A coal consumption prediction management system for a thermal power plant

Abstract

The present application relates to the technical fields of thermal power generation and energy management, in particular to a coal consumption prediction management system for thermal power plants, comprising: a multi-source data sensing and alignment module: collecting external dispatching instruction prediction curves, fuel inventory and attribute financial data, and equipment physical operation condition data, and generating multi-source aligned data; a hysteresis characteristic prediction module: combining the multi-source aligned data to solve the energy conversion time delay curve and the heat loss rate corresponding to different fuel blending schemes; an operation target optimization module: performing reverse deduction with the minimum whole-process energy loss and material consumption as the objective function, and generating dynamic fuel consumption guidance curves and blending guidance curves; a risk quantification and management module: calculating the external dispatching default probability to generate financial risk early warning information, and adjusting the dynamic fuel consumption guidance curves; the present application converts the boiler thermal inertia into management parameters, reduces fuel waste, avoids grid assessment penalties, and exhibits good system operation safety.

Description

Technical Field

[0001] This invention relates to the field of thermal power generation and energy management technology, specifically to a coal consumption prediction and management system for thermal power plants. Background Technology

[0002] Coal consumption forecasting for thermal power plants is a traditional method for operation management and cost control of thermal power plants. It can predict the physical amount of coal required by the generating units by analyzing data such as the physical operating conditions of the equipment and fuel inventory. Under stable operating conditions, it has a certain degree of forecasting accuracy. With the development of the electricity market and grid dispatch, coal consumption forecasting has gradually extended from simple physical quantity statistics to complex closed-loop management of system operation. However, when forecasting and managing coal consumption in complex scenarios such as sudden changes in grid load, traditional methods are still limited to pursuing the accuracy of static physical tonnage forecasts, ignoring the thermal inertia and nonlinear combustion hysteresis effects generated by the internal thermodynamic processes of the boiler. This leads to asynchrony and spatiotemporal distribution differences between commercial management data and underlying physical data due to inconsistent data sampling frequencies. This not only easily leads to fuel waste due to coal quality fluctuations, but also to grid assessment penalties due to unit response lags, resulting in low efficiency in optimizing operating parameters and controlling abnormal risks in thermal power plants. Summary of the Invention

[0003] The purpose of this invention is to provide a coal consumption prediction and management system for thermal power plants, which solves the following technical problems: This addresses the technical obstacle in traditional thermal power plant management where coal consumption forecasting is limited to pursuing accuracy in static physical tonnage predictions while ignoring the thermal inertia and combustion hysteresis effects generated by the internal thermodynamic processes of the boiler. By reconstructing the data processing architecture, it achieves cross-domain integration from static physical quantity prediction to dynamic feedforward financial optimization, minimizing total operating costs and reducing grid default risk while utilizing physical hysteresis effects to smooth load.

[0004] The objective of this invention can be achieved through the following technical solutions: A coal consumption prediction and management system for thermal power plants includes: a multi-source data sensing and alignment module, used to collect external dispatch command prediction curves, fuel inventory and attribute management data, and real-time physical operating condition data of equipment, and to perform timestamp alignment processing on the external dispatch command prediction curves, fuel inventory and attribute management data, and real-time physical operating condition data of equipment to generate multi-source aligned data; The hysteresis feature prediction module is used to input multi-source aligned data into a preset nonlinear hysteresis proxy model to calculate the energy conversion time delay curve and heat loss rate corresponding to different fuel ratio schemes. The target optimization module is used to back-derive dynamic fuel consumption guidance curves and ratio guidance curves based on the energy conversion time delay curve and heat loss rate, with the goal of minimizing total operating cost. The target function is the goal function of minimizing energy loss and material consumption throughout the process, which is used to comprehensively evaluate the input and output of energy and materials in monetary terms.

[0005] The risk quantification and management module is used to calculate the probability of external dispatch default based on the dynamic fuel consumption guidance curve and the ratio guidance curve, generate financial risk early warning information, and dynamically adjust the fuel consumption guidance curve according to the operational anomaly risk early warning information.

[0006] As a further aspect of the present invention: the multi-source data perception and alignment module is specifically used to: extract the time series features of the external scheduling instruction prediction curve; Extract calorific value, moisture content, and ash content characteristics from fuel inventory and attribute management data; extract main steam pressure and main steam temperature characteristics from real-time physical operating condition data of equipment. Based on the timestamps of time series features, the calorific value features, moisture features, ash content features, main steam pressure features, and main steam temperature features are resampled and aligned to generate multi-source aligned data.

[0007] As a further aspect of the present invention: the nonlinear hysteresis proxy model includes a long short-term memory network layer and a thermodynamic mechanism layer; Among them, the long short-term memory network layer is used to extract the temporal dependencies in multi-source aligned data and generate temporal feature vectors; The thermodynamic mechanism layer is used to perform thermal inertia compensation calculations on time-series feature vectors based on a preset physical heat capacity equivalent model, and outputs energy conversion time delay curves and heat loss rates.

[0008] As a further aspect of the present invention: the objective function for minimizing energy loss and material consumption throughout the entire process includes a fuel procurement cost sub-function, a pollutant emission constraint sub-function, and a scheduling response deviation sub-function; Among them, the fuel procurement cost sub-function is obtained by multiplying the preset calorific value equivalence coefficient with the dynamic fuel consumption guidance curve and summing the results; The pollutant emission constraint sub-function is obtained by multiplying the preset pollutant emission penalty weights with the pollutant emission amount predicted based on multi-source aligned data; The scheduling response deviation sub-function is obtained by multiplying the preset response default judgment coefficient by the external scheduling default probability.

