Power plant fuel blending combustion optimization method based on multi-source data and deep learning

Abstract

The invention provides a power plant fuel blending combustion optimization method based on multi-source data and deep learning, and the method comprises the steps: detecting key characteristic indexes of incoming fuel, and constructing a dynamically updated fuel characteristic database; based on the fuel characteristic database, the boiler operation parameters and the load prediction data, a real-time fuel blending scheme is generated through a multi-objective optimization algorithm; performing a virtual combustion test on the real-time fuel blending scheme by using the three-dimensional combustion model of the boiler, and adjusting the opening degree of a secondary air door in the real-time fuel blending scheme; the method comprises the following steps: recording a data chain of fuel in a whole life cycle from factory entry to combustion through a block chain technology, generating a tamper-resistant carbon emission report, and when unconventional fuel is subjected to blending combustion, adapting combustion characteristics of the unconventional fuel through a transfer learning algorithm and generating a safe blending combustion proportion threshold value. And a real-time fuel blending scheme is generated and optimized through virtual testing, so that the fuel blending combustion optimization efficiency is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel blending and combustion optimization, and in particular to a power plant fuel blending and combustion optimization method based on multi-source data and deep learning. Background Art

[0002] In the operation of coal-fired power plants, fuel blending and combustion management are crucial to the economic and environmental benefits of the power plant. Traditional fuel blending methods rely primarily on manual experience, resulting in delayed response and difficulty in accurately adapting the blending plan to changes in boiler operating conditions. At the same time, the quality of incoming coal is often tested using offline spot checks, which makes it impossible to obtain dynamic coal quality data in real time, resulting in the inability to adjust the blending strategy in a timely manner. Furthermore, the combustion process lacks precise prediction and optimization methods, resulting in high coal consumption for power generation, unstable pollutant emissions, severe equipment wear, and difficulty in effectively utilizing low-calorific value coal and unconventional fuels, which does not meet the energy conservation, emission reduction, and intelligent development needs of power plants.

[0003] Therefore, how to improve the optimization efficiency of fuel blending and combustion in power plants has become a technical problem that technical personnel in this field urgently need to solve. Summary of the Invention

[0004] The present invention provides a power plant fuel blending and combustion optimization method based on multi-source data and deep learning, which is used to solve the defect of low efficiency of power plant fuel blending and combustion optimization in the prior art.

[0005] In a first aspect, the present invention provides a method for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning, comprising: Spectral sensor arrays deployed on coal conveyor belts detect key characteristics of incoming fuel in real time, building a dynamically updated fuel characteristics database. Based on the fuel characteristics database, boiler operating parameters and load forecast data, a real-time fuel blending plan is generated through a multi-objective optimization algorithm; Using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation, a virtual combustion test is performed on the real-time fuel blending scheme, and the secondary air damper opening in the real-time fuel blending scheme is dynamically adjusted according to the combustion state of the virtual combustion test; Blockchain technology is used to record the data chain of the entire life cycle of fuel from entry into the factory to combustion, generating an unalterable carbon emission report. When blending unconventional fuels, the transfer learning algorithm is used to adapt its combustion characteristics and generate a safe blending ratio threshold.

[0006] According to the present invention, a method for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning is provided. The method comprises: Deploy spectral sensor arrays in a staggered manner across the cross section of the coal conveyor belt; The spectral sensor array is used to synchronously collect calorific value, volatile matter and sulfur content data of the entire cross-section of the coal flow as key characteristic indicators.

[0007] According to the present invention, a power plant fuel blending and combustion optimization method based on multi-source data and deep learning is provided, wherein a real-time fuel blending scheme is generated by a multi-objective optimization algorithm based on the fuel property database, boiler operating parameters and load forecast data, including: Inputting a fuel property database, boiler operating parameters, and load forecast data into a long short-term memory network model, and constraining a search space of a multi-objective optimization algorithm based on the output of the long short-term memory network model; The search space is used to find the Pareto optimal solution set of power supply coal consumption, nitrogen oxide emissions, and equipment wear coefficient, and to generate a real-time fuel blending plan.

[0008] According to a power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided by the present invention, the search space of the output-constrained multi-objective optimization algorithm based on the long short-term memory network model includes: Analyzing the blending ratio boundary threshold output by the long short-term memory network; The blending ratio boundary threshold is set as the decision variable range of the multi-objective optimization algorithm to determine the search space.

[0009] According to a power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided by the present invention, the secondary air damper opening in the real-time fuel blending scheme is dynamically adjusted according to the combustion state of the virtual combustion test, including: Injecting the parameters of the real-time blending scheme into a computational fluid dynamics simulation environment, and outputting a nitrogen oxide generation rate change curve and a coking risk index; When the coking risk index exceeds the preset threshold, a secondary air door opening adjustment instruction is generated; When the NOx generation rate of change curve exceeds the allowable range, the staged air supply compensation logic is activated.

[0010] According to a power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided by the present invention, generating an unalterable carbon emission report includes: Connect the incoming fuel purchase data, real-time blending ratio, spectrum calibration value and actual emission value by timestamp; Writing the incoming fuel purchase data, real-time blending ratio, spectral verification value, and actual emission value into a Merkle tree node of the blockchain; In response to a regulatory query request for each of the Merkle tree nodes, a carbon emission report with a digital signature is output.

