A method and system for optimizing energy efficiency of a gas-steam boiler

Abstract

The application discloses a kind of gas steam boiler energy supply energy efficiency optimization method and system, it is related to boiler control technical field, including, real-time acquisition gas component data, boiler operating parameter is synchronously obtained, gas component data and boiler operating parameter are generated by data preprocessing standardization input matrix;Real-time operation data flow of boiler is obtained, input into the neural differential equation prediction model of training completion, the state prediction sequence in future time domain is calculated, and is transmitted to dynamic optimization controller;In dynamic optimization controller, based on state prediction sequence, construct rolling horizon optimization problem, convert control instruction set into physical operating quantity by actuator, adjust gas flow, combustion air volume and feed water flow, after completing boiler operating state update, actual operation data is fed back to constraint update process.The application significantly improves the real-time performance, adaptability and physical consistency of boiler energy efficiency optimization by dynamically coupling physical constraints and data-driven.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of boiler control, in particular to a gas-steam boiler energy supply efficiency optimization method and system. BACKGROUND

[0002] In recent years, the energy efficiency optimization method for gas-steam boilers has gradually developed from traditional PID control to data-driven and physical model fusion. In the prior art, neural network-based predictive control (such as LSTM and Transformer) can handle nonlinear dynamic characteristics, while physical models (such as heat balance equations and combustion dynamics equations) provide explainability constraints. In addition, the combination of genetic programming and neural differential equations shows potential in complex system modeling, such as mining implicit physical laws through symbolic regression. However, existing methods mostly use static constraints or pure data-driven methods, lack dynamic coupling mechanisms, and are difficult to adapt to real-time working conditions such as gas composition fluctuations and load changes.

[0003] The core defect of the prior art is the separation of physical constraints and data-driven models. In traditional methods, physical constraints are usually hard-coded into control algorithms in the form of fixed equations, which cannot be dynamically adjusted with changes in fuel characteristics. While pure data-driven models can fit dynamic characteristics, they are prone to generate false correlations that violate the laws of thermodynamics. In addition, the real-time updating efficiency of constraint conditions in the rolling horizon optimization is insufficient, leading to control delay or overshoot, especially when the gas composition fluctuates frequently. SUMMARY

[0004] In view of the above existing problems, the present application is proposed.

[0005] Therefore, the present application provides a gas-steam boiler energy supply efficiency optimization method to solve the problems of poor dynamic adaptability of physical constraints and real-time control delay in gas-steam boiler energy efficiency optimization.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a gas-steam boiler energy supply efficiency optimization method, which comprises: collecting gas composition data in real time, synchronously acquiring boiler operation parameters, generating a standardized input matrix through data preprocessing of the gas composition data and the boiler operation parameters; based on the standardized input matrix, mining physical constraint conditions through a genetic programming algorithm, generating a constraint equation set in differential algebraic form, encoding the constraint equation as a Lagrange multiplier term to inject into a neural differential equation, and constructing a neural differential equation prediction model with physical constraints; acquiring a real-time operation data stream of the boiler, inputting the real-time operation data stream into the trained neural differential equation prediction model, calculating a state prediction sequence in a future time domain, and transmitting the state prediction sequence to a dynamic optimization controller; in the dynamic optimization controller, constructing a rolling horizon optimization problem based on the state prediction sequence, converting a control instruction set into a physical operation quantity through an actuator, adjusting gas flow, combustion air volume, and feed water flow, and after updating the boiler operation state, feeding actual operation data back to a constraint update process.

[0008] As a preferred scheme of the gas-steam boiler energy supply efficiency optimization method, the gas composition data comprises volume percentage concentrations of methane, ethane, propane, carbon monoxide, and hydrogen; and the boiler operation parameters comprise steam pressure, flue gas temperature, oxygen content, feed water flow, and load rate.

[0009] As a preferred scheme of the gas-steam boiler energy supply efficiency optimization method, the constraint equation set comprises a heat balance equation, an oxygen content limit, and a tube wall temperature gradient constraint.

