Operation balance optimization method for waste incineration thermal power plant

Abstract

The invention discloses an operation balance optimization method for a waste incineration thermal power plant. Efficient and environment-friendly waste treatment and green energy recovery are achieved through regulation and control of core parameters. Comprising an intelligent control system, a carbon cyclic utilization technology and a flue gas purification technology, an optimization model taking comprehensive net income maximization and carbon emission intensity minimization as targets is constructed, a garbage heat value, incinerator temperature, steam parameters, power generation efficiency, station service power consumption rate and the like are selected as core parameters, an operation balance optimization target function is constructed, and the comprehensive net income maximization and the carbon emission intensity minimization are realized. And through an intelligent control system, key parameters such as the incinerator temperature, the excess air coefficient and the flue gas treatment temperature are monitored and adjusted in real time, it is ensured that garbage is fully combusted in the incinerator, harmful gas emission is reduced, and the power generation efficiency is improved. Compared with the prior art, efficient operation of the waste incineration thermal power plant and maximum utilization of green energy are achieved, and remarkable economic and environmental benefits are achieved.

Description

Technical Field

[0001] The present invention relates to the field of waste incineration power generation and low-carbon technology, and is applicable to energy efficiency improvement, carbon emission management and green energy recovery scenarios of domestic waste incineration power generation systems. It is particularly applicable to incineration efficiency optimization and carbon footprint control under multi-parameter dynamic coupling conditions, and specifically relates to an operation balance optimization method for a waste incineration thermal power plant. Background Art

[0002] The core of the waste incineration power generation system is to achieve efficient decomposition and energy conversion of organic matter in garbage by precisely controlling the incineration temperature, air supply and flue gas treatment parameters. However, in actual operation, the drastic fluctuations in the calorific value of garbage (such as changes in the moisture content of kitchen waste, increases and decreases in the proportion of plastic products) will cause the incineration conditions to frequently deviate from the optimal operating range. When the incineration temperature is lower than 850°C, highly toxic pollutants such as dioxins are easily produced; and a high excess air coefficient will reduce the furnace temperature, aggravate the generation of nitrogen oxides, and cause heat loss. Traditional control strategies mostly rely on manual experience to set fixed parameter thresholds, which are difficult to cope with the complex and changeable real-world scenarios of garbage composition, resulting in power generation efficiency fluctuations exceeding ±15%, and annual auxiliary fuel replacement needs as high as 30%-40%, which significantly increases operating costs.

[0003] In terms of carbon emissions management, current technologies primarily rely on direct flue gas emissions, failing to effectively capture and recycle the carbon dioxide produced by incineration. Furthermore, the existing carbon trading system lacks a mature price transmission mechanism, leading to a significant gap between the allocation of carbon emission quotas and the actual cost of emissions reductions. This has created a lack of incentive for businesses to proactively implement carbon capture retrofits.

[0004] The limitations of traditional operation optimization strategies further restrict the effectiveness of the system. Most incineration plants use a single-loop regulation method based on PID controllers, which can only achieve steady-state control of local parameters and lack global perception of the dynamic characteristics of the incineration process. For example, there is a significant spatiotemporal coupling effect between the flue gas treatment system (such as SCR denitrification and bag dust removal) and the main incineration unit. Flue gas temperature fluctuations will affect the activity of the catalyst, while changes in particulate matter concentration will reversely adjust the power consumption of the dust removal fan. This complex nonlinear relationship makes it difficult to establish an accurate mathematical representation for traditional optimization models, and multi-objective decision-making (such as maximizing economic benefits and minimizing carbon emissions) often falls into the Pareto frontier solution dilemma. In addition, existing systems generally adopt a centralized control architecture, and cloud servers undertake all data processing and strategy generation tasks. When faced with massive real-time monitoring data (such as temperature, pressure, gas composition, etc.), there are problems such as high transmission delay and computing resource bottlenecks, which make it difficult to meet the millisecond-level response control requirements.

[0005] Purpose of the Invention

[0006] Purpose of the invention: In response to the problems in the background technology, the present invention provides an operation balance optimization method for a waste incineration thermal power plant to achieve full combustion of waste and efficient energy conversion, innovatively construct an "incineration-purification-conversion" closed-loop carbon cycle system, use P2G technology to reduce carbon emission intensity, and through a multi-objective optimization model and edge-cloud collaborative architecture, meet carbon emission constraints and break through the incineration system regulation bottleneck.

[0007] Technical solution: The present invention discloses a method for optimizing the operation balance of a waste incineration thermal power plant, comprising the following steps:

[0008] Step 1: Use high-precision sensors to monitor the calorific value of garbage, incinerator temperature, flue gas emission temperature, power generation efficiency, and plant power consumption rate in real time. Based on the real-time monitoring data, dynamically adjust the incinerator temperature, excess air coefficient, and flue gas treatment temperature to optimize the combustion process.

