CN-121995816-A - Garbage incineration power generation efficiency enhancement method and system based on multi-mode sensing fusion and predictive control

CN121995816ACN 121995816 ACN121995816 ACN 121995816ACN-121995816-A

Abstract

The invention discloses a garbage incineration power generation efficiency enhancing method and system based on multi-mode sensing fusion and predictive control. Aiming at the problems of control lag and unstable combustion caused by the fact that the characteristics of the garbage entering the furnace cannot be predicted in the prior art, the invention deploys a multi-mode sensing array consisting of an industrial camera, a near infrared spectrometer and an industrial microphone in a garbage feeding channel and acquires visual, spectral and acoustic data of the garbage in real time. And through the multi-mode fusion prediction model, the combustion characteristics such as the heat value, the moisture and the like of the garbage entering the furnace in the future are accurately predicted. The prediction characteristics are fused with real-time working condition data and are input into a combustion process optimization prediction model to generate optimal combination control parameters for the future. The parameter is used as a feedforward instruction to be issued to a distributed control system to adjust the combustion process in advance. The invention converts the lag feedback control into the active feedforward control, thereby fundamentally stabilizing combustion, improving steam quality and yield and obviously improving power generation efficiency.

Inventors

GAO XIN
ZHANG YU
WANG TONG
LIU HANG
LUO WENFENG
Tian Yening

Assignees

贵阳中电环保发电有限公司

Dates

Publication Date: 20260508
Application Date: 20251204

Claims (10)

1. The garbage incineration power generation efficiency increasing method based on multi-mode perception fusion and predictive control is characterized by comprising the following steps of: a. Driving a multi-mode sensing array deployed in a garbage feeding channel to synchronously acquire data of garbage to be charged in real time, wherein the multi-mode sensing array comprises an industrial camera, a near infrared spectrometer and an industrial microphone, and the acquired data comprises visual data, spectral data and acoustic data; b. inputting the visual data, the spectral data and the acoustic data into a preset multi-mode fusion prediction model to predict a group of combustion characteristic parameters of the garbage in the furnace in a specified future time period; c. acquiring working condition data of the current combustion process from a distributed control system in real time, and fusing the predicted combustion characteristic parameters and the real-time working condition data into a high-dimensional state feature vector; d. Inputting the high-dimensional state feature vector into a preset combustion process optimization prediction model to generate a group of optimal combination control parameters which aim at stabilizing combustion and maximizing steam yield and are oriented to the future; e. and sending the optimal combination control parameters to the distributed control system as a feedforward control instruction so as to drive an executing mechanism connected with the distributed control system to adjust the combustion process in advance.
2. The method of claim 1, wherein the multimodal fusion prediction model comprises three branches independent of each other, a convolutional neural network CNN branch for extracting spatial texture features from the visual data, a one-dimensional convolutional neural network 1D-CNN branch for extracting component absorption peak features from the spectral data, and a convolutional recurrent neural network CRNN branch for extracting time-sequential dynamic features from the acoustic data.
3. The method according to claim 1 or 2, wherein the processing of the acoustic data further comprises extracting a mel-frequency cepstrum coefficient (MFCCs) or a sound spectrum map capable of reflecting deep physical characteristics such as garbage density, blocking, etc. from the original audio signal acquired by the industrial microphone, and inputting the mel-frequency cepstrum coefficient or sound spectrum map as the acoustic feature of the multi-modal fusion prediction model.
4. The method of claim 1, wherein the predicted combustion characteristics parameters include at least predicted heating value, predicted moisture and predicted density or bulk, and the optimal combination control parameters include at least target feed rate, target grate speed for each grate segment, target primary air total and distribution ratio for each plenum, and target secondary air total and distribution ratio for each nozzle.
5. The method of claim 1, further comprising the step of closed loop self-optimization of continuously monitoring and collecting actual operating condition data fed back by the distributed control system after the feedforward control command is issued, comparing the actual operating condition data with a predicted target to calculate an error, and utilizing the error to perform online learning or periodic retraining on the multimodal fusion prediction model and the combustion process optimization prediction model.
6. The utility model provides a msw incineration electricity generation synergistic system based on multimode perception fuses and predictive control which characterized in that includes: The multi-mode sensing array is deployed in the garbage feeding channel and comprises an industrial camera, a near infrared spectrometer and an industrial microphone, and is used for acquiring visual data, spectral data and acoustic data of garbage; the data interface is used for acquiring working condition data of the combustion process from the distributed control system; a processor, coupled to the array and the data interface, configured to: a. Inputting the acquired visual, spectral and acoustic data into a multi-mode fusion prediction model to predict a group of combustion characteristic parameters of the garbage entering the furnace; b. Fusing the combustion characteristic parameters and the working condition data into a high-dimensional state feature vector; c. inputting the state feature vector into a combustion process optimization prediction model to generate a group of optimal combination control parameters facing the future; d. And sending the optimal combination control parameters to a distributed control system as a feedforward control instruction through the data interface so as to adjust the combustion process in advance.
7. The system of claim 6, wherein the multimodal fusion prediction model, when implemented in the processor, includes three parallel processing branches, a convolutional neural network CNN branch for extracting spatial texture features from the visual data, a one-dimensional convolutional neural network 1D-CNN branch for extracting component absorption peak features from the spectral data, and a convolutional neural network CRNN branch for extracting temporal dynamic features from the acoustic data.
8. The system of claim 6 or 7, wherein the processor, when processing the acoustic data, is further configured to extract a mel-frequency cepstrum coefficient (MFCCs) or a spectrogram capable of reflecting deep physical characteristics such as garbage density, block size, etc. from the original audio signal acquired by the industrial microphone, and input the extracted features as acoustic features of the multi-modal fusion prediction model.
9. The system of claim 6, wherein the combustion characteristic parameters include at least a predicted heating value, a predicted moisture content, and a predicted density or block size, and the optimal combination control parameters include at least a target feed rate, a target grate speed for each grate segment, a target primary air total amount, and a distribution ratio for each plenum, and a target secondary air total amount, and a distribution ratio for each nozzle.
10. The system of claim 6, wherein the processor is further configured to execute closed loop self-optimization logic to continuously receive actual operating condition data fed back by the distributed control system after issuing the feedforward control instruction, to compare the actual operating condition data with a prediction target to calculate an error, and to use the error to learn online or retrain the multimodal fusion prediction model and the combustion process optimization prediction model.