[0009] As a further aspect of the present invention: the target optimization module is specifically used to: construct an optimization space containing multiple candidate fuel ratio schemes; Substitute the candidate fuel ratio schemes into the nonlinear hysteresis proxy model to obtain the candidate energy conversion time delay curve and candidate heat loss rate for each candidate fuel ratio scheme. Substitute the candidate energy conversion time delay curves and candidate heat loss rates into the objective function that minimizes total operating costs to calculate the predicted total operating cost. The candidate fuel blending scheme with the lowest predicted total operating cost is selected as the blending guidance curve, and the corresponding consumption is used as the dynamic fuel consumption guidance curve.

[0010] As a further aspect of the present invention: when calculating the probability of external scheduling default, the risk quantification and management module is specifically used to: calculate the time phase difference between the energy conversion time delay curve and the external scheduling instruction prediction curve; Calculate the integral value of the time phase difference within the preset assessment time window; divide the integral value by the preset maximum tolerable hysteresis constant to obtain the external scheduling default probability.

[0011] As a further aspect of the present invention: when generating financial risk warning information, the risk quantification and management module is specifically used to: obtain a preset upper limit threshold and a lower limit threshold for default risk, wherein the upper limit threshold for default risk is greater than the lower limit threshold for default risk. If the probability of external scheduling default is greater than the upper limit threshold of default risk, a high-risk warning message is generated; if the probability of external scheduling default is less than the lower limit threshold of default risk, a safe operation confirmation message is generated. If the probability of external scheduling default is greater than or equal to the lower threshold of default risk and less than or equal to the upper threshold of default risk, then medium-risk monitoring information is generated.

[0012] As a further aspect of the present invention: when the risk quantification and management module adjusts the dynamic fuel consumption guidance curve based on the operational anomaly risk warning information, it is specifically used to: respond to the high-risk warning information, increase the proportion of high-calorific-value fuel according to a preset time advance before the load surge time point predicted by the external dispatch instruction, and generate a corrected dynamic fuel consumption guidance curve. In response to moderate risk monitoring information, maintain the dynamic fuel consumption guidance curve unchanged and increase the data collection frequency; in response to safe operation confirmation information, output execution instructions according to the dynamic fuel consumption guidance curve.

[0013] The beneficial effects of this invention are: 1. This invention extracts multi-dimensional features of external scheduling instructions, fuel attributes, and real-time operating conditions of equipment through a multi-source data perception and alignment module, and performs resampling and alignment based on the scheduling time axis. This mechanism effectively breaks the temporal misalignment between power plant commercial management data and underlying equipment physical data, eliminates the asynchronicity between multi-source heterogeneous data, provides a unified benchmark with strict physical causal relationship for subsequent accurate prediction and optimization decision-making, and enhances the system's anti-interference capability under complex operating conditions. 2. This invention utilizes a nonlinear hysteresis surrogate model to integrate the time-series fitting capability of artificial intelligence with the hard constraints of thermodynamic mechanisms. By quantifying the hidden physical hysteresis effect in the process of coal combustion and thermal energy conversion, the traditional static physical tonnage of coal is transformed into dynamic coal thermal financial equivalent, accurately characterizing the energy decay and time delay of different coal qualities under specific operating conditions. This solves the prediction bias problem caused by neglecting thermal inertia in traditional predictions, ensuring that the output results still conform to the dynamic laws when there are sudden changes in coal quality or drastic load fluctuations. 3. This invention utilizes a target optimization module to comprehensively quantify and optimize fuel procurement, environmental governance, and default risk costs with the goal of minimizing total operating costs. The system proactively evaluates the financial consequences of various fuel ratio schemes and automatically seeks the optimal solution between utilizing boiler thermal inertia to smooth loads and reducing procurement costs, thus realizing the transformation from static open-loop state prediction to dynamic feedforward closed-loop control command generation. This not only reduces fuel waste caused by coal quality fluctuations but also significantly reduces overall power generation costs without increasing hardware investment. 4. This invention constructs a risk quantification and management module, which transforms the underlying thermodynamic hysteresis risk into readable financial early warning indicators. By calculating the probability of default and implementing a graded early warning mechanism, the system can inject high-calorific-value fuel in advance before a sudden load surge by reverse deduction, and use the time difference to physically offset the response lag caused by thermal inertia. This intelligent closed-loop control method avoids grid dispatch default penalties and significantly improves the operational safety, operating condition adaptability, and process adaptability of thermal power plants under the scenario of sudden changes in grid load. Attached Figure Description

[0014] The invention will now be further described with reference to the accompanying drawings.

[0015] Figure 1 This is a schematic diagram of a module of a coal consumption prediction and management system for thermal power plants provided in an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0017] Please see Figure 1 As shown, a coal consumption prediction and management system for thermal power plants includes: a multi-source data sensing and alignment module, used to collect external dispatch command prediction curves, fuel inventory and attribute management data, and real-time physical operating condition data of equipment, and to perform timestamp alignment processing on the external dispatch command prediction curves, fuel inventory and attribute management data, and real-time physical operating condition data of equipment to generate multi-source aligned data; The hysteresis feature prediction module is used to input multi-source aligned data into a preset nonlinear hysteresis proxy model to calculate the energy conversion time delay curve and heat loss rate corresponding to different fuel ratio schemes. The target optimization module is used to back-derive dynamic fuel consumption guidance curves and ratio guidance curves based on the energy conversion time delay curve and heat loss rate, with the goal of minimizing total operating cost. The target function is the goal function of minimizing energy loss and material consumption throughout the process, which is used to comprehensively evaluate the input and output of energy and materials in monetary terms.