[0009] Step 2: Install a CO2 collection device in the flue gas treatment system and combine it with P2G technology to combine the collected CO2 with H2 to generate CH4, which is reused as auxiliary fuel in the incineration process to achieve carbon recycling;

[0010] Step 3: Establish an objective function that maximizes the comprehensive net benefit and minimizes the carbon emission intensity, which is composed of electricity sales revenue, carbon emission costs, auxiliary fuel purchase costs, flue gas treatment and P2G power operation costs;

[0011] Step 4: Based on the unit price of electricity, CO2 emissions from waste incineration, flue gas treatment time per unit of waste, incinerator temperature, excess air coefficient, and flue gas temperature, a spatiotemporal coupled neural network (ST-GNN) model is established. A continuous action space is designed to impose safety constraints on the objective function, which is then converted into an immediate reward.

[0012] Step 5: Introduce the improved MOEA collaborative optimization algorithm to perform dual-objective Pareto frontier analysis on maximizing comprehensive net benefits and minimizing carbon emission intensity, and obtain the optimization results. The improved MOEA collaborative optimization algorithm introduces a NSGA-III+MOPO hybrid framework. The first 30% of iterations use NSGA-III to generate the initial population, and the last 70% of iterations use MOPO to introduce DRL policy gradient update through multi-objective strategy optimization. Dynamic strategy adjustment is introduced in multi-objective optimization to balance the conflicting goals of maximizing comprehensive net benefits and minimizing carbon emission intensity.

[0013] Step 6: Based on the final optimization results, continuously adjust and optimize the garbage calorific value, incinerator temperature, flue gas treatment temperature, and P2G equipment operating parameters to ensure that the system is always in the best operating state.

[0014] Furthermore, the objective function of maximizing the comprehensive net benefit and minimizing the carbon emission intensity, which is composed of electricity sales revenue, carbon emission costs, auxiliary fuel purchase costs, flue gas treatment and P2G power operation costs, is as follows:

[0015]

[0016] Where: comprehensive net income Z, carbon emission intensity Q, R elec represents the unit price of electricity (yuan / kWh), η gen Indicates power generation efficiency, which is the amount of power generated per unit of garbage (kWh / ton), C fuel represents the auxiliary fuel cost (yuan / ton), η plant Indicates the power consumption rate of the plant, which is the proportion of the power consumption in the plant to the total power generation. emission represents the carbon emission cost (yuan / ton CO2), E CO2 Indicates the CO2 emissions from waste incineration, C operation represents the flue gas treatment and P2G power operating cost (yuan / hour), H treatment Indicates the flue gas treatment time per unit of garbage (hours / ton), D total Indicates the total amount of waste processed (tons).

[0017] Furthermore, step 4 establishes an intelligent control model that deeply integrates the spatiotemporal coupled neural network ST-GNN model with deep reinforcement learning DRL, and processes the objective function and the collected parameters as follows:

[0018] 1) Define the decision variable incinerator temperature adjustment range ΔT adj , excess air coefficient correction value Δα adj , P2G power regulation ratio ΔP P2G When the incinerator temperature rises, the combustion efficiency is improved, the power generation is increased, the power sales benefits are also increased, and the fuel consumption is increased. By optimizing the incinerator temperature, the power generation benefits and fuel costs are balanced and carbon emissions are reduced. When the excess air coefficient correction value Δα adj When it is increased, the combustion efficiency is improved, the heat loss is improved, and Δα is optimized adj Suppressing the formation of nitrogen oxides while avoiding thermal efficiency loss; when P2G power increases, CO2 conversion fuel increases, auxiliary fuel demand decreases, but P2G energy consumption increases, reducing carbon emission intensity through carbon recycling, while also balancing P2G electricity consumption;

[0019] 2) Real-time collection of six-dimensional state data: s t =[T furnace ,α air ,T flue ,R elec (t),E CO2 (t),Htreatment (t)], where T furnace ,α air ,T flue Represent the incinerator temperature, excess air coefficient, flue gas temperature, R elec Indicates the unit price of electricity (yuan / kWh), E CO2 represents the CO2 emissions from waste incineration, H treatment It represents the flue gas treatment time per unit of garbage (hours / ton), and is standardized using a sliding window to eliminate dimensional differences.