[0011] According to the present invention, a method for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning is provided. The method adapts the combustion characteristics and generates a safe blending ratio threshold through a transfer learning algorithm, including: Construct the physical property vectors of unconventional fuels, including calorific value, volatile matter, fixed carbon and sulfur content; Loading a weight matrix of a pre-trained boiler model based on the calorific value, volatile matter, fixed carbon, and sulfur content; Calculating a pollutant emission sensitivity gradient based on the weight matrix; Based on the pollutant emission sensitivity gradient, the safe blending ratio threshold is updated.

[0012] According to the present invention, a power plant fuel blending and combustion optimization method based on multi-source data and deep learning also includes: Receive the speed pulse signal generated by the encoder of the coal conveyor belt and synchronously trigger the sampling timing of the spectrum sensor array; Perform spectral data cleaning at the edge computing node to remove outliers caused by coal flow faults and compensate for moisture interference errors; The Kalman filter fuses the multi-sensor readings and outputs the sliding average of calorific value, volatile matter, and sulfur content; When the standard deviation of the sliding average exceeds the set threshold, a coal quality abnormality alarm instruction is sent to the DCS system.

[0013] According to the present invention, a power plant fuel blending and combustion optimization method based on multi-source data and deep learning also includes: Continuously compare real-time power supply coal consumption with dynamic thresholds; When the comparison result exceeds the tolerance deviation, the real-time fuel blending plan is regenerated and the virtual combustion test is performed again.

[0014] In a second aspect, the present invention provides a power plant fuel blending and combustion optimization system based on multi-source data and deep learning, comprising: The detection module is used to detect the key characteristics of incoming fuel in real time through a spectral sensor array deployed on the coal conveyor belt, and to build a dynamically updated fuel characteristics database; A generation module, configured to generate a real-time fuel blending plan through a multi-objective optimization algorithm based on the fuel property database, boiler operating parameters and load forecast data; a testing module configured to perform a virtual combustion test on the real-time fuel blending scheme using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation, and dynamically adjust the secondary air damper opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test; The optimization module is used to record the data chain of the entire life cycle of fuel from entry into the factory to combustion through blockchain technology, generate an unalterable carbon emission report, and when blending unconventional fuels, adapt its combustion characteristics through transfer learning algorithms and generate a safe blending ratio threshold.

[0015] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the above-described methods for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning.

[0016] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning.

[0017] In a fifth aspect, the present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-mentioned methods for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning.

[0018] The present invention provides a power plant fuel blending and combustion optimization method based on multi-source data and deep learning, comprising: real-time detection of key characteristic indicators of incoming fuel by a spectral sensor array deployed on a coal conveyor belt, and construction of a dynamically updated fuel characteristic database; generation of a real-time fuel blending scheme by a multi-objective optimization algorithm based on the fuel characteristic database, boiler operating parameters and load forecast data; a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation, a virtual combustion test of the real-time fuel blending scheme, and dynamic adjustment of the secondary air damper opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test; recording the data chain of the fuel from entry to combustion throughout its life cycle by blockchain technology, and generating an unalterable carbon emission report; when blending unconventional fuels, adapting its combustion characteristics by a transfer learning algorithm and generating a safe blending ratio threshold; since a real-time fuel blending scheme is generated based on fuel characteristics, boiler operating parameters and load forecast data, and optimized by virtual testing, the efficiency of fuel blending and combustion optimization is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1is a flow chart of a method for optimizing fuel blending and combustion in a power plant based on multi-source data and deep learning, provided in this embodiment; Figure 2 Schematic diagram of the structure of the power plant fuel blending and combustion optimization system based on multi-source data and deep learning provided by this embodiment; Figure 3 Schematic diagram of the structure of the electronic device provided in this embodiment. DETAILED DESCRIPTION

[0021] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0022] Figure 1 This is a flow chart of the power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided in this embodiment.

[0023] like Figure 1 As shown, the power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided by the embodiment of the present invention can be executed by the edge computing node and the power plant distributed control system (DCS) in collaboration, and mainly includes the following steps: 101. By deploying a spectral sensor array on the coal conveyor belt, the key characteristic indicators of the fuel entering the factory are detected in real time, and a dynamically updated fuel characteristic database is constructed.

[0024] In a specific implementation, an array of micro-spectral sensors is installed on the coal conveyor belts of a coal-fired power plant at specific spacing and layouts to ensure full coverage of the coal flow and enable real-time monitoring of coal quality. This array can monitor key indicators such as calorific value, volatile matter, and sulfur content of incoming coal in real time. The sensors transmit the detected data to the system, which establishes a dynamic fuel property database (coal quality database) to store, update, and analyze coal quality data in real time, providing basic data support for subsequent fuel blending and combustion optimization. Simultaneously, edge computing devices are deployed to perform preliminary processing and analysis of the sensor data, and to communicate and exchange data with the power plant's existing DCS system, ensuring stable and real-time data transmission.

[0025] 102. Based on the fuel characteristics database, boiler operating parameters and load forecast data, a real-time fuel blending plan is generated through a multi-objective optimization algorithm.

