining the trained neural network model and non-linear dynamic model, the gas production at the next moment can be predicted according to the current process parameters. Since noise and the like will interfere with the non-linear dynamic model, and the number of parameters in the non-linear dynamic model is relatively small, the predicted gas production at the next moment is not accurate enough. The present invention combines the neural network model and the non-linear dynamic model to predict the gas production at the next moment. Specifically, the neural network model is calculated according to the latest process parameters every preset time to obtain the gas production at multiple subsequent moments, while the non-linear dynamic model directly predicts the gas production at the next moment, simultaneously avoiding the problems of excessive resource consumption required for running the neural network model and inaccurate prediction of the non-linear dynamic model. The non-linear dynamic model predicts the gas production at the next moment, and at the same time, the gas production corresponding to this moment is found in the gas production sequence predicted by the neural network model, and the difference between the two is judged. If the difference is too large, the trained neural network model is re-run according to the current process parameters, and the non-linear dynamic model is re-generated. The non-linear dynamic model is fitted according to the historically collected values. If the difference is relatively large, a new f(y k , u k ) will be generated by re-fitting, that is, the non-linear dynamic model is re-generated using the multiple steam intakes and multiple gas productions closest to the current moment.

[0059] Input the data consisting of the latest process parameters into the neural network model to obtain the gas generation amount at the next moment, and input the steam inlet amount and gas generation amount at the current moment into a new non-linear dynamic model to obtain the gas generation amount at the next moment, and determine the prediction interval based on the deviation. Among them, the latest process parameters are the process parameters closest to the time when the neural network model is used for prediction. For example, if the neural network model needs to be run for prediction at the 10th minute, it is the process parameters closest to the 10th minute that have been collected.

[0060] S3. Solve the optimization objective function of the non-linear dynamic model to obtain the steam inlet amount control value, and use the steam inlet amount control value to adjust the steam inlet valve.

[0061] Define the optimization objective function as:

[0062]

[0063] Among them, N is the prediction interval, y i is the gas generation amount at the i-th moment predicted by the neural network model, r is the expected gas generation amount, y i ′ is the gas generation amount at the i-th moment predicted by the non-linear dynamic model, u i is the steam inlet amount at the i-th moment, and Q and R are weight matrices.

[0064] Use a quadratic programming solver to solve the steam inlet amount sequence to minimize the optimization objective function.

[0065] Retain the steam inlet amount closest to the current in the steam inlet amount sequence as the steam inlet amount control value at the next moment.

[0066] As Figure 4 shown, the number of gas generation amounts predicted by the neural network model is more than that generated by the non-linear dynamic model. When calculating the optimization objective function, align the gas generation amount predicted at the next moment by the non-linear dynamic model and the gas generation amount predicted by the neural network model in time, and use the predicted gas generation amount corresponding to the next moment after alignment as the gas generation amount at the next moment of the neural network model. The number of predicted gas generation amounts generated by the neural network model is more than the number of gas generation amounts predicted by the non-linear dynamic model.

[0067] Moreover, the inputs of the neural network model and the non-linear dynamic model are different. The input of the neural network model is multiple sequences composed of the collected process parameter values of the same size as the sample, and the input of the non-linear dynamic model is the steam inlet amount and gas generation amount at the current moment.

[0068] Embodiment 2. The present invention provides a steam-gas ratio regulation system for a sludge pyrolysis gasification furnace based on artificial intelligence, as Figure 5As shown, the system includes the following modules:

[0069] A model construction module, which is used to collect the process parameters of the sludge pyrolysis gasification furnace, construct the training samples of the neural network model, where the label of the training samples is the generated gas volume sequence, and the generated gas volume sequence corresponds to multiple future time points; determine the input and output of the non-linear dynamic model of the pyrolysis gasification furnace, and construct the non-linear dynamic model of the pyrolysis gasification furnace. The independent variables of the non-linear dynamic model are the steam inlet volume at the current moment and the generated gas volume at the current moment, and the dependent variable is the generated gas volume at the next moment.

[0070] A parameter prediction module, which is used to calculate the generated gas volume at the next moment of the non-linear dynamic model according to the current process parameters, and then determine the generated gas volume at the same moment as the next moment in the generated gas volume sequence of the neural network model. If the deviation between the two is less than the preset value, a prediction interval is determined based on the deviation. Otherwise, the trained neural network model is re-run according to the latest process parameters, and the non-linear dynamic model is regenerated. The generated gas volume at the next moment is calculated according to the current process parameters and the new non-linear dynamic model.

[0071] A regulation module, which is used to solve the optimization objective function of the non-linear dynamic model to obtain the steam inlet volume control value, and use the steam inlet volume control value to adjust the steam inlet valve.

[0072] Preferably, the process parameters at least include the steam inlet volume, the feeding speed, the temperature, the pressure, the generated gas volume, and the content of the preset gas components in the generated gas. The preset gas at least includes carbon monoxide, hydrogen, and methane.