[0020] 3) Establish a spatiotemporal coupled neural network ST-GNN model to model the spatiotemporal dependencies of each unit in the incineration system:

[0021]

[0022] Among them, x v is the node v feature, including six-dimensional state data, N(v) is a set of spatial adjacent nodes, outputting the hidden state h v It is used to predict the dynamic response of the system. W1 is the spatial coupling weight of the neighbor node state, which is used to capture the spatial dependence between the incineration system units. W2 is the time coupling weight of the current node state, which is used to model the time dependence of the historical state. W3 is the global weight of the node feature. is the hidden state of node v in the lth layer, representing its spatiotemporal dynamic characteristics, represents the hidden state of the neighbor node u in layer l, and σ represents the activation function;

[0023] 4) Action space design and constraint processing: design a continuous action space and impose safety constraints on the objective function:

[0024]

[0025] Among them, T min 、T max Indicates the lower and upper safety limits of the incinerator temperature, α min , α max Indicates the reasonable lower and upper limits of the excess air coefficient, P 2G,min 、P 2G,max Indicates the P2G device power and its regulation boundary;

[0026] 5) Construct a reward function, convert the objective function into an immediate reward, and add an L2 regularization term to suppress oscillations:

[0027]

[0028] Where λ is the regularization coefficient.

[0029] Furthermore, the improved MOEA collaborative optimization algorithm is introduced in step 5 to perform a dual-objective Pareto frontier analysis on maximizing comprehensive net benefits and minimizing carbon emission intensity, as follows:

[0030] 1) Introducing the improved MOEA collaborative optimization, the dual-objective Pareto frontier of maximizing net benefits and minimizing carbon emission intensity is performed; introducing the NSGA-III+MOPO hybrid framework, the first 30% of iterations use NSGA-III to generate the initial population, and the last 70% of iterations introduce DRL policy gradient update through MOPO, the formula is:

[0031]

[0032] Among them, A t is the advantage function, α is the learning rate, θ is the parameter of the policy network, a t Used to generate actions, such as the incinerator temperature adjustment amplitude, T is the total number of time steps, indicating the duration of a single incineration process, is the gradient of the policy network parameter θ, π θ is the policy function;

[0033] 2) Probabilistic constraint modeling to quantify the uncertainty of carbon footprint. The formula is as follows:

[0034]

[0035] Among them, target The carbon emission intensity target value, δ, represents the confidence level. A δ value of 0.05 indicates a 95 percent probability of meeting the constraint. Based on the assumption that historical waste calorific value fluctuation data is normally distributed, 1,000 scenarios were generated and carbon emission constraints were calculated with a confidence level of 1-δ. An increase in waste calorific value increases in incineration efficiency, energy generation per unit of waste η, and electricity sales revenue. A decrease in waste calorific value requires an increase in incineration temperature or excess air coefficient to maintain combustion, leading to increased fuel costs.

[0036] 3) A hierarchical decision-making mechanism is adopted, and a lightweight DRL agent is used on the edge to perform local optimization in real time with a sampling frequency of 5Hz. The cloud uploads running data to the learning server every 30 minutes to update the global strategy parameters.

[0037] Furthermore, the steps of improving the MOEA collaborative optimization algorithm are as follows:

[0038] S1: Model the environment and use a one-stage delay differential equation to describe the combustion process:

[0039]

[0040] Where τ is the time constant, K is the gain coefficient, and ΔQ fuelis the fuel heat input, T setpoint Respectively represent the set temperature;

[0041] S2: ST-GNN training: Input historical state sequence Σ=[s0,s1,...,s T ], output the predicted system response, which includes the power generation efficiency change Δη gen , construct the loss function in is the predicted change of power generation efficiency, γ is the weight attenuation coefficient;

[0042] S3: DRL strategy training, collecting experience trajectories {s t ,a t ,r t ,s t+1}, calculate the advantage function A t =Q(s t ,a t )-V(s t ), update the policy network function: The truncation ratio ò=0.2 to prevent the policy from being updated too much; s t is the state vector, the current state of the environment, a t is the action vector, the decision action at the current moment, r t is the reward signal, the immediate benefit of the current action, s t+1 is the state at the next moment, Q(s t ,a t ) is the action value function, the expected cumulative reward of performing action a in state s, V(s t ) is the expected long-term value of the state value function in state s; θ policy The neural network weights generated for the policy network parameter control actions, such as the incinerator temperature adjustment policy; β represents the probability distribution;

[0043] S4: MOEA multi-objective optimization generates initial solutions covering the parameter space of incinerator temperature, excess air coefficient, flue gas temperature, and P2G power. Fast non-dominated sorting NSGA-II is used to extract the Pareto frontier, and the top 20% individuals are retained for the next generation.

[0044] S5: Online adaptive adjustment: Federated learning updates, aggregates local model gradients at the edge every 30 minutes, and updates the global strategy: where n k is the local data volume of the kth incineration plant, N is the total data volume; θ global represents the global angle, θ k represents the angle or local model parameter of the kth sample;

[0045] S6: If the number of iterations is reached or the termination condition is met, the iteration is terminated; otherwise, go to S4;

[0046] S7: Get the Z and Q corresponding to the optimal fitness value, that is, maximize the comprehensive net benefit Z while minimizing the carbon emission intensity Q.

