Thermoelectric decoupling and carbon capture collaborative optimization method

Abstract

The invention belongs to the technical field of energy and environmental protection, and particularly relates to a thermoelectric decoupling and carbon capture collaborative optimization method. Thermoelectric system parameters, carbon capture system parameters and external dynamic data are collected in real time, after data cleaning, normalization and fusion processing, a thermoelectric load prediction model and a carbon capture efficiency prediction model based on deep learning are constructed, and high-precision prediction is achieved. And a multi-target joint optimization model is further established, an optimal scheduling strategy is solved by adopting an improved particle swarm algorithm, and an instruction is issued through an industrial control system, so that dynamic collaborative distribution of heat energy resources between the thermoelectricity and carbon capture systems is realized. The system also introduces a closed-loop feedback mechanism, automatically triggers model retraining based on prediction deviation, and ensures long-term operation stability. According to the invention, deep coupling and intelligent cooperation of thermoelectric decoupling and carbon capture are realized, the energy consumption of carbon capture is obviously reduced, the energy utilization efficiency is improved by 15-25%, and the comprehensive operation cost is reduced by 10-18%.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of energy and environmental protection, and particularly relates to a heat-electricity decoupling + carbon capture collaborative optimization method. BACKGROUND

[0002] In the field of energy and environmental protection, the heat and power cogeneration system of traditional coal-fired power plants and high energy-consuming enterprises generally has the problem of close heat and power coupling and insufficient regulation flexibility. When the heat demand is prioritized, the power generation efficiency is easily restricted, and when the power generation benefit is focused on, it is difficult to adapt to the dynamic fluctuation of the heat load, resulting in low overall energy utilization efficiency. At the same time, the existing carbon capture system consumes a large amount of energy and operates independently of the heat and power system, which not only increases the energy consumption and economic cost of enterprises, but also may further affect the stability of the heat and power system due to energy competition. In addition, under the guidance of the "double carbon" goal, enterprises need to balance carbon emission control, energy utilization efficiency, system stability and economic cost, but the existing technology lacks a collaborative scheduling mechanism for heat and power decoupling and carbon capture, and it is difficult to intelligently optimize in combination with external dynamic factors such as real-time electricity price, carbon price, environmental temperature and production plan, resulting in weak adaptability of the system to external working condition changes, and the overall balance of energy utilization, carbon reduction effect and economy cannot be achieved. Therefore, a technical solution for collaborative optimization of heat and power decoupling and carbon capture is urgently needed. SUMMARY

[0003] The application aims to provide a heat and power decoupling + carbon capture collaborative optimization method, which uses the excess alternative heat load after the coal-fired power unit is decoupled to drive the carbon capture system, reduces the energy consumption of carbon capture, improves the carbon capture efficiency, reduces carbon emissions, combines an AI scheduling algorithm, dynamically adjusts the heat load distribution according to real-time heat and power demand, realizes efficient decoupling of heat and power, improves energy utilization efficiency, and reduces enterprise energy cost and carbon emission cost.

[0004] In order to achieve the above-mentioned goal, the technical solution adopted by the application is as follows:

[0005] A heat and power decoupling + carbon capture collaborative optimization method, comprising the following steps:

[0006] 1. Equipment modification stage

[0007] The heat and power cogeneration system, carbon capture system and supporting facilities of the existing coal-fired power plant or high energy-consuming enterprise are comprehensively modified to lay a hardware foundation for collaborative optimization:

[0008] Heat and power cogeneration system modification: optimize the existing heat and power cogeneration equipment, focus on improving the flexibility of heat and power output regulation, break the traditional strong coupling relationship between heat and power, and ensure that the "excess alternative heat load" can be stably generated to meet the heat and power demand of the system, and ensure that this part of the heat load can be allocated to the carbon capture system as needed.

[0009] Carbon capture system installation and modification: Install an efficient carbon capture system while reserving heat energy input interfaces at key locations in the system to ensure that the abundant alternative heat load can be smoothly connected to provide heat energy support for the carbon capture process.