[0073] Collect the values of each process parameter to form a collection value sequence of the process parameters. Align the collection value sequences of all process parameters according to the collection time, and the later the collection time in the collection value sequence, the later it is.

[0074] Starting from the k-th of all the aligned collection value sequences, take the k collection values closest to the (k + 1)-th collection value and before the (k + 1)-th collection value in each collection value sequence as a subsequence in the training sample. All subsequences form a sample. The (k + 1)-th to (k + m)-th values of the collection value sequence corresponding to the generated gas volume form the generated gas volume sequence; increase k by 1 to get the next sample until k = M - m; where k and m are positive integers greater than 3, and m < k < M, and M is the length of the aligned collection value sequence.

[0075] Preferably, the regeneration of the non-linear dynamic model is specifically as follows:

[0076] Regenerate the non-linear dynamic model by using multiple steam inlet volumes and multiple generated gas volumes closest to the current moment.

[0077] Input the data consisting of the latest process parameters into the neural network model to obtain the gas generation amount at the next moment, and input the steam inlet amount and the gas generation amount at the current moment into a new non-linear dynamic model to obtain the gas generation amount at the next moment, and determine the prediction interval based on the deviation.

[0078] Preferably, the determination of the prediction interval based on the deviation is specifically:

[0079] Calculate the ratio of the deviation to the preset value, and determine the size of the prediction interval based on the ratio.

[0080] Preferably, the solution of the optimization objective function of the non-linear dynamic model to obtain the steam inlet amount control value is specifically:

[0081] Define the optimization objective function as:

[0082]

[0083] where N is the prediction interval, y i is the gas generation amount at the i-th moment predicted by the neural network model, r is the desired gas generation amount, y i ′ is the gas generation amount at the i-th moment predicted by the non-linear dynamic model, u i is the steam inlet amount at the i-th moment, and Q and R are weight matrices.

[0084] Use a quadratic programming solver to solve the steam inlet amount sequence to minimize the optimization objective function.

[0085] Retain the steam inlet amount closest to the current one in the steam inlet amount sequence as the steam inlet amount control value at the next moment.

[0086] Embodiment 3: The present invention also provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the method as described in Embodiment 1.

[0087] Embodiment 4: The present invention also provides a computer device, which at least includes a memory and a processor, and a computer program is stored on the memory, and the computer program, when executed by the processor, implements the method as described in Embodiment 1.

[0088] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of adding a necessary general hardware platform. Of course, it can also be implemented by a combination of hardware and software. Based on such an understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a computer product. The present invention can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Other embodiments can also be adopted; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for controlling the steam-gas ratio of a sludge pyrolysis gasifier based on artificial intelligence, characterized in that: The method includes the following steps: Collect the process parameters of the sludge pyrolysis gasification furnace, construct the training samples of the neural network model, the label of the training samples is the generated gas volume sequence, and the generated gas volume sequence corresponds to multiple future time points; determine the input and output of the non-linear dynamic model of the pyrolysis gasification furnace, construct the non-linear dynamic model of the pyrolysis gasification furnace, the independent variables of the non-linear dynamic model are the steam inlet volume at the current moment and the generated gas volume at the current moment, and the dependent variable is the generated gas volume at the next moment; Calculate the generated gas volume at the next moment of the non-linear dynamic model according to the current process parameters, and then determine the generated gas volume at the same moment as the next moment in the generated gas volume sequence of the neural network model. If the deviation between the two is less than the preset value, determine the prediction interval based on the deviation. Otherwise, re-run the trained neural network model according to the latest process parameters, and re-generate the non-linear dynamic model. Calculate the generated gas volume at the next moment according to the current process parameters and the new non-linear dynamic model; Solve the optimization objective function of the non-linear dynamic model to obtain the steam inlet volume control value, and use the steam inlet volume control value to adjust the steam inlet valve.

2. The method according to claim 1, characterized in that The process parameters at least include the steam inlet volume, the feeding speed, the temperature, the pressure, the generated gas volume, and the content of the preset gas components in the generated gas. The preset gas at least includes carbon monoxide, hydrogen, and methane; Collect the values of each process parameter to form the collected value sequence of the process parameters, align the collected value sequences of all process parameters according to the collection time, and the later the collection time is in the collected value sequence; Starting from the k-th of all the aligned collected value sequences, take the k collected values closest to the (k + 1)-th collected value and before the (k + 1)-th collected value in each collected value sequence as a subsequence in the training sample. All subsequences form a sample, and the (k + 1)-th to (k + m)-th values of the collected value sequence corresponding to the generated gas volume form the generated gas volume sequence; increase k by 1 to obtain the next sample until k = M - m; where k and m are positive integers greater than 3, and m < k < M, and M is the length of the aligned collected value sequence.