[0047] Beneficial effects:

[0048] 1. The present invention uses high-precision sensors to monitor core parameters in real time, and uses an intelligent control system to dynamically adjust key parameters such as the incinerator temperature and excess air coefficient, thereby optimizing the combustion process and significantly improving power generation efficiency. This effectively reduces power generation efficiency fluctuations, reduces dependence on auxiliary fuels, reduces the cost of purchasing auxiliary fuels, and increases electricity sales revenue.

[0049] 2. This invention precisely controls incinerator temperature and other parameters to ensure complete combustion of waste, reducing the generation of harmful gases such as dioxins and nitrogen oxides. During flue gas treatment, a CO2 collection device is installed and combined with P2G technology to achieve carbon recycling, converting carbon dioxide into methane as an auxiliary fuel, reducing carbon emission costs and dependence on fossil fuels.

[0050] 3. This invention establishes an intelligent control framework that deeply integrates spatiotemporal coupled neural networks (ST-GNN) and deep reinforcement learning (DRL). It can fully perceive the spatiotemporal dependencies of various units in the incineration system (incinerator, flue gas purification, P2G). In this way, dynamic collaborative optimization among multiple units of the incineration system is achieved, effectively coping with the complex and changing composition of garbage, solving the problem that traditional control strategies are difficult to cope with complex working conditions, and improving the overall operational efficiency of the system.

[0051] 4. This invention adopts a hierarchical decision-making mechanism that collaborates with the edge and the cloud. A lightweight DRL agent is deployed at the edge to perform local optimization in real time at a sampling frequency of 5Hz, enabling rapid response to on-site changes. An optimization model is established to maximize comprehensive net benefits, taking into account factors such as electricity sales revenue, carbon emission costs, and auxiliary fuel purchase costs. The introduction of probabilistic constraint models and multi-objective optimization methods, such as the NSGA-III + MOPO hybrid framework, addresses the dilemma of multi-objective decision-making (such as maximizing economic benefits and minimizing carbon emissions) that often falls into the Pareto frontier solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 Schematic diagram of the system structure of the present invention;

[0053] Figure 2 It is the algorithm flow chart of the present invention;

[0054] Figure 3 It is a diagram of the incineration facility of the present invention. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention.

[0056] The present invention discloses an operation balance optimization method for a waste incineration thermal power plant, which specifically includes the following steps:

[0057] Step 1: High-precision sensors monitor core parameters such as waste calorific value, incinerator temperature, steam parameters, power generation efficiency, and plant power consumption in real time to ensure data accuracy and real-time availability. An intelligent control system dynamically adjusts key parameters such as incinerator temperature, excess air ratio, and flue gas treatment temperature based on real-time monitoring data to optimize the combustion process, improve power generation efficiency, and reduce harmful gas emissions.

[0058] Step 2: Install a CO2 collection device in the flue gas treatment system. Combined with P2G technology, the collected CO2 is combined with H2 through chemical or biological reactions to generate CH4, which is reused as auxiliary fuel in the incineration process, realizing carbon recycling and reducing carbon emission costs.

[0059] Step 3: Establish an optimization model to maximize the comprehensive net profit, which is composed of electricity sales revenue, carbon emission costs, auxiliary fuel purchase costs, flue gas treatment and P2G power operation costs. The optimization results are obtained through simulation calculations and fed back to the control system to continuously adjust and optimize the operating parameters to ensure that the system is always in the best operating state. This realizes the efficient operation of the waste incineration thermal power plant and maximizes the utilization of green energy, with significant economic and environmental benefits.

[0060] During the data collection and detection phase, the system uses high-precision sensors to monitor the core parameters of the waste incineration thermal power plant in real time. These parameters are key factors affecting waste incineration efficiency, environmental performance, and economic benefits. Fluctuations in the calorific value of waste directly affect the combustion stability and heat output of the incinerator, while controlling the incinerator temperature is crucial for complete waste combustion and the generation of harmful gases. By monitoring these parameters in real time, the system can promptly obtain the operating status of the waste incineration process, providing data support for subsequent optimization and control. The system also monitors flue gas emissions, including information such as the concentration and flow rate of harmful gases. This helps evaluate the treatment effectiveness of the flue gas purification system and ensure compliance with environmental standards. Furthermore, monitoring equipment status is a crucial component of data collection. By monitoring equipment operating parameters and status information, abnormalities can be detected promptly, potential failure risks can be prevented, and the stable operation of the system can be ensured.