[0010] Abundant alternative heat load delivery system construction: Lay special insulated pipelines, configure high-efficiency heat exchangers and automatic control valves, and build a heat load delivery channel connecting the thermal decoupling system and the carbon capture system. High-efficiency insulation materials are used to reduce heat loss during delivery, and automatic control valves are used to adjust the heat load delivery amount in real time. If excessive heat loss is found during subsequent operation, the heat loss can be controlled within a reasonable range by increasing the thickness of the insulation layer or shortening the pipeline delivery distance to ensure the heat energy demand of the carbon capture system. The formula for calculating the heat loss rate of the pipeline is as follows:

[0011]

[0012] In the formula: is the total heat loss of the pipeline in time , with the unit being kJ; is the thermal conductivity of the insulation material, with the unit being W / (m·K), which is determined according to the type of insulation material; is the pipeline length, with the unit being m; is the temperature of the abundant alternative heat load in the pipeline, with the unit being K; is the external environment temperature, with the unit being K; is the outer diameter of the insulation layer, with the unit being m; is the inner diameter of the pipeline, with the unit being m; is the heat load delivery time, with the unit being h;

[0013] Sensor system deployment: High-precision sensors are installed at key equipment such as boilers, steam turbines, heat exchangers, and key nodes such as adsorption towers, regeneration towers, and flue gas pipelines in the carbon capture system to collect system operation parameters in real time, providing data sources for subsequent data processing and model calculation.

[0014] AI scheduling system platform construction: An AI scheduling system platform is constructed, including a data server, a model calculation module, and a control instruction output interface. The data server is used to store real-time collected and historically accumulated data; the model calculation module is responsible for running prediction models and joint optimization models; and the control instruction output interface is used to convert optimization results into executable control instructions to realize the connection with the industrial control system.

[0015] 2. Real-time data collection and transmission

[0016] Based on the deployed sensor system and AI scheduling system platform, full-dimensional data collection and real-time transmission are carried out:

[0017] Two types of core system parameters and external dynamic data are collected. Among them, the thermal power system parameters include power generation, heat supply, steam pressure, steam temperature, fuel consumption and equipment running status, etc.; the carbon capture system parameters include CO2 capture amount, system energy consumption, adsorbent temperature and saturation, regeneration steam consumption, etc.; the external data covers environmental temperature, weather forecast, real-time electricity price, carbon price and enterprise production plan, such as phased capacity adjustment, heat demand change, etc.

[0018] Through industrial Ethernet or special wireless transmission module, all collected data are transmitted to the data center of AI scheduling system in real time, ensuring the timeliness and integrity of data transmission, and avoiding the influence of subsequent optimization calculation due to data delay or loss.

[0019] 3、Data preprocessing

[0020] The collected raw data is processed to eliminate interference factors and integrate multi-source data, providing high-quality data basis for model construction:

[0021] Data cleaning: The original data is screened by using outlier identification algorithm to eliminate abnormal values caused by sensor failure, transmission interference and other factors, ensuring the authenticity and reliability of the data.

[0022] Data normalization: The min-max normalization method is used to convert the original data of different magnitudes and units into standardized data of uniform range, eliminate the influence of data magnitude difference on model training, and ensure the accuracy of model calculation.

[0023] Multi-source data fusion: The thermal power system data, carbon capture system data and external data are integrated by using weighted fusion method. In the fusion process, according to the credibility of each data source, the corresponding weight is allocated, the weight of the data source with higher credibility is larger, and finally a unified data analysis model is constructed.

[0024] Data rule mining: Using big data analysis technology, the processed historical data is deeply mined to extract system operation rules, load change trend and system response characteristics, providing experience knowledge for subsequent model training.

[0025] 4、AI prediction model establishment

[0026] Two types of prediction models are constructed by using deep learning algorithm to realize accurate prediction of thermal power load and carbon capture efficiency, and the model performance is continuously optimized through online learning:

[0027] Thermal power load prediction model: A bidirectional long short-term memory network combined with an attention mechanism is used to construct the model. The input features of the model include historical thermal power load data, external temperature, and power grid load prediction values. The prediction time scale is divided into short-term and medium-term. The short-term prediction error is controlled at a low level through historical data training, and the medium-term prediction error is kept within a reasonable range to provide a prediction basis for thermal power system output adjustment.

[0028] Carbon capture efficiency prediction model: A gated recurrent unit model is used to construct the model. The input features include thermal energy input, absorbent concentration, flue gas parameters, etc. The model is trained to ensure high prediction accuracy, providing a reference for parameter adjustment of the carbon capture system.