3. The method according to claim 1, characterized in that The re-generation of the non-linear dynamic model is specifically: Re-generate the non-linear dynamic model by using multiple steam inlet volumes and multiple generated gas volumes closest to the current moment; Input the data composed of the latest process parameters into the neural network model to obtain the generated gas volume at the next moment, and input the steam inlet volume and the generated gas volume at the current moment into the new non-linear dynamic model to obtain the generated gas volume at the next moment, and determine the prediction interval based on the deviation.

4. The method according to claim 1 or 3, characterized in that The determination of the prediction interval based on the deviation is specifically: Calculate the ratio of the deviation to the preset value, and determine the size of the prediction interval based on the ratio.

5. The method according to claim 1, characterized in that The solution of the optimization objective function of the non-linear dynamic model to obtain the steam inlet volume control value is specifically: The optimization objective function is defined as: ; Where N is the prediction interval, is the amount of gas generated at the i-th moment predicted by the neural network model, r is the expected amount of gas generated, is the amount of generated gas at the i-th moment predicted by the nonlinear dynamic model, is the steam intake at the i-th moment, Q and R are weight matrices; Use a quadratic programming solver to solve the steam inlet volume sequence to minimize the optimization objective function; Retain the steam inlet volume closest to the current in the steam inlet volume sequence as the steam inlet volume control value at the next moment.

6. A sludge pyrolysis gasification furnace steam-gas ratio control system based on artificial intelligence, characterized in that: The system includes the following modules: A model construction module, configured to collect process parameters of a sludge pyrolysis gasification furnace and construct training samples for a neural network model. Labels of the training samples are sequences of gas production amounts, and the sequences of gas production amounts correspond to multiple future time points. Determine inputs and outputs of a nonlinear dynamic model of the pyrolysis gasification furnace, and construct the nonlinear dynamic model of the pyrolysis gasification furnace. An independent variable of the nonlinear dynamic model is the steam inlet amount at the current moment and the gas production amount at the current moment, and a dependent variable is the gas production amount at the next moment. A parameter prediction module, configured to calculate the gas production amount at the next moment of the nonlinear dynamic model according to current process parameters, and then determine the gas production amount at the same moment as the next moment in the sequence of gas production amounts of the neural network model. If a deviation between the two is less than a preset value, determine a prediction interval based on the deviation. Otherwise, re-run the trained neural network model according to the latest process parameters, re-generate the nonlinear dynamic model, and calculate the gas production amount at the next moment according to the current process parameters and the new nonlinear dynamic model. A regulation module, configured to solve an optimization objective function of the nonlinear dynamic model to obtain a steam inlet amount control value, and adjust a steam inlet valve by using the steam inlet amount control value.

7. The system according to claim 6, characterized in that The process parameters at least include a steam inlet amount, a feeding speed, a temperature, a pressure, a gas production amount, and contents of preset gas components in the generated gas. The preset gas at least includes carbon monoxide, hydrogen, and methane. Collect values of each process parameter to form a sequence of collected values of the process parameters, align the sequences of collected values of all the process parameters according to the collection time, and the later the collection time is in the sequence of collected values, the later the collection time is. Starting from the k-th one of all the aligned sequences of collected values, use the k collected values before the (k + 1)-th collected value and closest to the (k + 1)-th collected value in each sequence of collected values as a subsequence in the training sample. All the subsequences form a sample. The (k + 1)-th to (k + m)-th values of the sequence of collected values corresponding to the gas production amount form a sequence of gas production amounts. Increase k by 1 to obtain the next sample until k = M - m, where k and m are positive integers greater than 3, m < k < M, and M is the length of the aligned sequence of collected values.

8. The system according to claim 6, characterized in that The re-generation of the nonlinear dynamic model is specifically as follows: Re-generate the nonlinear dynamic model by using multiple steam inlet amounts and multiple gas production amounts closest to the current moment. Input data formed by the latest process parameters into the neural network model to obtain the gas production amount at the next moment, and input the steam inlet amount and the gas production amount at the current moment into the new nonlinear dynamic model to obtain the gas production amount at the next moment, and determine a prediction interval based on the deviation.

9. The system according to claim 6, characterized in that The solution of the optimization objective function of the nonlinear dynamic model to obtain a steam inlet amount control value is specifically as follows: The optimization objective function is defined as: ; Where N is the prediction interval, is the amount of gas generated at the i-th moment predicted by the neural network model, r is the expected amount of gas generated, is the amount of generated gas at the i-th moment predicted by the nonlinear dynamic model, is the steam intake at the i-th moment, Q and R are weight matrices; Use a quadratic programming solver to solve a sequence of steam inlet amounts to minimize the optimization objective function. Retain the steam inlet amount closest to the current one in the sequence of steam inlet amounts as the steam inlet amount control value at the next moment.

10. A computer storage device, wherein a computer program is stored on the storage device, characterized in that: The computer program, when executed by a processor, implements the method according to any one of claims 1-5.

Images