[0010] The constraint equation is encoded as a Lagrange multiplier term and injected into the neural differential equation, and the specific steps are as follows,

[0011] The heat balance equation, the oxygen content limit, and the tube wall temperature gradient constraint are input into a differential algebraic encoder, the partial derivatives of the constraint equation to the hidden state are calculated through a gradient field generation layer, and a constraint gradient tensor is output;

[0012] The constraint gradient tensor is dynamically coupled with the forward propagation result of the neural differential equation, and a constraint-enhanced gradient flow is generated through a Lagrange multiplier weighting layer;

[0013] The coupled gradient flow is input into an adjoint optimizer, joint backpropagation calculation is performed, and the neural network parameters and the Lagrange multiplier are synchronously updated;

[0014] The optimized gradient flow is input into a prediction corrector, and the heat balance equation, the oxygen content limit, and the tube wall temperature gradient constraint are forced to be satisfied simultaneously through an interior point method solver.

[0015] As a preferred scheme of the energy efficiency optimization method of the gas-steam boiler, the input layer of the neural differential equation prediction model receives real-time sensor data streams, the hidden layer describes the combustion dynamics process through a neural differential equation, the output layer generates state prediction values in a future time period, the state prediction values include thermal efficiency, nitrogen oxide concentration, and pipe wall stress distribution, the neural differential equation prediction model parameters are optimized based on the adjoint method during the training process, and the loss function fuses the prediction error and the constraint violation penalty.

[0016] As a preferred scheme of the energy efficiency optimization method of the gas-steam boiler, the control variables of the rolling horizon optimization problem include gas valve opening, fan speed, and feed water pump frequency, control instruction sets are generated in real time through an optimization algorithm, and the control instruction sets are issued to the actuator.

[0017] As a preferred scheme of the energy efficiency optimization method of the gas-steam boiler, the actual operation data refers to real-time working condition verification data after the gas-steam boiler executes the control instruction.

[0018] The real-time working condition verification data includes control response data, working condition verification parameters, and energy efficiency indicators.

[0019] As a preferred scheme of the energy efficiency optimization method of the gas-steam boiler, the constraint updating process refers to a closed-loop optimization process of dynamically correcting physical constraint conditions based on real-time operation data of the boiler.

[0020] In a second aspect, the present application provides an energy efficiency optimization system for a gas-steam boiler, comprising a data acquisition module, a constraint modeling module, a real-time prediction module, and an optimization control module. The data acquisition module is used to acquire gas composition data in real time, synchronously acquire boiler operation parameters, and generate a standardized input matrix from the gas composition data and the boiler operation parameters through data preprocessing. The constraint modeling module is used to mine physical constraint conditions through a genetic programming algorithm based on historical operation data and the standardized input matrix, generate a constraint equation set in differential algebraic form, encode the constraint equation as a Lagrange multiplier item to inject into a neural differential equation, and construct a neural differential equation prediction model with physical constraints. The real-time prediction module is used to acquire real-time operation data streams of the boiler, input the real-time operation data streams into the trained neural differential equation prediction model, calculate a state prediction sequence in a future time domain, and transmit the state prediction sequence to a dynamic optimization controller. The optimization control module is used to construct a rolling horizon optimization problem in the dynamic optimization controller based on the state prediction sequence, convert a control instruction set into a physical operation quantity through an actuator, adjust gas flow, combustion air volume, and feed water flow, complete an update of the boiler operation state, and feed actual operation data to a constraint updating process.

[0021] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the method for optimizing energy efficiency of a gas-steam boiler according to the first aspect of the present application.

[0022] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements any step of the method for optimizing energy efficiency of a gas-steam boiler according to the first aspect of the present application.

[0023] The present application has the following beneficial effects: based on genetic programming, differential algebraic constraint equations are dynamically mined to realize deep coupling between physical laws and data-driven models; constraint gradient tensors are dynamically injected by Lagrange multiplier weighting layers to significantly reduce constraint violation detection delay in combination with edge computing acceleration and buffering mechanisms; an improved hierarchical solving strategy realizes fast rolling optimization with dynamic constraints to improve control instruction response speed; reverse sign deduction verification of constraint equations ensures the physical rationality of prediction results; constraint equations are adaptively updated when gas composition suddenly changes to maintain prediction accuracy and control stability. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0025] Fig. 1 Flowchart of the method for optimizing energy efficiency of a gas-steam boiler.

[0026] Fig. 2 Coupling diagram of physical constraint modeling and neural differential equation in embodiment 1.

[0027] Fig. 3 Flowchart of dynamic rolling time domain optimization control in embodiment 1.