[0061] In the parameter analysis and optimization phase, the system conducts an in-depth analysis of the core parameters of the waste incineration thermal power plant based on the collected data and constructs an operational balance optimization objective function. Through the intelligent control system, the system can adjust key parameters such as the incinerator temperature, excess air coefficient, and flue gas treatment temperature in real time. For example, by optimizing the control strategy of the incinerator temperature, it can ensure that the garbage is fully burned in the furnace, reduce the emission of harmful gases, and improve power generation efficiency. Adjustment of the excess air coefficient can balance the oxygen supply required for combustion and avoid incomplete combustion or energy waste due to insufficient or excessive air. The system also combines advanced data analysis technology and optimization algorithms. The system can predict the operating results under different parameter combinations, thereby finding the optimal parameter settings and maximizing the overall net benefit.

[0062] Carbon recycling is an important means of achieving low-carbon operation in waste incineration thermal power plants. In this process, the system collects the carbon dioxide generated during the incineration process through a CO2 collection device installed in the flue gas treatment system. Subsequently, using P2G (Power to Gas) technology, the collected CO2 is converted into CH4 and reused as an auxiliary fuel in the incineration process. This process not only achieves the recycling of carbon resources and reduces carbon emission costs, but also improves the system's energy efficiency. The CO2 collection device separates carbon dioxide from the flue gas through physical or chemical absorption methods. P2G technology then uses electricity to convert carbon dioxide and water into methane and oxygen. This process can store energy and use it as fuel when needed. In this way, the system not only reduces its dependence on fossil fuels, but also reduces carbon emissions and achieves the recovery and utilization of green energy.

[0063] The benefit evaluation phase primarily focuses on three aspects: revenue calculation, cost accounting, and overall net profit. The system calculates various revenue and cost indicators by monitoring and recording the waste incineration thermal power plant's operating data in real time. Revenue calculations include revenue from electricity sales and carbon emissions trading, while cost accounting covers fuel costs, equipment maintenance costs, and flue gas purification costs. By comparing data before and after optimization, the system can assess the impact of optimization measures on economic benefits.

[0064] The following objective function of operation balance optimization is established with the maximum net profit and the minimum carbon emission to facilitate the comparison of optimization effects under different processing volumes. The objective function is as follows:

[0065]

[0066] Where: comprehensive net income Z, carbon emission intensity Q, R elec represents the unit price of electricity (yuan / kWh), η gen Indicates power generation efficiency, which is the amount of power generated per unit of garbage (kWh / ton), C fuelrepresents the auxiliary fuel cost (yuan / ton), η plant Indicates the power consumption rate of the plant, which is the proportion of the power consumption in the plant to the total power generation. emission represents the carbon emission cost (yuan / ton CO2), E CO2 Indicates the CO2 emissions from waste incineration, C operation represents the flue gas treatment and P2G power operating cost (yuan / hour), H treatment Indicates the flue gas treatment time per unit of garbage (hours / ton), D total Indicates the total amount of waste processed (tons).

[0067] The optimization framework is as follows:

[0068] 1. Define your goals:

[0069] Maximize the comprehensive net benefit Z while minimizing the carbon emission intensity Q, satisfy the dynamic constraints, and define the decision variable incinerator temperature adjustment range ΔT adj , excess air coefficient correction value Δα adj , P2G power regulation ratio ΔP P2G When the incinerator temperature rises, the combustion efficiency is improved, the power generation is increased, the power sales benefits are also increased, and the fuel consumption is increased. By optimizing the incinerator temperature, the power generation benefits and fuel costs are balanced and carbon emissions are reduced. When the excess air coefficient correction value Δα adj When it is increased, the combustion efficiency is improved, the heat loss is improved, and Δα is optimized adj Inhibit the generation of nitrogen oxides while avoiding thermal efficiency loss; when P2G power increases, CO2 conversion fuel increases and auxiliary fuel demand decreases, but P2G energy consumption increases, reducing carbon emission intensity through carbon recycling while weighing P2G electricity consumption.

[0070] 2. Data collection and modeling:

[0071] Real-time collection of six-dimensional state data: s t =[T furnace ,α air ,T flue ,R elec (t),E CO2 (t),H treatment (t)], where T furnace ,α air ,T flue Represent the incinerator temperature, excess air coefficient, flue gas temperature, R elec Indicates the unit price of electricity (yuan / kWh), E CO2 represents the CO2 emissions from waste incineration, H treatment It represents the flue gas treatment time per unit of garbage (hours / ton), and is standardized using a sliding window to eliminate dimensional differences.

[0072] 3. State representation and dynamic modeling:

[0073] A spatiotemporal coupled neural network (ST-GNN) was established to model the spatiotemporal dependencies of the various units of the incineration system (incinerator, flue gas purification, P2G):

[0074]

[0075] Among them, x v is the node feature, N(v) is the set of spatial adjacent nodes, and the output hidden state h v Used to predict the dynamic response of the system.