[0026] A multi-objective optimization algorithm was constructed that considers factors such as boiler thermal efficiency, pollutant emissions, and equipment wear, and combined it with an LSTM neural network to predict load changes. The system draws on coal quality data from a dynamic coal quality database, along with boiler operating parameters, load forecast data, and other multi-source information. Through the collaborative operation of the multi-objective optimization algorithm and the LSTM neural network, it generates the optimal blending solution, achieving a response time of minutes to meet the real-time operational needs of the power plant.

[0027] 103. A three-dimensional boiler combustion model constructed using computational fluid dynamics simulation is used to conduct a virtual combustion test on the real-time fuel blending scheme, and the secondary air damper opening in the real-time fuel blending scheme is dynamically adjusted according to the combustion state of the virtual combustion test.

[0028] A three-dimensional boiler combustion model is constructed using computational fluid dynamics (CFD) simulation technology. After determining the fuel blending scheme, the system uses this model to conduct virtual combustion tests before execution. By simulating the combustion process, key indicators such as coking tendency and NOx generation are predicted. Parameters such as the secondary air damper opening are automatically adjusted based on the predicted results, enabling pre-optimization of the combustion process and ensuring efficient and stable actual combustion.

[0029] 104. Blockchain technology is used to record the data chain of the entire life cycle of fuel from entry into the factory to combustion, generating an unalterable carbon emission report. When blending unconventional fuels, the transfer learning algorithm is used to adapt its combustion characteristics and generate a safe blending ratio threshold.

[0030] Specifically, lightweight blockchain technology is used to record data on each fuel batch, from its arrival at the plant to its combustion. This data includes, but is not limited to, fuel procurement information, coal quality testing data, blending process data, and combustion operation data. By ensuring the immutability of this data through blockchain technology, the system automatically generates carbon emissions reports, meeting environmental regulations and providing reliable data for internal plant management.

[0031] The system has been expanded to support the physical property database of unconventional fuels such as biomass pellets and sludge. When new fuel types need to be blended, the system uses a transfer learning algorithm to quickly learn and adapt the characteristics of the new fuel, and automatically generates a safe blending ratio threshold to ensure safe, stable operation and efficient combustion of the boiler when blending unconventional fuels.

[0032] By adopting the fuel blending and combustion optimization method implemented in this paper, through precise fuel blending and combustion optimization, the power coal consumption can be reduced by ≥2.1g / kWh, the power consumption of the coal mill can be reduced by 15%-20%, and the energy efficiency of the power plant can be effectively improved, thereby reducing production costs. The NOx fluctuation range can be suppressed to ≤±5mg / Nm³, ensuring that pollutant emissions are stable and meet standards. At the same time, the system automatically generates carbon emission reports to facilitate environmental protection supervision and help power plants achieve green development. The combustion process can be optimized to reduce equipment wear caused by incomplete combustion or unstable coal quality, extend the boiler heating surface cleaning cycle by 30%, and reduce equipment maintenance costs and downtime. It can effectively increase the utilization rate of low calorific value coal to more than 85%, and realize the safe and efficient blending of unconventional fuels such as biomass pellets and sludge, broaden the fuel source, and improve the comprehensive utilization rate of resources.

[0033] Furthermore, in this embodiment, a spectral sensor array deployed on the coal conveyor belt is used to detect key characteristic indicators of the fuel entering the factory in real time, including: staggered deployment of the spectral sensor array in the cross-section of the coal conveyor belt; using the spectral sensor array to synchronously collect calorific value, volatile matter and sulfur content data of the entire cross-section of the coal flow as key characteristic indicators.

[0034] Specifically, a spectral sensor array is deployed in a staggered arrangement in the cross-sectional direction of the coal conveyor belt, and multiple groups of sensor probes are set at equal intervals along the width of the belt. Each group of probes contains infrared spectral sensors, near-infrared spectral sensors, etc., forming an upper and lower cross-layer detection network to ensure that the entire cross-section is covered without blind spots when the coal flow passes through.

[0035] As the fuel is conveyed by the belt through the sensor array, each sensor simultaneously emits spectral signals (such as infrared and near-infrared light) and receives spectral signals reflected or transmitted by the fuel. Using a built-in spectral analysis module, the collected spectral data is preprocessed (e.g., filtering and noise reduction). By comparing it with a pre-set database of fuel characteristic spectra, the calorific value, volatile matter, and sulfur content of the entire cross-section of the coal flow are calculated in real time. The data collection frequency matches the belt's operating speed (e.g., 10 times per second) to ensure continuous capture of the dynamic coal flow's characteristic indicators.

[0036] The fuel property data obtained from the test is uploaded to the central control system in real time and automatically added to the fuel property database, forming a dynamic data set that is continuously updated over time, providing real-time data support for subsequent blending optimization.

[0037] The staggered sensor array enables full-section detection of the coal flow, avoiding the limitations of single-point sampling, improving data representativeness, and ensuring the accuracy of fuel characteristic indicators. Data acquisition frequency of one second can reflect fuel characteristic fluctuations in real time, providing an immediate data foundation for the dynamic generation of subsequent blending schemes, and shortening response time to seconds. The dynamically updated database combines historical data to form a fuel characteristic trend analysis, providing more accurate input parameters for the multi-objective optimization algorithm, and improving the adaptability of the blending scheme. This detection method is compatible with fuel characteristic analysis of different coal types and particle sizes. When the fuel type changes, the automatic update of the database can achieve self-adaptation of the detection model, reducing the cost of system recalibration.