[0018] The risk quantification and management module is used to calculate the probability of external dispatch default based on the dynamic fuel consumption guidance curve and the ratio guidance curve, generate financial risk early warning information, and dynamically adjust the fuel consumption guidance curve according to the operational anomaly risk early warning information.

[0019] This embodiment provides a coal consumption prediction and management system for thermal power plants, which aims to solve the technical obstacle in traditional thermal power plant management where coal consumption prediction is limited to pursuing the accuracy of static physical tonnage prediction while ignoring the thermal inertia and combustion lag effects generated by the internal thermodynamic process of the boiler. This coal consumption prediction and management system for thermal power plants achieves cross-domain integration from static physical quantity prediction to dynamic feedforward financial optimization by reconstructing the data processing architecture and prediction objectives. Specifically, the coal consumption prediction and management system for thermal power plants includes a multi-source data perception and alignment module, a hysteresis feature prediction module, an operation target optimization module, and a risk quantification and management module. The multi-source data perception and alignment module is configured to collect external dispatch command prediction curves representing future grid demand business constraints, fuel inventory and attribute management data representing cost and potential input parameters, and real-time physical operating condition data of equipment representing the current unit thermal state boundary conditions. Since the above data comes from different systems with significant differences in sampling frequency and time base, the multi-source data perception and alignment module performs timestamp alignment processing on external dispatch instruction prediction curves, fuel inventory and attribute management data, and real-time physical operation data of equipment to generate multi-source aligned data, so as to break down the spatiotemporal barriers between power plant commercial management data and underlying equipment physical data, and provide a unified spatiotemporal base for subsequent model calculations. The hysteresis feature prediction module inputs the multi-source aligned data generated by the multi-source data perception and alignment module into the preset nonlinear hysteresis proxy model to calculate the energy conversion time delay curve and heat loss rate corresponding to different fuel ratio schemes. In this collaborative mechanism, different fuel ratio schemes are dynamically generated by the subsequent operation target optimization module when constructing the optimization space and fed forward to the hysteresis feature prediction module. The hysteresis feature prediction module, as the core computing engine, receives these ratio schemes and performs iterative calculations in combination with multi-source aligned data. This quantifies the nonlinear physical hysteresis effect hidden in the coal combustion and thermal energy conversion process, converts the physical quantity of coal into dynamic coal thermal financial equivalent, and characterizes the time difference and energy decay of different coal qualities from entering the furnace to generating an effective power under specific operating conditions. Based on this, the target optimization module uses the energy conversion time delay curve and heat loss rate as the objective function to back-derive the dynamic fuel consumption guidance curve and the ratio guidance curve, so that the coal consumption prediction and management system of thermal power plants can actively find the optimal mathematical solution between using boiler thermal inertia to smooth the load and reducing the total daily procurement cost. The risk quantification and management module calculates the probability of external dispatch default based on dynamic fuel consumption guidance curves and ratio guidance curves, generates financial risk early warning information, and transforms the underlying thermodynamic hysteresis risk into a financial early warning indicator readable by management based on the operational anomaly risk early warning information and dynamic fuel consumption guidance curve, thus forming a closed-loop control. Through the cascading mechanism and synergistic effect of the above-mentioned multi-modules, the system transforms the physical obstacle of boiler thermal inertia into a usable management parameter, reduces fuel waste caused by coal quality fluctuations and incomplete combustion, avoids grid assessment penalties caused by unit response lag, and demonstrates good operational safety, operating condition adaptability, and process adaptability under grid load change scenarios.

[0020] In a preferred embodiment of the present invention, the multi-source data perception and alignment module is specifically used to: extract the time series features of the external scheduling instruction prediction curve; extract the calorific value features, moisture features, and ash features from the fuel inventory and attribute management data; extract the main steam pressure features and main steam temperature features from the real-time physical operating condition data of the equipment; and resample and align the calorific value features, moisture features, ash features, main steam pressure features, and main steam temperature features based on the timestamp of the time series features to generate multi-source aligned data.

[0021] This embodiment is a further specification of the multi-source data perception and alignment module. It mainly focuses on the detailed description of the refined feature extraction and resampling processing mechanism when the influence mechanism of different dimensions of data features on combustion hysteresis is different and the data granularity is extremely different. The multi-source data perception and alignment module extracts the time series features of the external scheduling instruction prediction curve to obtain the trend and inflection point of future load changes. At the same time, it extracts the calorific value, moisture and ash features that directly determine the fuel ignition delay period and burnout rate from the fuel inventory and attribute management data. It also extracts the main steam pressure and main steam temperature features that characterize the current heat storage capacity inside the boiler from the real-time physical operation data of the equipment. To eliminate the asynchronicity of multi-source data, the multi-source data perception and alignment module uses the timestamp of time series features as the benchmark, that is, the time axis of power grid dispatch as the main axis, and resamples and aligns the calorific value features, moisture features, ash content features, main steam pressure features, and main steam temperature features to generate multi-source aligned data. The resampling process employs linear interpolation or zero-order hold methods based on data smoothness requirements to ensure that each frame of data input into the prediction model has a strict physical causal correspondence. Feature-level resampling alignment is performed based on the scheduling time series, effectively eliminating the dimensional misalignment problem caused by low-frequency business management data and high-frequency underlying physical condition data. This improves the anti-interference capability and accuracy of subsequent hysteresis prediction under complex conditions and verifies the robustness of this technical solution in multi-source heterogeneous data fusion scenarios.