Description

Garbage incineration power generation efficiency enhancement method and system based on multi-mode sensing fusion and predictive control Technical Field The invention belongs to the technical field of garbage disposal and clean energy, and particularly relates to a control method and a system integrating a multisource sensing technology, a deep learning model and model predictive control (Model Predictive Control, MPC) thought, which are particularly suitable for intelligent and refined closed-loop optimization control of an urban household garbage incineration process so as to realize remarkable improvement of power generation efficiency. Background The main goal of the garbage incineration power generation is to convert chemical energy contained in garbage into electric energy with as high efficiency as possible. In actual production operation, a control system of the garbage incineration process is a key to achieve the aim. Currently, most waste incineration power plants commonly use a Distributed Control System (DCS) as a basic control platform. On the platform, an operator manually or semi-automatically adjusts key control variables such as feeding rate, speed of each section of the fire grate, air quantity and distribution of primary air and secondary air according to a series of process parameters such as temperature of each region of a furnace, temperature and pressure of main steam, smoke components (such as O 2 and CO content) and the like, and combines long-term accumulated operation experience. The control mode commonly adopted in the prior art can be summarized as a 'passive manual intervention control mode based on feedback of hysteresis process parameters'. Although this mode plays a role in guaranteeing the basic operation of a waste incineration power plant, its inherent technical drawbacks also greatly limit the further improvement of the energy conversion efficiency, in particular in the following aspects: 1. inherent hysteresis of control combustion is a complex physicochemical process with large inertia. When a batch of garbage with changed characteristics enters the hearth, the influence of the garbage on the combustion working condition (such as the reduction of the temperature of the hearth or the severe fluctuation of the temperature of the hearth) can be measured by a sensor such as a thermocouple, a pressure transmitter and the like and reflected on the DCS system after a period of time. The operator or automatic control loop then adjusts again, essentially a "sheep-mending" type of hysteresis compensation. The hysteresis makes the combustion condition always in a frequently fluctuating state, and the combustion condition is difficult to maintain in an optimal combustion interval, so that continuous fluctuation of steam quality is caused. 2. The black box is recognized on the characteristics of the material to be charged, and the urban household garbage is a typical heterogeneous and unsteady fuel. The key combustion characteristics of heat value, moisture content, density, ash, volatile matters, physical forms (such as block degree and looseness) and the like have great differences and uncertainties in different batches and even in the same batch. The traditional DCS control system can not acquire any quantitative or qualitative information of the garbage to be charged in advance, and can only treat the garbage as a black box with constant characteristics. This "unknowing" of fuel properties is the root cause of combustion instability. 3. The deep contradiction between the power generation efficiency and the stable operation is that the severe fluctuation of the combustion working condition directly leads to the unstable quality (mainly temperature and pressure) of the main steam generated by the boiler. Unstable steam enters the steam turbine to apply work, so that the operation efficiency and the power generation capacity of the steam turbine generator unit are seriously affected, and the thermal shock and the mechanical damage to key parts such as the steam turbine blades are caused in the long term, so that the service life of equipment is shortened. To ensure that the waste is burned out and meets stringent environmental emission standards (e.g., control dioxin production), operators tend to employ conservative strategies for excessive air distribution. However, excessive combustion air can carry away a large amount of heat in the furnace, directly resulting in reduced thermal efficiency of the boiler. This forms a difficult reconciliation between ensuring combustion stability and pursuing high thermal efficiency. 4. Over-reliance on personal experience of operators in the prior art framework, the operating level of incinerators is highly dependent on a small number of experienced and sophisticated operators. They can make relatively accurate decisions and adjustments by virtue of long-term observations and summaries of equipment characteristics and combustion phenomena. How