[0038] Furthermore, in this embodiment, a real-time fuel blending scheme is generated through a multi-objective optimization algorithm based on the fuel characteristic database, boiler operating parameters and load forecast data, including: inputting the fuel characteristic database, boiler operating parameters and load forecast data into a long short-term memory network model, and constraining the search space of the multi-objective optimization algorithm based on the output of the long short-term memory network model; using the search space to solve the Pareto optimal solution set of power supply coal consumption, nitrogen oxide emissions, and equipment wear coefficient to generate a real-time fuel blending scheme.

[0039] Specifically, the fuel characteristics database, including real-time updated data such as calorific value, volatile matter, and sulfur content; boiler operating parameters (such as main steam temperature, pressure, and oxygen content); and future load data predicted based on historical load data are organized into a time series format and fed into a long-short-term memory (LSTM) model. The LSTM model learns from historical data to uncover temporal dependencies between data. The LSTM model trains and predicts based on the input data, outputting correlation trends between fuel characteristics, operating parameters, and load changes. This output is then converted into constraints on the feasible solution range for the multi-objective optimization algorithm, limiting the search space, reducing inefficient computation, and improving optimization efficiency. With the optimization objectives of minimizing coal consumption, nitrogen oxide emissions, and equipment wear coefficient, a multi-objective optimization algorithm, such as particle swarm optimization (PSO), is employed within the search space defined by the LSTM model to determine the Pareto optimal solution set for these three parameters. From this set of optimal solutions, a real-time fuel blending plan is selected based on the power plant's current environmental requirements, economic objectives, and other factors.

[0040] The LSTM model leverages the memory properties of time series data to accurately predict trends in fuel and operating parameters, narrowing the search space for multi-objective optimization and significantly improving computational efficiency. By solving for the Pareto optimal solution set, it effectively balances power coal consumption, nitrogen oxide emissions, and equipment wear. Compared to traditional single-objective optimization, this reduces overall costs and improves environmental compliance. Blending plans generated based on real-time data and predicted trends can quickly respond to changes in fuel characteristics and load, ensuring efficient, environmentally friendly, and safe boiler operation at all times, reducing manual intervention and improving automation.

[0041] Furthermore, the search space of the output-constrained multi-objective optimization algorithm based on the long short-term memory network model in this embodiment includes: parsing the blending ratio boundary threshold output by the long short-term memory network; setting the blending ratio boundary threshold as the decision variable range of the multi-objective optimization algorithm to determine the search space.

[0042] Specifically, the fuel characteristics and load trend data output by the long-short-term memory (LSTM) model are processed by the data analysis module. Using pre-set rules and statistical methods, the thresholds for fuel blending ratios under different operating conditions are extracted. For example, based on the LSTM-predicted fuel demand during high-load periods, combined with fuel characteristic data, the minimum and maximum blending ratios for each coal type are calculated.

[0043] The blending ratio thresholds obtained through analysis are directly set as the decision variable ranges for the multi-objective optimization algorithm. Specifically, the blending ratio variables for each coal type are constrained within the corresponding threshold range, forming a multi-dimensional search space for feasible solutions. For example, if the LSTM outputs a threshold for the blending ratio of a certain coal type as [10%, 40%], the blending ratio of that coal type will always fluctuate within this range during the multi-objective optimization algorithm.

[0044] Through historical data backtracking and simulation verification, the set search space is ensured to meet the boiler's operational safety and efficiency requirements. If verification finds that the boundary threshold may lead to operational anomalies (such as coking risk), the threshold range is automatically adjusted and the search space is rebuilt.

[0045] By pre-screening the range of feasible solutions using LSTM outputs, the search space for the multi-objective optimization algorithm is reduced, redundant computations are minimized, and solution generation time is shortened, meeting the real-time control requirements of the power plant. Boundary thresholds are set based on historical data and predicted trends to ensure that the optimization decision variables remain within a safe and efficient operating range, avoiding combustion instability or equipment failure caused by extreme blending ratios and reducing operational risks. The LSTM model continuously learns from real-time data, and boundary thresholds are dynamically adjusted based on fuel characteristics and load changes, ensuring that the search space is always adapted to the current operating conditions, improving the dynamic adaptability and universality of the blending solution.

[0046] Further, in the embodiment, the secondary air damper opening degree in the real-time fuel blending scheme is dynamically adjusted according to the virtual combustion test, including: injecting the parameters of the real-time blending scheme in the computational fluid dynamics simulation environment, outputting the nitrogen oxide generation rate curve and the coking risk index; when the coking risk index exceeds the preset threshold, generating a secondary air damper opening degree adjustment instruction; when the nitrogen oxide generation rate curve breaks through the allowed interval, activating the staged air supply compensation logic.