[0028] Fig. 4 Closed-loop feedback and constraint dynamic updating mechanism diagram in embodiment 1. DETAILED DESCRIPTION

[0029] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0030] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0031] Second, the "one embodiment" or "an embodiment" described herein as containing various features, structures, or characteristics can be combined with one or more other embodiments to form new embodiments, where the new embodiments do not include all of the features, structures, or characteristics of the one or more other embodiments.

[0032] Embodiment 1, refer to Figs. 1-4 The embodiment provides a gas-steam boiler energy supply efficiency optimization method, including the following steps:

[0033] S1, real-time acquisition of gas component data, synchronous acquisition of boiler operation parameters, generation of a standardized input matrix by data preprocessing of the gas component data and the boiler operation parameters.

[0034] Specifically, the method includes the following steps:

[0035] The gas components (volume percentage concentrations of methane, ethane, propane, carbon monoxide, and hydrogen) are collected by an online gas chromatograph (1 Hz), the boiler parameters (steam pressure, flue gas temperature, oxygen content, feed water flow, and load rate) are obtained by a distributed control system (DCS system), and a unified data format is defined.

[0036] The clocks of each device are calibrated based on the NTP protocol, and all data are time-stamped; linear interpolation is performed on the low-frequency data of the gas components (e.g., 5s / time for the chromatograph) to generate a 1Hz continuous sequence.

[0037] A ring buffer is set in the edge gateway to detect and remove abnormal values outside 3σ in real time, a sliding window (60 seconds) is used to establish a parameter baseline, linear interpolation is used to complete ≤3 consecutive missing points, and >3 points trigger an alarm and are filled with the previous valid value.

[0038] Specifically, when deploying a ring buffer at the edge gateway, a thread-safe ring queue structure with a fixed size is adopted, and independent buffer zones are allocated for gas composition data and boiler parameters respectively (gas data buffer capacity of 60 seconds, boiler parameter buffer capacity of 10 seconds). Pre-allocated memory and mutual exclusion lock mechanisms are used to ensure safe multi-thread access. The data acquisition thread writes sensor data to the head of the buffer zone in timestamp order, and automatically covers the oldest data and records an alarm when the buffer zone is full. The preprocessing thread reads data from the tail of the buffer zone at a frequency of 1 Hz, and automatically triggers linear interpolation compensation when it finds that the timestamps are not continuous. The buffer zone has a built-in integrity check mechanism that checks the monotonicity of the timestamp and the reasonableness of the value of each data packet. It also supports the function of continuing transmission after network interruption, and can maintain at least 10 seconds of data caching during network interruption, and after recovery, it can be retransmitted in timestamp order. To optimize performance, the memory mapping technique is used to reduce data copying overhead, and batch reading (10 seconds of window data at a time) is used to reduce thread switching frequency, ultimately achieving reliable data buffering with end-to-end processing delay ≤20 ms and data loss rate ≤0.1% within 10 seconds of network interruption. All abnormal events and buffer zone states are recorded in real time to a local log database for monitoring and analysis.

[0039] The Z-Score parameter (mean μ, standard deviation σ) is further calculated by a sliding window (60 seconds) to normalize the boiler parameters in real time; the gas concentration is first compressed in dynamic range by logarithmic transformation, and then Z-Score is executed.

[0040] The normalized gas composition data (5 dimensions) and boiler operating parameters (5 dimensions) are spliced into a matrix by timestamp, with each row representing a time (1 Hz) and each column corresponding to a feature, generating a standardized input matrix U:

[0041] U = [x1, x2, …, x 10 ] T , U ∈ R N×10 ;

[0042] In the formula, N represents the time step of the real-time sliding window (typical value 30).

[0043] S2, based on the standardized input matrix, a set of constraint equations in the form of differential algebra is generated by a genetic programming algorithm to mine physical constraints, and the constraint equations are coded as Lagrange multiplier terms to inject into the neural differential equation, and a neural differential equation prediction model with physical constraints is constructed.

[0044] Specifically, the following steps are included:

[0045] A symbol space containing a variable set, an operator set and a constant pool is defined according to the standardized matrix X.