[0076] 4. Action space design and constraint processing:

[0077] Design a continuous action space and impose safety constraints on the objective function:

[0078]

[0079] 5. Constructing reward function:

[0080] Convert the objective function into an immediate reward and add an L2 regularization term to suppress oscillations:

[0081]

[0082] Where λ is the regularization coefficient.

[0083] 6. Introducing improved MOEA collaborative optimization:

[0084] The dual-objective Pareto frontier of net benefits and minimizing carbon emission intensity is introduced, and the NSGA-III+MOPO hybrid framework is introduced. The first 30% iterations use NSGA-III to generate the initial population; the last 70% iterations introduce DRL policy gradient update through MOPO, the formula is:

[0085]

[0086] Among them, A t is the advantage function, α is the learning rate, θ is the parameter of the policy network, a t Used to generate actions, such as the incinerator temperature adjustment amplitude, T is the total number of time steps, indicating the duration of a single incineration process, is the gradient of the policy network parameter θ, π θ is the policy function.

[0087] 7. Probabilistic constraint modeling to quantify the uncertainty of carbon footprint. The formula is as follows:

[0088]

[0089] Among them, target The carbon emission intensity target value, δ, represents the confidence level. A δ value of 0.05 indicates a 95 percent probability of meeting the constraint. Based on the assumption of a normal distribution for historical waste calorific value fluctuation data, 1,000 scenarios were generated, and carbon emission constraints with a confidence level of 1-δ were calculated. An increase in waste calorific value increases in incineration efficiency, unit waste power generation η, and electricity sales revenue. A decrease in waste calorific value requires an increase in incineration temperature or excess air ratio to maintain combustion, leading to increased fuel costs. Based on historical waste calorific value fluctuation data (assuming a normal distribution), 1,000 scenarios were generated, and carbon emission constraints with a confidence level of 1-δ were calculated.

[0090] 8. Adopt a hierarchical decision-making mechanism: A lightweight DRL agent is used on the edge to perform local optimization in real time, with a sampling frequency of 5Hz. The cloud uploads running data to the learning server every 30 minutes to update the global strategy parameters.

[0091] The algorithm steps are as follows:

[0092] Step 1: Model the environment and use a one-stage delay differential equation to describe the combustion process:

[0093]

[0094] Where τ is the time constant, K is the gain coefficient, and ΔQ fuel is the fuel heat input, T setpoint Respectively represent the set temperature.

[0095] Step 2: ST-GNN training: Input historical state sequence Σ=[s0,s1,...,s T ], output the predicted system response, which includes the power generation efficiency change Δη gen , construct the loss function in is the predicted change of power generation efficiency, γ is the weight attenuation coefficient;

[0096] Step 3: DRL strategy training, collecting experience trajectories {s t ,a t ,r t ,s t+1}, calculate the advantage function A t =Q(s t ,a t )-V(s t ), update the policy network function: The truncation ratio ò=0.2 to prevent the policy from being updated too much; s t is the state vector, the current state of the environment, a tis the action vector, the decision action at the current moment, r t is the reward signal, the immediate benefit of the current action, s t+1 is the state at the next moment, Q(s t ,a t ) is the action value function, the expected cumulative reward of performing action a in state s, V(s t ) is the expected long-term value of the state value function in state s; θ policy The neural network weights generated for the policy network parameter control actions, such as the incinerator temperature adjustment policy; β represents the probability distribution;

[0097] Step 4: MOEA multi-objective optimization generates an initial solution covering the parameter space of incinerator temperature, excess air coefficient, flue gas temperature, and P2G power. Fast non-dominated sorting NSGA-II is used to extract the Pareto frontier, and the top 20% individuals are retained to enter the next generation.

[0098] Step 5: Online adaptive adjustment: Federated learning updates, aggregates local model gradients at the edge every 30 minutes, and updates the global strategy: where n k is the local data volume of the kth incineration plant, N is the total data volume; θ global represents the global angle, θ k represents the angle or local model parameter of the kth sample;

[0099] Step 6: If the number of iterations is reached or the termination condition is met, terminate the iteration; otherwise, go to step 4;

[0100] Step 7: Obtain Z and Q corresponding to the optimal fitness value, that is, maximize the comprehensive net benefit Z while minimizing the carbon emission intensity Q.

[0101] The present invention aims to achieve efficient and environmentally friendly waste treatment and green energy recovery through the regulation of core parameters. This invention enables efficient operation of waste incineration thermal power plants and maximizes the use of green energy, with significant economic and environmental benefits.