[0047] Specifically, in the computational fluid dynamics (CFD) simulation environment, the fuel components, blending ratio, and coal quantity in the real-time fuel blending scheme are imported, combined with the boiler structure parameters and operating conditions, and a three-dimensional combustion model is constructed. After the model runs, the nitrogen oxide (NOx) generation rate curve and the coking risk index are output in real time and transmitted to the control system through a data interface. The control system monitors the coking risk index in real time, and if the index exceeds the preset threshold (such as the critical value determined by historical data and experiments), the secondary air damper opening degree adjustment logic is triggered: according to the coking risk degree, a secondary air damper opening degree adjustment instruction is automatically generated according to the preset strategy (such as hierarchical incremental), which is sent to the actuator through the DCS system to increase or decrease the secondary air volume, improve the flow field distribution in the furnace, and reduce the coking risk. When the NOx generation rate curve exceeds the allowed interval (such as the upper limit of the environmental emission standard), the staged air supply compensation logic is activated: the system automatically adjusts the partition opening degree of the secondary air damper to optimize the oxygen distribution in the furnace and suppress NOx generation by strengthening the staged combustion. At the same time, the damper opening degree is dynamically corrected according to the NOx concentration feedback to ensure that the emissions meet the standards.

[0048] By monitoring the coking risk in real time and actively adjusting the secondary air damper, the probability of high-temperature region coking can be reduced, the boiler ash removal cycle can be extended, and the equipment maintenance cost can be reduced. Based on the dynamic adjustment of the staged air supply according to the NOx generation rate, the nitrogen oxide emission concentration fluctuation range can be reduced, the environmental standard compliance rate can be significantly improved, and the operation load of the denitration system can be reduced. The virtual combustion test results are linked with the secondary air damper control to realize closed-loop optimization of fuel blending and combustion adjustment, compared with traditional manual control, the boiler thermal efficiency is improved, and the fuel adaptability is enhanced.

[0049] Further, in the embodiment, the non-tamperable carbon emission report is generated, including: concatenating the in-plant fuel purchase data, real-time blending ratio, spectral verification value, and actual emission value according to the time stamp; writing the in-plant fuel purchase data, real-time blending ratio, spectral verification value, and actual emission value into the Merkle tree node of the blockchain; in response to a regulatory query request for each Merkle tree node, outputting a carbon emission report with a digital signature.

[0050] Specifically, the system collects real-time incoming fuel procurement data (source, quantity, calorific value, etc.), real-time fuel blending ratios, fuel property verification values ​​detected by spectral sensors, and actual emissions data obtained by online monitoring equipment, forming a data chain in timestamp order. For example, a data set is generated every 15 minutes, containing information about incoming fuel batches, blending plans, and corresponding emissions data for that period. This integrated data set is written as a separate unit into a Merkle tree node on the blockchain. Each data set is encrypted using a hashing algorithm to generate a unique hash value, which is then aggregated step by step to form a Merkle tree structure. Using distributed ledger technology, this data is synchronously stored within the power plant and at authorized regulatory nodes to ensure data immutability. When regulatory authorities initiate a query request, the system locates the Merkle tree node based on the timestamp or specific conditions and extracts the corresponding data set. Using public-private key encryption, the data set is digitally signed and a carbon emissions report containing the complete data chain and signature information is generated, ensuring data traceability and authenticity.

[0051] Blockchain's distributed storage and hash encryption make carbon emissions data difficult to tamper with or falsify. Compared to traditional databases, this improves data credibility and meets the stringent data authenticity requirements of regulators. The Merkle tree structure enables rapid data location and verification, reducing regulatory query response times to seconds. Compared to traditional database retrieval efficiency, this significantly reduces data audit costs. The complete data chain records the entire fuel lifecycle, from procurement to combustion, automatically generating regulatory-compliant carbon emissions reports, eliminating manual reporting errors, ensuring that power plant carbon emissions data complies with environmental regulations, and reducing compliance risks.

[0052] Furthermore, in this embodiment, the combustion characteristics are adapted through a transfer learning algorithm and a safe blending ratio threshold is generated, including: constructing a physical property vector of the unconventional fuel, including calorific value, volatile matter, fixed carbon and sulfur content; based on the calorific value, volatile matter, fixed carbon and sulfur content, loading the weight matrix of the pre-trained boiler model; based on the weight matrix, calculating the pollutant emission sensitivity gradient; based on the pollutant emission sensitivity gradient, updating the safe blending ratio threshold.

[0053] Specifically, a spectral sensor array detects key characteristics of unconventional fuels, extracting data on calorific value, volatile matter, fixed carbon, and sulfur content, and combining them into a multidimensional physical property vector. For example, the physical property vector of a biomass fuel is [calorific value 15MJ / kg, volatile matter 70%, fixed carbon 18%, sulfur 0.1%], which serves as the basic input data for the algorithm.

[0054] Call a boiler combustion prediction model (such as a neural network model) trained on conventional fuel data and retain its trained weight matrix. Input the physical property vectors of unconventional fuels into the model and use the existing weights to perform preliminary feature extraction on the new fuel characteristics, avoiding training the model from scratch and significantly reducing the amount of computation.

[0055] Based on the pre-trained model, data for unconventional fuels with varying blending ratios is input, the combustion process is simulated, and the resulting emissions of pollutants such as nitrogen oxides and sulfur dioxide are output. Using a backpropagation algorithm, the sensitivity gradients of various physical parameters to pollutant emissions are calculated, quantifying the impact of changes in fuel properties on emissions. For example, if a sulfur dioxide sensitivity gradient of 0.8 is found when the sulfur content of a fuel increases by 1%, this indicates a significant impact of sulfur on emissions.