[0046] Specifically, the variable set: {x1, x2, …, x 10};(corresponding to 10-dimensional features, such as x1 = methane concentration, x2 = steam pressure, etc.); operator set: contains differential integral (∫), four arithmetic operations (+, -, x, ÷), and nonlinear functions constant pool: fixed constants (0, 1, π, e) and random numbers (range [-5, 5]);

[0047] Randomly generate 200 initial equations, with tree structure (depth ≤ 6) for population initialization, providing a basis for candidate constraint equations for subsequent evolution, for example, candidate constraint equation F(x) is represented as:

[0048]

[0049] In the formula, represents the rate of change of methane (CH4) concentration over time (instantaneous change rate), d is the differential operator, which means differentiating the infinitesimal change of the variable, and dt represents the infinitesimal change of time (i.e., time differentiation), represents the nonlinear influence of steam pressure (prototype of heat balance equation), x3 is the exhaust gas temperature, and x4 is the oxygen content, represents the exhaust heat loss per unit oxygen content, which implies the balance between combustion efficiency and oxygen utilization (oxygen content limit prototype), 0 is the constraint nature, i.e., following the mass dynamic mass-energy conservation;

[0050] For each candidate constraint equation F(x), the constraint violation Fitness is calculated by traversing the historical data, which is represented as:

[0051]

[0052] In the formula, T represents the length of the historical data time window for calculating the fitness (total sampling points), F(x(t)) represents the value of the candidate constraint equation at time t, represents the normalization coefficient, which is used to eliminate the influence of window length, so that the fitness of different T can be compared;

[0053] It should be noted that the closer the fitness is to 0, the stronger the constraint of the candidate constraint equation on the data.

[0054] Furthermore, by defining selection, crossover, mutation, and termination conditions, the evolution iteration operation is performed.

[0055] Specifically, selection: tournament selection (select the best from 5 individuals each time); crossover: subtree exchange (probability 60%); mutation: node replacement / insertion / deletion (probability 30%); termination condition: evolution for 50 generations or optimal fitness change rate <1e-4.

[0056] Further, through the iterative optimization of evolutionary operations (selection, crossover, mutation), the equations in the population gradually converge to candidate solutions with higher fitness. At this time, the optimal individual of the final generation needs to be mathematically simplified to eliminate redundant structures and improve the interpretability of the equation.

[0057] Specifically, through mathematical equivalence simplification (such as merging like terms), the final output obtains a simplified differential-algebraic equation set, which is in the form of:

[0058]

[0059] In the formula, represents the kth constraint equation set (such as the heat balance equation, oxygen content limit, and pipe wall temperature gradient constraint), k represents the index of the constraint equation, represents the rate of change of the state vector with time, which is obtained through discrete difference or neural network output, and K represents the total number of effective constraint equations mined through genetic programming;

[0060] It should be noted that the candidate constraint equation is an initial equation set randomly generated by genetic programming. These initial equations may contain redundant structures or parts that do not conform to physical laws, and need to be screened and optimized through fitness evaluation and evolutionary operations. The constraint equation set is the final effective equation retained after strict screening, mathematical simplification, and physical verification from the candidate equation. They not only satisfy data consistency (high fitness) and mathematical simplicity (no redundant terms), but also must comply with engineering laws such as mass / energy conservation. Finally, they are injected into neural differential equations in the standard form C k = 0 to directly guide the neural differential equation prediction model training and prediction.

[0061] The core difference between the two is that the candidate equation is an intermediate product in the evolution process, with a large number and the need for further verification. The constraint equation set is a refined and reliable result, with a simplified number and clear physical meaning, which together constitute a complete process from data-driven to physical law fusion.

[0062] The physical constraint equation (heat balance equation, oxygen content limit, and pipe wall temperature gradient constraint) is input into the differential-algebraic encoder, and the partial derivative of the constraint equation to the hidden state is calculated through the gradient field generation layer and outputs the constraint gradient tensor (dimension consistent with the hidden state);

[0063] The constraint gradient tensor is dynamically coupled with the forward propagation result of the neural differential equation, and the constraint-enhanced gradient flow is generated through the Lagrange multiplier weighting layer, which is represented as:

[0064]

[0065] In the formula, f θ(X,t) represents the original state derivative predicted by the neural network (with parameter θ). This indicates taking the partial derivative of the constraint equation with respect to X. Represents the k-th constraint equation The total differential, Let μ be the differential of the state vector X, representing the infinitesimal change in the state. k For Lagrange multipliers;

[0066] Perform joint backpropagation to synchronously update the neural network parameters θ and Lagrange multipliers μ. k ;

[0067] Forced to satisfy the condition using the interior point method solver (For example, the heat balance equation, oxygen content limit, and pipe wall temperature gradient constraint all hold simultaneously)

[0068] A constraint-enhanced gradient flow is generated using Lagrange multiplier weighted layers to ensure that the state derivative satisfies physical constraints. A neural differential equation network architecture is constructed, consisting of a 4-layer fully connected network (64 hidden layers, Swish activation function), with the input being... The output is The constraint gradient term is dynamically calculated during forward propagation.