[0102] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with this technology to understand the content of the present invention and implement it accordingly, and they cannot limit the scope of protection of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the scope of protection of the present invention. Although the embodiments of the present invention have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for optimizing the operation balance of a waste incineration thermal power plant, characterized in that: The steps include: Step 1: Use high-precision sensors to monitor the calorific value of garbage, incinerator temperature, flue gas emission temperature, power generation efficiency, and plant power consumption rate in real time. Based on the real-time monitoring data, dynamically adjust the incinerator temperature, excess air coefficient, and flue gas treatment temperature to optimize the combustion process. Step 2: Install a CO2 collection device in the flue gas treatment system and combine it with P2G technology to combine the collected CO2 with H2 to generate CH4, which is reused as auxiliary fuel in the incineration process to achieve carbon recycling; Step 3: Establish an objective function that maximizes the comprehensive net benefit and minimizes the carbon emission intensity, which is composed of electricity sales revenue, carbon emission costs, auxiliary fuel purchase costs, flue gas treatment and P2G power operation costs; Step 4: Based on the unit price of electricity, CO2 emissions from waste incineration, flue gas treatment time per unit of waste, incinerator temperature, excess air coefficient, and flue gas temperature, a spatiotemporal coupled neural network (ST-GNN) model is established. A continuous action space is designed to impose safety constraints on the objective function, which is then converted into an immediate reward. Step 5: Introduce the improved MOEA collaborative optimization algorithm to perform dual-objective Pareto frontier analysis on maximizing comprehensive net benefits and minimizing carbon emission intensity, and obtain the optimization results. The improved MOEA collaborative optimization algorithm introduces a NSGA-III+MOPO hybrid framework. The first 30% of iterations use NSGA-III to generate the initial population, and the last 70% of iterations use MOPO to introduce DRL policy gradient update through multi-objective strategy optimization. Dynamic strategy adjustment is introduced in multi-objective optimization to balance the conflicting goals of maximizing comprehensive net benefits and minimizing carbon emission intensity. Step 6: Based on the final optimization results, continuously adjust and optimize the garbage calorific value, incinerator temperature, flue gas treatment temperature, and P2G equipment operating parameters to ensure that the system is always in the best operating state.

2. The operation balance optimization method of a waste incineration thermal power plant according to claim 1 is characterized in that: The objective function of maximizing the comprehensive net income and minimizing the carbon emission intensity, which consists of electricity sales income, carbon emission costs, auxiliary fuel purchase costs, flue gas treatment and P2G power operation costs, is as follows: Where: comprehensive net income Z, carbon emission intensity Q, R elec represents the unit price of electricity (yuan / kWh), η gen Indicates power generation efficiency, which is the amount of power generated per unit of garbage (kWh / ton), C fuel represents the auxiliary fuel cost (yuan / ton), η plant Indicates the power consumption rate of the plant, which is the proportion of the power consumption in the plant to the total power generation. emission represents the carbon emission cost (yuan / ton CO2), E CO2 Indicates the CO2 emissions from waste incineration, C operation represents the flue gas treatment and P2G power operating cost (yuan / hour), H treatment Indicates the flue gas treatment time per unit of garbage (hours / ton), D total Indicates the total amount of waste processed (tons).

3. The operation balance optimization method of a waste incineration thermal power plant according to claim 2 is characterized in that: Step 4 establishes an intelligent control model that deeply integrates the spatiotemporal coupled neural network ST-GNN model with deep reinforcement learning DRL, and processes the objective function and the collected parameters as follows: 1) Define the decision variable incinerator temperature adjustment range ΔT adj , excess air coefficient correction value Δα adj , P2G power regulation ratio ΔP P2G When the incinerator temperature rises, the combustion efficiency is improved, the power generation is increased, the power sales benefits are also increased, and the fuel consumption is increased. By optimizing the incinerator temperature, the power generation benefits and fuel costs are balanced and carbon emissions are reduced. When the excess air coefficient correction value Δα adj When it is increased, the combustion efficiency is improved, the heat loss is improved, and Δα is optimized adj Suppressing the formation of nitrogen oxides while avoiding thermal efficiency loss; when P2G power increases, CO2 conversion fuel increases, auxiliary fuel demand decreases, but P2G energy consumption increases, reducing carbon emission intensity through carbon recycling, while also balancing P2G electricity consumption; 2) Real-time collection of six-dimensional state data: s t =[T furnace ,α air ,T flue ,R elec (t),E CO2 (t),H treatment (t)], where T furnace ,α air ,T flue Respectively represent the incinerator temperature, excess air coefficient, flue gas temperature, R elec Indicates the unit price of electricity (yuan / kWh), E CO2 represents the CO2 emissions from waste incineration, H treatment It represents the flue gas treatment time per unit of garbage (hours / ton), and is standardized using a sliding window to eliminate dimensional differences. 3) Establish a spatiotemporal coupled neural network ST-GNN model to model the spatiotemporal dependencies of each unit in the incineration system: Among them, x v is the node v feature, including six-dimensional state data, N(v) is a set of spatial adjacent nodes, outputting the hidden state h v It is used to predict the dynamic response of the system. W1 is the spatial coupling weight of the neighbor node state, which is used to capture the spatial dependence between the incineration system units. W2 is the time coupling weight of the current node state, which is used to model the time dependence of the historical state. W3 is the global weight of the node feature. is the hidden state of node v in the lth layer, representing its spatiotemporal dynamic characteristics, represents the hidden state of the neighbor node u in layer l, and σ represents the activation function; 4) Action space design and constraint processing: design a continuous action space and impose safety constraints on the objective function: Among them, T min 、T max Indicates the lower and upper safety limits of the incinerator temperature, α min , α max Indicates the reasonable lower and upper limits of the excess air coefficient, P 2G,min 、P 2G,max Indicates the P2G device power and its regulation boundary; 5) Construct a reward function, convert the objective function into an immediate reward, and add an L2 regularization term to suppress oscillations: Where λ is the regularization coefficient.