[0056] Based on the results of sensitivity gradient analysis, combined with boiler operating safety standards and environmental emission limits, an iterative optimization algorithm dynamically adjusts the safe blending ratio threshold. For example, if the high sulfur content of an unconventional fuel causes the emission sensitivity to exceed the limit, the maximum blending ratio is automatically reduced and the new threshold is fed back to the fuel blending solution generation module to ensure a safe and controllable combustion process.

[0057] Transfer learning eliminates model retraining and reduces the time required to adapt to unconventional fuel characteristics from weeks to hours, significantly improving the flexibility of power plant fuel blending. Based on quantitative sensitivity gradient analysis, the impact of unconventional fuel blending on emissions and safety can be predicted in advance. Compared to setting empirical values, the accuracy of safe blending thresholds is improved, reducing the risk of coking and over-emissions. Automatically generating blending thresholds reduces manual experimental costs and supports rapid evaluation of multiple unconventional fuels, helping power plants expand fuel sources, reduce fuel procurement costs, and improve economic efficiency.

[0058] Furthermore, this embodiment also includes: receiving the speed pulse signal generated by the coal conveyor belt encoder to synchronously trigger the sampling timing of the spectral sensor array; performing spectral data cleaning at the edge computing node to eliminate outliers caused by coal flow faults and compensate for moisture interference errors; fusing multi-sensor readings through a Kalman filter to output the sliding average of calorific value, volatile matter, and sulfur content; when the standard deviation of the sliding average exceeds the set threshold, sending a coal quality abnormality alarm instruction to the DCS system.

[0059] Specifically, an encoder is installed on the coal conveyor belt to generate a real-time speed pulse signal, which is then connected to the edge computing node. The node calculates the belt speed based on the pulse frequency and uses this as a reference to synchronously trigger the sampling control circuit of the spectral sensor array. For example, when the encoder detects a belt speed of 1.5 m / s, spectral data collection is triggered every 10 cm to ensure full cross-section coverage of the coal flow. After receiving the raw data from the spectral sensor array, the edge computing node performs two steps: 1. By setting a normal fluctuation range for coal quality parameters (e.g., calorific value fluctuation of ±2 MJ / kg), it uses statistical methods to identify and delete abnormal data caused by coal flow faults; 2. Based on a pre-established moisture-spectral absorption relationship model, the raw spectral data is reverse-corrected to eliminate moisture content fluctuations that interfere with the detection of indicators such as calorific value and volatile matter.

[0060] Multiple spectral sensors are deployed to form a redundant detection network, and the output data from each sensor is fed into a Kalman filter. Based on the system's state transition equations and observation equations, the filter dynamically predicts and updates coal quality parameter estimates, outputting sliding averages (e.g., averaging 10 data sets) of calorific value, volatile matter, and sulfur content to improve data stability. The standard deviation of the sliding averages is calculated. If the standard deviation of a parameter (such as calorific value) exceeds a preset threshold (e.g., 0.5 MJ / kg), indicating abnormal coal quality fluctuations, the edge computing node immediately sends an alarm to the distributed control system (DCS), prompting operators to adjust the blending strategy.

[0061] Synchronous encoder sampling ensures data collection is perfectly aligned with belt operation, preventing missed detections. Kalman filter fusion reduces detection errors and narrows data fluctuations. A standard deviation threshold monitoring mechanism instantly triggers alarms when coal quality changes suddenly (such as a sudden drop in calorific value), improving response speed compared to manual inspections and ensuring boiler combustion stability. Outlier rejection and moisture compensation algorithms effectively eliminate interfering factors such as uneven coal flow and ambient humidity, increasing the pass rate of key characteristic indicators and providing reliable data support for blending optimization.

[0062] Furthermore, based on the above embodiment, this embodiment also includes: continuously comparing the real-time power supply coal consumption with the dynamic threshold; when the comparison result exceeds the tolerance deviation, regenerating the real-time fuel blending plan and re-performing the virtual combustion test.

[0063] Specifically, the system uses a machine learning algorithm to calculate and generate dynamic thresholds for real-time power coal consumption based on historical boiler operating data, current load forecasts, and fuel characteristics. Simultaneously, the DCS collects actual boiler power coal consumption data in real time, with a frequency tailored to operating conditions (e.g., every 5 minutes). This real-time coal consumption data is continuously compared against the dynamic thresholds, and the relative deviation between the two is calculated. For example, a tolerance of ±3% is set. If actual coal consumption exceeds this range, it is considered an abnormal operating condition, triggering a system response mechanism. When a coal consumption deviation outside the tolerance range is detected, the system automatically retrieves the latest fuel characteristics database, boiler operating parameters, and load forecast data, re-inputs them into the long-short-term memory network model and multi-objective optimization algorithm, and generates a new real-time fuel blending plan. The new plan parameters are then injected into a computational fluid dynamics simulation environment for virtual combustion testing to verify their impact on power coal consumption, NOx emissions, and equipment wear. If the test results do not meet expectations, the optimization and testing process is repeated until a blending plan that meets the requirements is generated.