[0022] In a preferred embodiment of the present invention, the nonlinear hysteresis proxy model includes a long short-term memory network layer and a thermodynamic mechanism layer; wherein, the long short-term memory network layer is used to extract the temporal dependencies in multi-source aligned data and generate a temporal feature vector; the thermodynamic mechanism layer is used to perform thermal inertia compensation calculation on the temporal feature vector based on a preset physical heat capacity equivalent model, and output the energy conversion time delay curve and heat loss rate.

[0023] This embodiment is a further specification of the internal topology of the nonlinear hysteresis agent model, aiming to overcome the industry pain points that pure data-driven neural networks are prone to deviations under unseen working conditions and that pure physical mechanism models are computationally intensive and cannot meet the needs of real-time prediction and management. The nonlinear hysteresis surrogate model comprises a long short-term memory network layer and a thermodynamic mechanism layer. The long short-term memory network layer is used to extract temporal dependencies in multi-source aligned data and generate temporal feature vectors. This process utilizes the fitting ability of neural networks to high-dimensional nonlinear data to quickly capture the implicit correlation between historical operations and the current state. The thermodynamic mechanism layer performs thermal inertia compensation calculations on the temporal feature vectors based on a pre-defined physical heat capacity equivalent model, and outputs the energy conversion time delay curve and heat loss rate. The physical heat capacity equivalent model here simplifies the complex boiler heating surface to a form with a specific time constant. With damping coefficient In the energy storage stage, the physical boundary constraints and compensation corrections of the feature vector output by the neural network are applied using the mechanism law; in order to realize the above thermal inertia compensation calculation, the physical heat capacity equivalent model extracts the dynamic gradient representing the change of input heat in the time series feature vector and uses it as the input of the first-order inertial stage. A specific time constant is dynamically adjusted based on the current main steam pressure and temperature characteristics. With damping coefficient The time constant It determines the delay length of heat transfer and the damping coefficient. The magnitude of heat attenuation is determined; the specific adjustment logic is as follows: a two-dimensional lookup table matrix of main steam pressure and time constant, and main steam temperature and damping coefficient is pre-established in the controller. When the real-time collected main steam pressure is lower than the rated pressure, the time constant is increased according to a mapping rule proportional to the pressure deviation to simulate the slow response when the boiler has insufficient heat storage. The mathematical expression of this mapping rule is: ; in, For a specific time constant; The reference time constant for rated operating conditions; This is the preset pressure deviation compensation coefficient, in units of... ; Rated main steam pressure; Real-time main steam pressure; Similarly, when the real-time collected main steam temperature is lower than the rated temperature, the damping coefficient is increased according to a mapping rule proportional to the temperature deviation to simulate the increased attenuation during heat transfer. The mathematical expression of this mapping rule is: ; in, The damping coefficient; The reference damping coefficient under rated operating conditions; This is the preset temperature deviation compensation coefficient, in units of... ; The rated main steam temperature; This refers to the real-time main steam temperature. The dynamic gradient is smoothed and phase-shifted in the time domain using a discretized first-order inertial differential equation; specifically, in the first... One time step; Calculate the transient heat output after compensation at the k-th time step. According to the thermal inertia compensation calculation formula: ; in, For time steps; This represents the transient heat output after compensation at the k-th time step. This refers to the transient heat output from the previous time step; is the damping coefficient, used to characterize the energy gain attenuation during heat transfer; Here is the fuel heat input value at the k-th time step; where, The calculation formula is: ; in, For the first The fuel mass flow rate entering the boiler at each time step; The lower heating value of fuel; It is obtained by linearly mapping the temporal feature vector output by the Long Short-Term Memory network layer through a fully connected layer; the formula realizes the physical simulation of thermal inertia through a first-order discretized hysteresis element; Linear mapping in fully connected layers refers to mapping the high-dimensional abstract feature vector output from the Long Short-Term Memory (LSTM) network layer to the nominal value of instantaneous fuel thermal energy input with energy dimensions through a weight matrix. The calculation process includes time constants. attenuation factor The specific calculation formula is as follows:

[0024] in, It is the attenuation factor; The system data sampling period is defined by this formula. The compensated transient heat output curve in the time domain is calculated using this formula. By comparing the low calorific value of the fuel at the input end with the transient heat at the output end, the energy conversion time delay curve and heat loss rate are clearly obtained. When the time sequence feature vector shows that the input heat increases in a step, the mechanism model uses a dynamically adjusted time constant to make the predicted output heat increase slowly according to an exponential law, thereby forcibly correcting the instantaneous abrupt output that may not conform to the thermodynamic laws that the long short-term memory network layer may produce at the physical boundary. The in-situ construction method, which integrates artificial intelligence time series prediction with thermodynamic mechanisms, retains the high-efficiency computing speed of neural networks while introducing physical laws as hard constraints. This allows the nonlinear hysteresis surrogate model to still output energy conversion time delay curves that conform to physical reality when facing boundary conditions such as extreme coal quality changes or drastic fluctuations in grid load, demonstrating strong adaptive capability for dynamic mismatch.