4. The method for optimizing the operation balance of a waste incineration thermal power plant according to claim 3, characterized in that: Step 5 introduces the improved MOEA collaborative optimization algorithm to perform dual-objective Pareto frontier analysis on maximizing comprehensive net benefits and minimizing carbon emission intensity, as follows: 1) Introducing the improved MOEA collaborative optimization, the dual-objective Pareto frontier of maximizing net benefits and minimizing carbon emission intensity is performed; introducing the NSGA-III+MOPO hybrid framework, the first 30% of iterations use NSGA-III to generate the initial population, and the last 70% of iterations introduce DRL policy gradient update through MOPO, the formula is: Among them, A t is the advantage function, α is the learning rate, θ is the parameter of the policy network, a t Used to generate actions, such as the incinerator temperature adjustment amplitude, T is the total number of time steps, indicating the duration of a single incineration process, is the gradient of the policy network parameter θ, π θ is the policy function; 2) Probabilistic constraint modeling to quantify the uncertainty of carbon footprint. The formula is as follows: in, The carbon emission intensity target value, δ, represents the confidence level. A δ value of 0.05 indicates a 95 percent probability of meeting the constraint. Based on the assumption that historical waste calorific value fluctuation data is normally distributed, 1,000 scenarios were generated and carbon emission constraints were calculated with a confidence level of 1-δ. An increase in waste calorific value increases in incineration efficiency, energy generation per unit of waste η, and electricity sales revenue. A decrease in waste calorific value requires an increase in incineration temperature or excess air coefficient to maintain combustion, leading to increased fuel costs. 3) A hierarchical decision-making mechanism is adopted, and a lightweight DRL agent is used on the edge to perform local optimization in real time with a sampling frequency of 5Hz. The cloud uploads running data to the learning server every 30 minutes to update the global strategy parameters.

5. The operation balance optimization method of a waste incineration thermal power plant according to claim 4 is characterized in that: The steps of improving MOEA collaborative optimization algorithm are as follows: S1: Model the environment and use a one-stage delay differential equation to describe the combustion process: Where τ is the time constant, K is the gain coefficient, and ΔQ fuel is the fuel heat input, T setpoint Respectively represent the set temperature; S2: ST-GNN training: Input historical state sequence Σ=[s0,s1,...,s T ], output the predicted system response, which includes the power generation efficiency change Δη gen , construct the loss function in is the predicted change of power generation efficiency, γ is the weight attenuation coefficient; S3: DRL strategy training, collecting experience trajectories {s t ,a t ,r t ,s t+1 }, calculate the advantage function A t =Q(s t ,a t )-V(s t ), update the policy network function: The cut-off ratio Prevent policy updates from being too large; t is the state vector, the current state of the environment, a t is the action vector, the decision action at the current moment, r t is the reward signal, the immediate benefit of the current action, s t+1 is the state at the next moment, Q(s t ,a t ) is the action value function, the expected cumulative reward of performing action a in state s, V(s t ) is the expected long-term value of the state value function in state s; θ policy The neural network weights generated for the policy network parameter control actions, such as the incinerator temperature adjustment policy; β represents the probability distribution; S4: MOEA multi-objective optimization generates initial solutions covering the parameter space of incinerator temperature, excess air coefficient, flue gas temperature, and P2G power. Fast non-dominated sorting NSGA-II is used to extract the Pareto frontier, and the top 20% individuals are retained for the next generation. S5: Online adaptive adjustment: Federated learning updates, aggregates local model gradients at the edge every 30 minutes, and updates the global strategy: where n k is the local data volume of the kth incineration plant, N is the total data volume; θ global represents the global angle, θ k represents the angle or local model parameter of the kth sample; S6: If the number of iterations is reached or the termination condition is met, the iteration is terminated; otherwise, go to S4; S7: Get the Z and Q corresponding to the optimal fitness value, that is, maximize the comprehensive net benefit Z while minimizing the carbon emission intensity Q.

Images