[0064] Through real-time monitoring and automatic adjustments, fluctuations in power supply coal consumption can be minimized, ensuring boiler efficiency remains within the optimal range. This reduces fuel costs compared to traditional manual adjustments. An automatic comparison and solution regeneration mechanism deploys new solutions within 10 minutes of an abnormal coal consumption event, improving response time and effectively preventing efficiency drops caused by fluctuating coal quality or changing operating conditions. Combined with virtual combustion testing and verification, this system ensures that new solutions reduce coal consumption while also addressing environmental emissions and equipment safety. This creates a closed-loop control system of "monitoring-judgment-optimization-verification," enhancing system stability and reliability.

[0065] Based on the same general inventive concept, the present invention also protects a power plant fuel blending and combustion optimization system based on multi-source data and deep learning. The power plant fuel blending and combustion optimization system based on multi-source data and deep learning described below and the power plant fuel blending and combustion optimization method based on multi-source data and deep learning described above can be referenced to each other.

[0066] Figure 2 This is a structural diagram of the power plant fuel blending and combustion optimization system based on multi-source data and deep learning provided in this embodiment.

[0067] like Figure 2 As shown, this embodiment provides a power plant fuel blending and combustion optimization system based on multi-source data and deep learning, including: Detection module 201 is used to detect key characteristic indicators of incoming fuel in real time through a spectral sensor array deployed on the coal conveyor belt, and to build a dynamically updated fuel characteristic database; A generation module 202 is used to generate a real-time fuel blending plan through a multi-objective optimization algorithm based on the fuel characteristic database, boiler operating parameters and load forecast data; A testing module 203 is configured to perform a virtual combustion test on the real-time fuel blending scheme using a three-dimensional boiler combustion model constructed using computational fluid dynamics simulation, and dynamically adjust the secondary air damper opening in the real-time fuel blending scheme based on the combustion state of the virtual combustion test; The optimization module 204 is used to record the data chain of the entire life cycle of the fuel from entry into the factory to combustion through blockchain technology, generate an unalterable carbon emission report, and when blending unconventional fuels, adapt its combustion characteristics through the transfer learning algorithm and generate a safe blending ratio threshold.

[0068] Figure 3 Schematic diagram of the structure of the electronic device provided in this embodiment.

[0069] like Figure 3 As shown, the electronic device may include: a processor (processor) 310, a communication interface (Communications Interface) 320, a memory (memory) 330 and a communication bus 340, wherein the processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call the logic instructions in the memory 330 to execute a power plant fuel blending and combustion optimization method based on multi-source data and deep learning, the method including: real-time detection of key characteristic indicators of incoming fuel through a spectral sensor array deployed on a coal conveyor belt, and construction of a dynamically updated fuel characteristic database; based on the fuel characteristic database, boiler operating parameters and load forecast data, generating a real-time fuel blending scheme through a multi-objective optimization algorithm; using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation to perform a virtual combustion test on the real-time fuel blending scheme, and dynamically adjusting the secondary air damper opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test; using blockchain technology to record the data chain of the fuel from entry into the plant to combustion throughout its life cycle, and generating an unalterable carbon emission report; when blending unconventional fuels, adapting its combustion characteristics through a transfer learning algorithm and generating a safe blending ratio threshold.

[0070] Furthermore, the logic instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0071] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the power plant fuel blending and combustion optimization method based on multi-source data and deep learning provided by the above methods. The method includes: using a spectral sensor array deployed on a coal conveyor belt to detect key characteristic indicators of incoming fuel in real time, and constructing a dynamically updated fuel characteristic database; based on the fuel characteristic database, boiler operating parameters and load forecast data, generating a real-time fuel blending scheme through a multi-objective optimization algorithm; using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation to perform a virtual combustion test on the real-time fuel blending scheme, and dynamically adjusting the secondary air damper opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test; using blockchain technology to record the data chain of the fuel from entry to combustion throughout its life cycle, and generate an unalterable carbon emission report. When blending unconventional fuels, adapt its combustion characteristics through a transfer learning algorithm and generate a safe blending ratio threshold.

[0072] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having stored thereon a computer program, which, when executed by a processor, implements a power plant fuel blending combustion optimization method based on multi-source data and deep learning provided by the above method, the method comprising: detecting key characteristic indicators of incoming plant fuel in real time through a spectrum sensor array deployed on a coal belt conveyor, and constructing a dynamically updated fuel characteristic database; generating a real-time fuel blending scheme based on the fuel characteristic database, boiler operating parameters and load prediction data through a multi-objective optimization algorithm; using a three-dimensional combustion model of the boiler constructed by computational fluid dynamics simulation to perform virtual combustion testing on the real-time fuel blending scheme, and dynamically adjusting the secondary air damper opening degree in the real-time fuel blending scheme according to the combustion state of the virtual combustion testing; recording the data chain of the fuel from the incoming plant to the combustion full life cycle through the blockchain technology, generating an unalterable carbon emission report, and when blending unconventional fuel, adapting its combustion characteristics through a transfer learning algorithm and generating a safe blending ratio threshold.