[0025] In a preferred embodiment of the present invention, the objective function for minimizing energy loss and material consumption throughout the entire process includes a fuel procurement cost sub-function, a pollutant emission constraint sub-function, and a scheduling response deviation sub-function; wherein, the fuel procurement cost sub-function of the objective function for minimizing energy loss and material consumption throughout the entire process is obtained by multiplying a preset calorific value equivalence coefficient and a dynamic fuel consumption guidance curve and summing the results; The objective function for minimizing energy loss and material consumption throughout the process includes a fuel thermal energy input pollutant emission constraint sub-function, which is obtained by multiplying a preset pollutant emission penalty weight with the pollutant emission amount predicted based on multi-source aligned data; and a scheduling response deviation sub-function, which is obtained by multiplying a preset response default judgment coefficient with the external scheduling default probability.

[0026] This embodiment is a further concretization of the mathematical structure of the objective function for minimizing total operating costs, serving as the basis for a multidimensional correlation mapping that transforms physical operating states into core financial indicators of the management system. The objective function for minimizing total operating costs includes sub-functions for fuel procurement costs, environmental governance costs, and default risk costs. Its calculation logic integrates costs from these three dimensions. The specific objective function formula is expressed as follows: ; in, This is a comprehensive evaluation index for the total energy loss and material consumption throughout the prediction cycle, with units representing its dimensions. ; For the time step index of the discretization; This refers to the total number of time steps within the future forecast period, for example, using 15-minute time steps, totaling [number] for the entire day. One time step; For the first The preset fuel unit price for each time step; Preset pollutant emission penalty weights; For the first Pollutant emissions at each time step; This is a preset default judgment coefficient for the response; The probability of external scheduling default; For the first The expected fuel consumption physical quantity corresponding to each time step, in tons; the fuel procurement cost sub-function is reflected as follows: ; The environmental governance cost sub-function is reflected as follows: ; To clarify the calculation boundaries, pollutant emissions The specific calculation formula is constructed as follows: ; in, For pollutant emissions; This refers to the physical quantity of expected fuel consumption. The comprehensive ash mass fraction under the current fuel blending scheme is extracted from multi-source aligned data; The comprehensive sulfur mass fraction under the current fuel blending scheme is extracted from multi-source aligned data; The emission conversion rate constant from ash to particulate matter is preset based on the current operating conditions of the dust removal and desulfurization equipment. The emission conversion rate constant from sulfur to sulfur dioxide is preset based on the current operating conditions of the dust removal and desulfurization equipment; The default risk cost subfunction is reflected as By specifying the time steps and the corresponding cumulative interval This formula strictly aligns the computational dimensions of each sub-item in the time domain; By constructing a comprehensive cost objective function comprising three sub-functions, the probability of default is directly transformed into a financial penalty term and incorporated into the optimization process. This enables the system to automatically conduct a quantitative comprehensive financial assessment between the penalty risk of using inferior coal and ensuring compliance with response standards by using high-quality coal, thereby improving the financial optimization efficiency of the total operating cost minimization objective function in complex market environments.

[0027] In a preferred embodiment of the present invention, the target optimization module is specifically used to: construct an optimization space containing multiple candidate fuel ratio schemes; substitute the candidate fuel ratio schemes into a nonlinear hysteresis proxy model to obtain the candidate energy conversion time delay curve and candidate heat loss rate corresponding to each candidate fuel ratio scheme; substitute the candidate energy conversion time delay curve and candidate heat loss rate into the total operating cost minimization objective function to calculate the predicted total operating cost; select the candidate fuel ratio scheme with the smallest predicted total operating cost as the ratio guidance curve, and use the corresponding consumption as the dynamic fuel consumption guidance curve.