[0029] After the model training is completed, new real-time operation data is continuously received for online learning and parameter adaptive adjustment. When the deviation between the predicted value and the actual value exceeds the set threshold for two consecutive collection periods, or when the system operating conditions change suddenly, such as fuel type replacement or equipment restart after maintenance, the model parameter update is automatically started to ensure that the model always matches the actual system operating state and maintains high prediction accuracy.

[0030] 5. Joint optimization model construction and solution

[0031] A joint optimization model is established with multi-objective optimization as the core. Intelligent algorithms are used to solve the optimal operation strategy to achieve multi-objective balance of the system:

[0032] Joint optimization model construction: The joint optimization model driven by rich alternative heat load for decoupling of carbon capture and heat power is constructed with the goals of maximizing energy utilization efficiency, minimizing carbon emissions, maximizing system stability, and minimizing economic cost. The model objective function uses a multi-objective weighted function form to convert the four objectives into a unified optimization index, as follows:

[0033]

[0034] Where: : Comprehensive energy utilization efficiency; : Carbon emissions per unit of electricity generation; : System stability index, defined as the inverse of the fluctuation rate of key parameters such as steam pressure and temperature; : Comprehensive operating cost; : The weight coefficients of each target are dynamically adjusted according to real-time operating conditions. For example, when the electricity price peaks, the weight of the "lowest economic cost" target is appropriately increased to ensure that the model can adapt to external market fluctuations and operating condition changes.

[0035] Model solution algorithm: An improved particle swarm optimization algorithm is used to solve the joint optimization model, with the speed and position update formula as follows:

[0036]

[0037] wherein: : the velocity and position of the i-th particle in the j-th generation; : the velocity and position of the i-th particle in the j-th generation; : the velocity and position of the i-th particle in the j-th generation; : inertia weight, linearly decreasing with iteration number; : learning factor, usually set as 2.0; : random number in [0, 1]; : individual optimal solution; : global optimal solution. The algorithm realizes the search of optimal solution by setting the velocity and position updating rules of particles: the inertia weight linearly decreases with iteration number, ensuring wide search range in the early stage and fast convergence speed in the later stage; the learning factor is set as a fixed value, guiding the particles to approach the individual optimal solution and the global optimal solution; the search randomness is introduced by random number, avoiding the algorithm falling into local optimum. Through multiple iterations, the optimal operation strategy meeting the requirements of multi-objective optimization is finally obtained.

[0038] 6. Optimization scheduling instruction issuing

[0039] The solution results of the joint optimization model are converted into executable control instructions, which are issued to each execution device through the industrial control system to realize the dynamic adjustment of system operation parameters: the AI scheduling system decomposes the optimal operation strategy into specific control instructions, including adjusting the operation load of the thermal power decoupling device (such as increasing or decreasing the boiler output, adjusting the steam turbine admission amount), adjusting the distribution ratio of the surplus alternative heat load (such as increasing or decreasing the heat load amount delivered to the carbon capture system), controlling the start-stop and operation parameters of the carbon capture system (such as the operation switching of the adsorption tower / regeneration tower, the adsorbent circulation amount adjustment), optimizing the heat input amount of the steam reheater, etc. The control instructions are issued to the corresponding execution devices (such as valves, pumps, fans, main equipment, etc.) through the industrial control system, ensuring that the devices adjust the operation state according to the instructions, realizing the dynamic collaborative distribution of heat energy resources between the thermal power system and the carbon capture system, and achieving the optimal operation state.

[0040] 7. Closed-loop optimization

[0041] Through the comparison analysis of actual operation results and predicted values, the model parameters are fed back and optimized, forming a closed loop of "prediction-optimization-execution-feedback-correction", and continuously improving the optimization effect of the system: the root mean square error and the average absolute percentage error are used as two indicators to quantitatively calculate the deviation of the thermal power load prediction result and the carbon capture efficiency prediction result from the actual operation data, and the matching degree of the model prediction accuracy and the actual operation state is intuitively reflected. When the thermal power load prediction error is continuously more than 3% for 2 collection periods, or the carbon capture efficiency prediction error is continuously more than 2% for 2 collection periods, the model parameter correction is automatically triggered; the correction process uses the small batch gradient descent method to iteratively adjust the hidden layer weight of the AI prediction model, and the sample batch size of each iteration is 50-100 groups of historical and real-time fusion data; the corrected model needs to pass the 3-period stability verification, and the thermal power load prediction error is stable ≤3% and the carbon capture efficiency prediction error is stable ≤2% for 3 consecutive collection periods, so as to replace the original model for subsequent optimization calculation; if it fails to pass the verification, it will be repeatedly iterated and adjusted until the stability requirement is met, ensuring that the model always maintains high prediction and optimization capability.