[0073] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0074] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and necessary universal hardware platforms, and of course, can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in terms of contribution to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0075] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for optimizing fuel blending and combustion in power plants based on multi-source data and deep learning, characterized in that: include: Spectral sensor arrays deployed on coal conveyor belts detect key characteristics of incoming fuel in real time, building a dynamically updated fuel characteristics database. Based on the fuel characteristics database, boiler operating parameters and load forecast data, a real-time fuel blending plan is generated through a multi-objective optimization algorithm; Using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation, a virtual combustion test is performed on the real-time fuel blending scheme, and the secondary air damper opening in the real-time fuel blending scheme is dynamically adjusted according to the combustion state of the virtual combustion test; Blockchain technology is used to record the data chain of the entire life cycle of fuel from entry into the factory to combustion, generating an unalterable carbon emission report. When blending unconventional fuels, the transfer learning algorithm is used to adapt its combustion characteristics and generate a safe blending ratio threshold.

2. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 1 is characterized in that: The spectral sensor array deployed on the coal conveyor belt conveyor detects key characteristic indicators of the incoming fuel in real time, including: Deploy spectral sensor arrays in a staggered manner across the cross section of the coal conveyor belt; The spectral sensor array is used to synchronously collect calorific value, volatile matter and sulfur content data of the entire cross-section of the coal flow as key characteristic indicators.

3. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 1 is characterized in that: The method of generating a real-time fuel blending scheme based on the fuel property database, boiler operating parameters and load forecast data through a multi-objective optimization algorithm includes: Inputting a fuel property database, boiler operating parameters, and load forecast data into a long short-term memory network model, and constraining a search space of a multi-objective optimization algorithm based on the output of the long short-term memory network model; The search space is used to find the Pareto optimal solution set of power supply coal consumption, nitrogen oxide emissions, and equipment wear coefficient, and to generate a real-time fuel blending plan.

4. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 3 is characterized in that: The search space of the output-constrained multi-objective optimization algorithm based on the long short-term memory network model includes: Analyzing the blending ratio boundary threshold output by the long short-term memory network; The blending ratio boundary threshold is set as the decision variable range of the multi-objective optimization algorithm to determine the search space.

5. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 1 is characterized in that: The dynamically adjusting the secondary air valve opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test includes: Injecting the parameters of the real-time blending scheme into a computational fluid dynamics simulation environment, and outputting a nitrogen oxide generation rate change curve and a coking risk index; When the coking risk index exceeds the preset threshold, a secondary air door opening adjustment instruction is generated; When the NOx generation rate of change curve exceeds the allowable range, the staged air supply compensation logic is activated.

6. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 1 is characterized in that: The generation of an unalterable carbon emission report includes: Connect the incoming fuel purchase data, real-time blending ratio, spectrum calibration value and actual emission value by timestamp; Writing the incoming fuel purchase data, real-time blending ratio, spectral verification value, and actual emission value into a Merkle tree node of the blockchain; In response to a regulatory query request for each of the Merkle tree nodes, a carbon emission report with a digital signature is output.

7. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to claim 1 is characterized in that: The method of adapting the combustion characteristics and generating a safe blending ratio threshold through a transfer learning algorithm includes: Construct the physical property vectors of unconventional fuels, including calorific value, volatile matter, fixed carbon and sulfur content; Loading a weight matrix of a pre-trained boiler model based on the calorific value, volatile matter, fixed carbon, and sulfur content; Calculating a pollutant emission sensitivity gradient based on the weight matrix; Based on the pollutant emission sensitivity gradient, the safe blending ratio threshold is updated.

8. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to any one of claims 1 to 7, characterized in that: Also includes: Receive the speed pulse signal generated by the encoder of the coal conveyor belt and synchronously trigger the sampling timing of the spectrum sensor array; Perform spectral data cleaning at the edge computing node to remove outliers caused by coal flow faults and compensate for moisture interference errors; The Kalman filter fuses the multi-sensor readings and outputs the sliding average of calorific value, volatile matter, and sulfur content; When the standard deviation of the sliding average exceeds the set threshold, a coal quality abnormality alarm instruction is sent to the DCS system.

9. The power plant fuel blending and combustion optimization method based on multi-source data and deep learning according to any one of claims 1 to 7, characterized in that: Also includes: Continuously compare real-time power supply coal consumption with dynamic thresholds; When the comparison result exceeds the tolerance deviation, the real-time fuel blending plan is regenerated and the virtual combustion test is performed again.

10. A power plant fuel blending and combustion optimization system based on multi-source data and deep learning, characterized in that: include: The detection module is used to detect the key characteristics of incoming fuel in real time through a spectral sensor array deployed on the coal conveyor belt, and to build a dynamically updated fuel characteristics database; A generation module, configured to generate a real-time fuel blending plan through a multi-objective optimization algorithm based on the fuel property database, boiler operating parameters and load forecast data; a testing module configured to perform a virtual combustion test on the real-time fuel blending scheme using a three-dimensional boiler combustion model constructed by computational fluid dynamics simulation, and dynamically adjust the secondary air damper opening in the real-time fuel blending scheme according to the combustion state of the virtual combustion test; The optimization module is used to record the data chain of the entire life cycle of fuel from entry into the factory to combustion through blockchain technology, generate an unalterable carbon emission report, and when blending unconventional fuels, adapt its combustion characteristics through transfer learning algorithms and generate a safe blending ratio threshold.

Images