[0028] This embodiment is a further specification of the specific execution logic of the running target optimization module; the running target optimization module constructs an optimization space containing multiple candidate fuel ratio schemes based on the available coal types in the current coal yard and the physical constraints of the coal feeder; Specifically, when constructing the optimization space, the target optimization module uses a preset ratio adjustment step size of 5% as the discrete interval to generate a discrete fuel ratio matrix as the optimization space within the hard physical upper and lower limits of the maximum co-firing ratio and minimum stable combustion load of the coal feeder. After the optimization space is established, the target optimization module substitutes the candidate fuel ratio schemes into the nonlinear hysteresis proxy model to obtain the candidate energy conversion time delay curve and candidate heat loss rate corresponding to each candidate fuel ratio scheme. In the actual operation data stream, the target optimization module sends the generated candidate fuel ratio scheme as a pre-input parameter to the hysteresis feature prediction module, which triggers the hysteresis feature prediction module to call the nonlinear hysteresis proxy model in combination with multi-source aligned data for calculation, thereby obtaining the candidate energy conversion time delay curve and candidate heat loss rate corresponding to each candidate fuel ratio scheme, ensuring the logical closed loop between the optimization command and the hysteresis prediction. The target optimization module substitutes the candidate energy conversion time delay curves and candidate heat loss rates into the objective function of minimizing total operating costs to calculate the predicted total operating cost. The target optimization module uses a heuristic search algorithm or a global traversal mechanism to select the candidate fuel ratio scheme with the smallest total operating cost prediction as the ratio guidance curve, and uses the corresponding consumption as the dynamic fuel consumption guidance curve. In this process, for example, when using particle swarm optimization algorithm as the heuristic search algorithm, each candidate fuel ratio scheme is regarded as a particle, and its position vector represents the ratio of each coal type. The fitness value is the total operating cost prediction value. In each iteration, the particle adjusts the search direction and step size of the allocation ratio based on its own historical optimal cost and the global optimal cost of the group. Specifically, in each iteration, the particle dynamically adjusts the search direction and update step size by calculating the gradient difference between the individual local optimal solution and the global optimal solution, based on its own historical optimal cost and the global optimal cost of the group, combined with preset inertia weights, learning factors and random numbers, thereby updating the current allocation ratio position to quickly converge to the optimal allocation ratio scheme with the minimum total operating cost. To ensure that particles do not violate the hard constraints of physical equipment during the optimization process, the target optimization module introduces a boundary reflection and penalty mechanism after each particle position update: if the proportion of a certain coal type calculated by a certain particle exceeds the maximum blending ratio of the coal feeder, the proportion of that dimension will be forcibly truncated to the boundary value of the maximum blending ratio, and the excess proportion will be redistributed according to the calorific value weight of other coal types in the current scheme. Simultaneously calculate the estimated heat generation after combination. If the estimated heat output Below the lower limit of heat capacity corresponding to the minimum stable combustion load Then, a large penalty term is imposed on the fitness value of the particle, that is, the penalty value is added to the total operating cost prediction value calculated based on the objective function. This penalty term is used to logically suppress candidate schemes that violate stable combustion constraints, and its calculation formula is as follows: ; in, The penalty value ensures the uniformity of the fitness value after the penalty calculation, forcing the particle to move away from this infeasible region in the next iteration; The maximum penalty coefficient constant is a monetary unit that converts the square of the difference in calorific value to the total operating cost forecast. To estimate the calorific value, the ratio of each coal type in the current iteration within the total number of coal types involved in the blending is multiplied by its corresponding baseline lower calorific value and then summed. This is the lower limit of heat capacity corresponding to the minimum stable combustion load; Specifically, the maximum penalty coefficient constant The formula for determining it is: ; in, The maximum penalty coefficient constant is used to ensure that the penalty value is significantly larger than the normal total operating cost forecast, effectively eliminating infeasible matching schemes. A preset safety margin factor is set, for example, a value between 1.5 and 2.0; The maximum single financial loss amount is obtained from historical operating data regarding financial losses caused by unplanned downtime due to insufficient heat generation. This represents the lower limit of heat capacity corresponding to the historical minimum stable combustion load. This optimization process upgrades the system from passive prediction to proactive prescription. By evaluating the final financial consequences of multiple coal blending schemes in advance, it can reduce the overall power generation cost by optimizing the spatiotemporal distribution of fuel ratios without increasing hardware investment. This embodiment demonstrates the high cost-effectiveness and process adaptability of the operation target optimization module in fuel scheduling strategy generation.

[0029] In a preferred embodiment of the present invention, the risk quantification and management module, when calculating the probability of external scheduling default, is specifically used to: calculate the time phase difference between the energy conversion time delay curve and the external scheduling instruction prediction curve; calculate the integral value of the time phase difference within a preset assessment time window; and divide the integral value by a preset maximum tolerable hysteresis constant to obtain the probability of external scheduling default.

[0030] This embodiment is a further specification of the core algorithm of the risk quantification and management module when calculating the probability of external dispatch default; the risk quantification and management module extracts the energy conversion time delay curve and the external dispatch command prediction curve, and calculates the time phase difference between the two, which represents the lag between the actual available power output and the grid dispatch command time difference; The risk quantification and management module calculates the integral value of the time phase difference within a preset assessment time window derived from the automatic generation control (AGC) command assessment cycle of the power grid, thereby quantifying the severity of the accumulated lag. The risk quantification and management module divides the integral value by a preset maximum tolerable hysteresis constant set according to the generator set nameplate parameters and historical grid connection protocols, thereby obtaining the normalized external dispatch default probability; let the instantaneous phase difference be... The probability of external scheduling failure The mathematical expression is: ; in, and These are the start and end times of the assessment time window, respectively. This is the preset maximum tolerance hysteresis constant; To clarify the physical dimensions and implementation logic of this probability calculation, when calculating the time phase difference, the system extracts the target load value required by the external scheduling instruction prediction curve at any specific moment, and finds the actual response timestamp corresponding to the actual output load reaching the same target load value on the energy conversion time delay curve. The absolute value of the time difference between the two is the instantaneous phase difference at that moment, and its dimension is s. Within the interval from the start to the end of the preset assessment time window, the instantaneous phase difference is continuously accumulated and integrated with respect to time. In the actual engineering discretization execution of the computer system, this integration operation is equivalently transformed into a discrete summation process that multiplies the instantaneous phase difference corresponding to each sampling point within the time window with the system data sampling period and then accumulates the results. Since the dimension of both the instantaneous phase difference and the integral time element is time, the dimension of the obtained integral value is the square of time. The use of the integral value with the dimension of the square of time here is intended to amplify the penalty weight for the unit's continuous lag in response to dispatch instructions through the second-order time accumulation effect, so that it can more sensitively reflect the risk of default. Furthermore, the setting of the preset maximum tolerance hysteresis constant is based on the penalty-free hysteresis limit specified in the historical grid connection protocol of the Automatic Generation Control (AGC) system. The specific value can be directly determined by extracting the product of the maximum response delay time allowed in the protocol, for example, set to 120 seconds, and the assessment time window length, for example, 15 minutes (900 seconds). For example, the preset maximum tolerance hysteresis constant is the square of 108,000 seconds. In this case, the dimension of the preset maximum tolerance hysteresis constant in the denominator is also the square of time. The integral value obtained above is divided by the preset maximum tolerance hysteresis constant to completely cancel out the dimension, ensuring that the external dispatch default probability value obtained by the final normalized calculation is a pure dimensionless probability value. This provides data support for subsequent management decisions that strictly adhere to physical laws and mathematical consistency, reflecting the technical advantages of the risk quantification and management module in cross-domain feature mapping.