[0042] The present application has the following beneficial effects:

[0043] 1. Efficient utilization of thermal energy

[0044] The traditional carbon capture system needs to consume the power generation or heat supply energy of the thermal power system, resulting in secondary waste of energy; the present application directly drives the carbon capture system by the "surplus alternative heat load" generated by the thermal power decoupling device, so that the additional energy consumption of the carbon capture system is reduced by more than 60%, and the comprehensive energy utilization efficiency is improved by 15%-25% compared with the traditional independent system.

[0045] 2. Flexible thermal power output

[0046] The modified combined heat and power system breaks the strong coupling relationship, and can dynamically adjust the output according to the real-time heat and power demand, such as ensuring the power generation efficiency when giving priority to heat supply, adapting to the heat load fluctuation when focusing on power generation, combining the accurate prediction of the AI heat and power load prediction model, avoiding the problems of oversupply or undersupply, and further improving the operation efficiency of the heat and power system itself.

[0047] 3. Dynamically adapt to market fluctuations

[0048] The combined optimization model dynamically adjusts the weight according to the real-time electricity price and carbon price, such as increasing the "lowest economic cost" weight when the electricity price peaks, and strengthening the "carbon emission minimization" target when the carbon price is high, so as to avoid economic losses caused by fixed operation strategy in market fluctuations, and reduce the comprehensive operation cost of the system by 10%-18%. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The present application is a method flowchart. Detailed Implementation

[0050] The present invention will be further described in detail below with reference to specific embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example: Application case of a 300MW coal-fired cogeneration power plant

[0052] 1. Equipment Modification Phase

[0053] The existing combined heat and power (CHP) system will be upgraded by adding a low-pressure cylinder zero-output device and a heat storage tank to improve the decoupling capability of heat and power. This will enable the system to stably generate approximately 50 MWth of surplus alternative heat load while meeting basic heat and power needs. An amine-based carbon capture system will be installed, designed with a CO2 capture capacity of 10 t / h and a capture efficiency of ≥90%. A heat input interface will be added to the regeneration tower. A 60m long insulated pipe (0.3m inner diameter, 0.5m outer diameter of the insulation layer, using aluminum silicate insulation material with a thermal conductivity of 0.05 W / (m·K)) will be laid, along with a plate heat exchanger and an electric regulating valve. Temperature, pressure, and flow sensors will be deployed at key nodes such as the boiler, turbine, and adsorption tower, with a data sampling frequency of 1 time / second. An AI scheduling platform will be built, using a cloud server to deploy data storage and model calculation modules, enabling data interaction with the power plant's DCS system.

[0054] 2. Real-time data acquisition and transmission

[0055] Collect parameters for the thermal power system: power generation (200–300MW), heating capacity (0–100MWth), main steam pressure (8–12MPa), steam temperature (500–540℃), and coal consumption (approximately 100t / h). Collect parameters for the carbon capture system: CO2 capture rate, regeneration steam consumption (15–25t / h), and adsorbent saturation (0.2–0.8). Collect external data: ambient temperature, electricity price fluctuations (0.3–0.8 RMB / kWh), carbon price (50–100 RMB / ton), and production plan. Upload the data to the AI ​​scheduling platform in real time via industrial Ethernet, ensuring data transmission latency <1s.

[0056] 3. Data Preprocessing

[0057] An isolated forest algorithm was used to identify and remove outlier data points. Min-max normalization was used to unify various parameters to the [0,1] interval. The bi-peak characteristics (morning and evening peaks) of daily heat and power load were mined from historical data to provide prior knowledge for the prediction model.

[0058] 4. AI prediction model establishment

[0059] The thermoelectric load prediction model adopts a BiLSTM+Attention structure, the input includes past 24-hour load, external temperature, power grid dispatching instruction, and the output is 8-hour load prediction, short-term error ≤3%. The carbon capture efficiency prediction model adopts a GRU network, the input features are thermal energy input, flue gas CO2 concentration, and absorbent state, and the prediction accuracy is ≥98%.