[0031] In a preferred embodiment of the present invention, the risk quantification and management module, when generating financial risk early warning information, is specifically used to: obtain a preset upper limit threshold and a lower limit threshold for default risk, wherein the upper limit threshold for default risk is greater than the lower limit threshold for default risk; if the external scheduling default probability is greater than the upper limit threshold for default risk, then generate high-risk early warning information; if the external scheduling default probability is less than the lower limit threshold for default risk, then generate safe operation confirmation information; if the external scheduling default probability is greater than or equal to the lower limit threshold for default risk and less than or equal to the upper limit threshold for default risk, then generate medium-risk monitoring information.

[0032] This embodiment is a further specification of the classification and grading mechanism for generating financial risk early warning information by the risk quantification and management module; the risk quantification and management module obtains the preset upper limit threshold and the preset lower limit threshold of default risk pre-set by the power plant management personnel based on risk preferences and historical assessment data, and the upper limit threshold of default risk is greater than the lower limit threshold of default risk; The risk quantification and management module compares the calculation results in real time. If the probability of external dispatch default is greater than the upper limit threshold of default risk, it indicates that the current coal allocation plan is facing a breach of the critical threshold of penalty in the upcoming load change, and a high-risk warning message is generated. If the probability of external dispatch default is less than the lower limit threshold of default risk, it indicates that the unit's response capability is sufficient, and a safe operation confirmation message is generated. If the probability of external scheduling default is greater than or equal to the lower threshold of default risk and less than or equal to the upper threshold of default risk, then medium-risk monitoring information is generated. The introduction of a dual-threshold grading mechanism enables the risk quantification and management module to transform continuous probability values ​​into discrete classification and early warning information with clear management semantics. This filters out invalid alarm interference caused by minor fluctuations, reduces the monitoring and processing load of operators, and allows management to quickly focus on high-risk periods that endanger financial benefits. This verifies the robustness of the grading mechanism under complex operating conditions.

[0033] In a preferred embodiment of the present invention, when the risk quantification and management module adjusts the dynamic fuel consumption guidance curve based on the operational anomaly risk warning information, it is specifically used to: respond to high-risk warning information, increase the proportion of high-calorific-value fuel according to a preset time advance before the load surge time point predicted by the external dispatch instruction, and generate a corrected dynamic fuel consumption guidance curve; respond to medium-risk monitoring information, keep the dynamic fuel consumption guidance curve unchanged and increase the data acquisition frequency; and respond to safe operation confirmation information, output execution instructions according to the dynamic fuel consumption guidance curve.

[0034] This embodiment is a further specification of the risk quantification and management module forming closed-loop control instructions based on the dynamic fuel consumption guidance curve of the abnormal operation risk warning information. In response to the high-risk warning information, the risk quantification and management module increases the proportion of high-calorific-value fuel according to the preset time advance before the load surge time point of the external dispatch instruction prediction curve, and generates a corrected dynamic fuel consumption guidance curve to store heat for the boiler in advance. In the above execution logic, the reverse derivation process of the preset time advance is specifically to extract the maximum delay time required to reach the target load under the current working condition from the energy conversion time delay curve of the risk quantification and management module, and combine it with the pre-calibrated physical transmission delay time from the coal feeder to the burner conveyor belt, and add the two to obtain the total lag time as the preset time advance. At the moment when the load surge time indicated by the external dispatch command prediction curve is advanced by the preset time advance, the current proportion is modified according to the preset high-calorific-value coal compensation ratio coefficient, generating a revised dynamic fuel consumption guidance curve; its mathematical formula is expressed as follows: ; in, Compensation amount for high-calorific-value coal; The step amplitude of load surges in the prediction curve for external dispatch instructions; The lower heating value is the benchmark lower heating value for high-calorific-value coal; Boiler thermal efficiency coefficient; Boiler thermal efficiency coefficient These are empirical efficiency constants obtained in advance through unit performance tests under different load ranges; The preset unit conversion and coal consumption rate conversion constant is expressed mathematically as follows: ; in, Used to convert power units to mass units on an hourly scale; 3600 is the conversion factor between hours and seconds; 1000 is the conversion factor between tons and kilograms; The mechanical transmission efficiency constant is used; by... Placing it in the numerator ensures that the amount of compensation coal is proportional to the magnitude of the sudden increase in load, which conforms to the law of conservation of energy; Meanwhile, the risk quantification and management module responds to medium-risk monitoring information, keeps the dynamic fuel consumption guidance curve unchanged and increases the data acquisition frequency to strengthen the close monitoring of combustion status; specifically, the data acquisition frequency of the main steam pressure characteristics and main steam temperature characteristics in the real-time physical operation data of the equipment is automatically increased from once every 60 seconds under normal operating conditions to once every 10 seconds, in order to shorten the dead time of the feedback control cycle of the underlying distributed control system. When responding to the confirmation of safe operation, the risk quantification and management module determines that the current strategy is optimal and outputs execution instructions to the underlying distributed control system according to the dynamic fuel consumption guidance curve. For high-risk states, the advance compensation co-firing strategy uses the time difference to inject high-calorific-value fuel in advance, so that when the grid load suddenly increases, the boiler can complete the heat accumulation and release, which physically offsets the thermal inertia hysteresis and realizes an intelligent closed loop from risk detection to risk mitigation. This demonstrates the high adaptability and engineering practical value of the risk quantification and management module in dynamic feedforward control.

[0035] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A coal consumption prediction and management system for thermal power plants, characterized in that, include: The multi-source data perception and alignment module is used to collect external dispatch command prediction curves, fuel inventory and attribute management data, and real-time physical operating condition data of equipment, and to perform timestamp alignment processing on the external dispatch command prediction curves, the fuel inventory and attribute management data, and the real-time physical operating condition data of equipment to generate multi-source aligned data. The hysteresis feature prediction module is used to input the multi-source aligned data into a preset nonlinear hysteresis proxy model to calculate the energy conversion time delay curve and heat loss rate corresponding to different fuel ratio schemes. The target optimization module is used to back-derive, based on the energy conversion time delay curve and the heat loss rate, with the minimization of total operating cost as the objective function, to generate dynamic fuel consumption guidance curves and ratio guidance curves; the objective function is the objective function for minimizing energy loss and material consumption throughout the process, which is used to comprehensively evaluate the input and output of energy and materials in monetary terms. The risk quantification and management module is used to calculate the probability of external dispatch default based on the dynamic fuel consumption guidance curve and the ratio guidance curve, generate operation anomaly risk warning information, and adjust the dynamic fuel consumption guidance curve according to the operation anomaly risk warning information.

2. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, The multi-source data perception and alignment module is specifically used to: extract the time series features of the external scheduling instruction prediction curve; Extract the calorific value, moisture content, and ash content characteristics from the fuel inventory and attribute management data; Extract the main steam pressure and main steam temperature characteristics from the real-time physical operating data of the equipment; Based on the timestamp of the time series features, the calorific value feature, the moisture feature, the ash content feature, the main steam pressure feature, and the main steam temperature feature are resampled and aligned to generate the multi-source aligned data.

3. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, The nonlinear hysteresis proxy model includes a long short-term memory network layer and a thermodynamic mechanism layer; The long short-term memory network layer is used to extract the temporal dependencies in the multi-source aligned data and generate temporal feature vectors. The thermodynamic mechanism layer is used to perform thermal inertia compensation calculations on the time-series feature vector based on a preset physical heat capacity equivalent model, and output the energy conversion time delay curve and the heat loss rate.

4. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, The objective function for minimizing energy loss and material consumption throughout the entire process includes a fuel procurement cost sub-function, a pollutant emission constraint sub-function, and a scheduling response deviation sub-function. The fuel procurement cost sub-function is obtained by multiplying and summing the preset fuel unit price and the dynamic fuel consumption guidance curve; The objective function for minimizing energy loss and material consumption throughout the process includes a fuel thermal energy input pollutant emission constraint sub-function, which is obtained by multiplying a preset pollutant emission penalty weight by the pollutant emission amount predicted based on the multi-source aligned data. The scheduling response deviation sub-function is obtained by multiplying the preset response default judgment coefficient by the external scheduling default probability.

5. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, The target optimization module is specifically used to: construct an optimization space containing multiple candidate fuel ratio schemes; Substitute the candidate fuel ratio schemes into the nonlinear hysteresis proxy model to obtain the candidate energy conversion time delay curve and candidate heat loss rate for each candidate fuel ratio scheme. Substitute the candidate energy conversion time delay curve and the candidate heat loss rate into the objective function for minimizing total operating cost to calculate the predicted value of total operating cost; The candidate fuel blending scheme with the lowest predicted total operating cost is selected as the blending guidance curve, and the corresponding consumption is used as the dynamic fuel consumption guidance curve.

6. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, When calculating the probability of external scheduling default, the risk quantification and management module is specifically used to: calculate the time phase difference between the energy conversion time delay curve and the external scheduling instruction prediction curve; Calculate the integral value of the time phase difference within a preset assessment time window; divide the integral value by a preset maximum tolerable hysteresis constant to obtain the external scheduling default probability.

7. The coal consumption prediction and management system for thermal power plants according to claim 1, characterized in that, When generating financial risk warning information, the risk quantification and management module is specifically used to: obtain a preset upper limit threshold and a lower limit threshold for default risk, wherein the upper limit threshold for default risk is greater than the lower limit threshold for default risk. If the probability of external scheduling default is greater than the upper limit threshold of default risk, a high-risk warning message is generated; If the probability of external scheduling failure is less than the lower limit threshold of failure risk, then a safe operation confirmation message is generated. If the probability of external scheduling default is greater than or equal to the lower threshold of default risk and less than or equal to the upper threshold of default risk, then medium-risk monitoring information is generated.

8. The coal consumption prediction and management system for thermal power plants according to claim 7, characterized in that, When the risk quantification and management module generates the dynamic fuel consumption guidance curve based on the operational anomaly risk warning information, it is specifically used to: respond to the high-risk warning information, increase the proportion of high-calorific-value fuel according to a preset time advance before the load surge time point of the external dispatch instruction prediction curve, and generate a corrected dynamic fuel consumption guidance curve. In response to the moderate risk monitoring information, the dynamic fuel consumption guidance curve is kept unchanged and the data acquisition frequency is increased; In response to the safety operation confirmation information, an execution command is output according to the dynamic fuel consumption guidance curve.

Images