[0060] 5. Joint optimization model construction and solution

[0061] The objective function is:

[0062]

[0063] The initial weights are set as w1=0.3, w2=0.3, w3=0.2, and w4=0.2, and are dynamically adjusted according to real-time electricity price and carbon price. An improved particle swarm optimization algorithm (population size 50, iteration 100 times) is used to solve, and the optimal strategy is obtained: 35 MWth excess heat load is allocated to the carbon capture system, and the power generation power is maintained at 260 MW.

[0064] 6. Optimization scheduling instruction issuance

[0065] The AI system generates instructions: adjust the boiler load to 85%, open the carbon capture heat load valve to 70%, and adjust the regeneration steam flow. The instructions are issued to the actuator through the DCS system to realize dynamic allocation of heat load.

[0066] 7. Closed-loop optimization

[0067] RMSE and MAPE are calculated every 15 minutes, if the thermoelectric load prediction RMSE or MAPE is greater than 3% for 2 consecutive periods, the model is modified. The model weights are updated using the small batch gradient descent method (batch size 80), and after modification, it needs to be verified for 3 periods (error stability ≤3%) before it can be used for subsequent optimization.

[0068] Implementation effect: After applying this method, the carbon capture system energy consumption of the power plant is reduced by 65%, the comprehensive energy efficiency is improved by 18%, the annual operating cost is reduced by about 15%, and the carbon emission intensity is reduced by 20%.

[0069] Finally, it should be noted that the above only describes the preferred embodiments of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent replacements to some technical features, as long as they are within the spirit and principles of the present application. Any modification, equivalent replacement, improvement, etc. made within the scope of the present application shall be included in the protection scope of the present application.

Claims

1. A thermoelectric decoupling + carbon capture synergistic optimization method, characterized in that: Includes the following steps: S1 Equipment Retrofit: Retrofit existing cogeneration systems in coal-fired power plants or high-energy-consuming enterprises, optimize the cogeneration decoupling equipment to enable flexible adjustment of cogeneration output and generation of "surplus alternative heat load"; install a high-efficiency carbon capture system and reserve heat input interfaces; construct a surplus alternative heat load transmission system, lay insulated pipes, configure heat exchangers and control valves, and connect the cogeneration decoupling system and the carbon capture system; install sensors at key nodes of the cogeneration system and the carbon capture system for real-time data collection; and build an AI scheduling system platform, including a data server, model calculation module, and control command output interface. S2 Real-time Data Acquisition and Transmission: Continuously acquires parameters of the thermoelectric system and carbon capture system through sensors, integrates external data, and transmits all data to the AI ​​scheduling system data center in real time; S3 data preprocessing: Cleaning, denoising, and normalizing the collected raw data, and removing outliers; integrating multi-source data to build a unified data analysis model, and using big data analysis technology to mine the operating patterns, load change trends and system response characteristics in historical data; S4 establishes AI prediction models: Deep learning algorithms are used to build thermoelectric load prediction models and carbon capture efficiency prediction models. The models are trained based on historical data and continuously receive new data for online learning and parameter adaptive adjustment to improve prediction accuracy. S5 constructs and solves a joint optimization model: a joint optimization model for carbon capture and thermoelectric decoupling driven by surplus alternative heat load is established, with the goal of "maximizing energy utilization efficiency, minimizing carbon emissions, maximizing system stability, and minimizing economic cost", and intelligent optimization algorithm is used to solve the optimal operation strategy; S6 issues optimized scheduling instructions: The AI ​​scheduling system converts the optimization results into specific control instructions, which are then sent to each executing device through the industrial control system. S7 closed-loop optimization: Compare the actual running results with the predicted values, calculate the deviation, and use the feedback data to retrain and correct the parameters of the AI ​​prediction model and the joint optimization model.

2. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S1, when excess alternative heat load is transported through the pipeline system and heat exchange device, high-efficiency insulation materials and automatic control valves are used to reduce heat loss. The formula for calculating the pipeline heat loss rate is as follows: ， In the formula: For the pipeline in time Total heat loss within the interior, expressed in kJ; The thermal conductivity of the insulation material is expressed in W / (m·K), and its value is determined based on the type of insulation material. This refers to the pipe length, in meters (m). The temperature of the excess alternative heat load inside the pipeline, expressed in K; The external ambient temperature of the pipeline, in Kelvin (K). The outer diameter of the insulation layer is in meters (m). The inner diameter of the pipe is expressed in meters (m). This refers to the heat load delivery time, expressed in hours (h). If heat loss is too high, increase the thickness of the insulation layer or reduce the pipeline transportation distance to ensure that heat loss is controlled within 10% and to guarantee the heat demand of the carbon capture system.

3. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S2, the parameters of the thermal power system include: power generation, heat supply, steam pressure, steam temperature, fuel consumption, and equipment operating status; the parameters of the carbon capture system include: CO2 capture amount, system energy consumption, adsorbent temperature and saturation, and regeneration steam consumption; external data include: ambient temperature, weather forecast, real-time electricity price, carbon price, and enterprise production plan.

4. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S3, data cleaning employs outlier detection algorithms to remove abnormal data, including the 3σ principle and the isolated forest algorithm. Data normalization uses the min-max normalization method, with the following formula: ， The original data, The minimum value of the data. For the maximum value of the data, The data is normalized; the multi-source data fusion adopts a weighted fusion method. During the fusion process, weights are assigned according to the credibility of each data source. Credibility is determined based on data acquisition frequency, sensor accuracy and historical error rate.

5. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S4, the thermal power load prediction model is constructed using a bidirectional long short-term memory network combined with an attention mechanism. The input features include historical thermal power load data, external temperature, and grid load prediction values. The prediction time scale is divided into short-term and medium-term, with short-term prediction error not exceeding 3% and medium-term prediction error not exceeding 5%. The carbon capture efficiency prediction model adopts a gated cyclic unit model. The input features include thermal energy input, absorbent concentration, and flue gas parameters, with a prediction accuracy of not less than 98%.

6. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S5, the objective function of the joint optimization model is a multi-objective weighted function, expressed as follows: ， in: Overall energy utilization efficiency; Carbon emissions per unit of electricity generated; System stability index, defined as the reciprocal of the volatility of key parameters such as steam pressure and temperature; Overall operating costs; .

7. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S5, the weight coefficients of the joint optimization model are dynamically adjusted according to real-time electricity price, carbon price, equipment operating status and environmental conditions to achieve adaptive optimization of multi-objective weights and ensure that the system maintains a balance between economic efficiency and environmental protection under external market fluctuations and changes in operating conditions.

8. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S5, the joint optimization model solution algorithm adopts an improved particle swarm optimization algorithm, and its velocity and position update formulas are as follows: ， In the formula: : No. The particle in the first The speed and position of the generation; Inertia weight, which decreases linearly with the number of iterations; Learning factor, usually set to 2.0; Random numbers in the interval [0,1]; Individual optimal solution; : Global optimal solution.

9. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S6, the control commands include: adjusting the operating load of the thermoelectric decoupling equipment, adjusting the distribution ratio of surplus alternative heat load, controlling the start-up and shutdown and operating parameters of the carbon capture system, and optimizing the heat input of the steam reheater, so as to realize the dynamic and coordinated distribution of thermal energy resources between the thermoelectric system and the carbon capture system.

10. The thermoelectric decoupling + carbon capture synergistic optimization method according to claim 1, characterized in that: In S7, the root mean square error and the mean absolute percentage error are used as dual indicators to quantify the deviations between the predicted thermal power load and carbon capture efficiency and the actual operating data. The calculation formulas are as follows: Root mean square error: In the formula The sample size is calculated for a single bias. These are the actual operating parameter values ​​of the system. For AI model predictions; Mean absolute percentage error: ; When the prediction error of thermal power load exceeds 3% for two consecutive acquisition cycles, or the prediction error of carbon capture efficiency exceeds 2% for two consecutive acquisition cycles, the model parameters are automatically corrected. The correction process uses the mini-batch gradient descent method to iteratively adjust the weights of the hidden layer of the AI ​​prediction model. The sample batch size for each iteration is 50-100 sets of historical and real-time fused data. The corrected model needs to pass a 3-cycle stability verification. If the prediction error of thermal power load is stable at ≤3% and the prediction error of carbon capture efficiency is stable at ≤2% for three consecutive acquisition cycles, it can replace the original model for subsequent optimization calculations. If the verification fails, the iterative adjustment is repeated until the stability requirements are met.

Images