[0069] 1000 time windows (60 seconds of data per window) are randomly sampled from the normalized matrix U;

[0070] The loss function is defined as follows:

[0071]

[0072] In the formula, X pred X(t) represents the predicted state vector (10-dimensional) at time t. true (t) is the true state vector at time t (from historical data), |||| 2 This represents the Euclidean norm (L2 norm);

[0073] The AdamW optimizer (learning rate 3e-4, weight decay 1e-5) is used, and the parameters are dynamically updated based on the adjoint method.

[0074] Dynamically update Lagrange multipliers:

[0075]

[0076] In the formula, ρ (n) This represents the global constraint penalty coefficient, which increases with the number of training epochs n. This is a truncation function, indicating that... The range is restricted to the interval [-1, 1].

[0077] Specifically, the truncation function is represented as:

[0078]

[0079] More preferably, by defining the truncation function, it is possible to prevent a sharp violation of a single constraint (such as a sensor failure leading to a sudden change) from triggering a sharp fluctuation in μ k , thereby forcing the constraint to be satisfied.

[0080] If the violation of a certain constraint decreases by <5% in 10 consecutive iterations, its weight is decayed by 50%;

[0081] Calculate the average violation of each constraint equation on the test set, and compare the violation reduction ratio of the constraint model and the baseline model.

[0082] Calculate the prediction mean square error (MSE) and mean absolute percentage error (MAPE) of key parameters such as steam pressure and flue gas temperature.

[0083] It should be noted that the average violation of the constraint equation on the test set is calculated and the violation reduction ratio of the constraint model and the baseline model is compared to verify the correction effect of the physical constraint on the prediction result (such as the heat balance deviation decreasing from 15% to 3%), and to ensure that the model follows the engineering rules; the calculation of the MSE and MAPE of the key parameters verifies the numerical accuracy (such as the flue gas temperature error within ±2℃). These two steps form a double check: the former guarantees physical reasonableness, and the latter guarantees data accuracy, which together provide a reliable foundation for subsequent "symbolic backstepping verification" - only constraint equations with low violation and low error can pass the symbolic backstepping verification of their consistency with the principles of combustion reactions (such as the relationship between oxygen content and flue gas temperature in the constraint equation conforming to the heat loss law of combustion), and ultimately prove that the constraint equations generated by the algorithm truly reflect the physical mechanism of the boiler, rather than false associations caused by data noise, thereby ensuring the engineering credibility of the prediction model in control optimization.

[0084] Symbolic backstepping is performed on the constraint equation , and its physical meaning is verified in combination with the chemical reaction equation of boiler combustion (such as the correlation between methane consumption rate and oxygen supply), to ensure compliance with engineering rules such as heat balance equation and oxygen content limit, for example:

[0085]

[0086] It should be noted that this verification step ensures that the data-driven constraint equation is consistent with the physical and chemical principles of the combustion process, avoiding the generation of false relationships that are mathematically reasonable but physically incorrect, thereby improving the engineering credibility of the neural differential equation prediction model and confirming that the constraint equation indeed reflects the real physical mechanism in the boiler operation, rather than just a data fitting result.

[0087] S3, obtaining a real-time operation data stream of the boiler, inputting into the trained neural differential equation prediction model, calculating a state prediction sequence in a future time domain, and transmitting to a dynamic optimization controller.

[0088] Specifically, the following steps are included:

[0089] A lightweight data synchronization service is deployed on the edge gateway, which subscribes to real-time boiler operation parameters (steam pressure, exhaust gas temperature, etc.) from the DCS system through the OPC UA protocol, and receives gas composition data from the online gas chromatograph through the Modbus TCP protocol.

[0090] Based on the timestamp, a sliding time window (window length 30 seconds, step 1 second) is used to dynamically align the sampling time deviation of the gas composition data and the boiler operation parameters, eliminating the timing misalignment problem caused by sensor response delay (such as oxygen content sensor lag 2 seconds).

[0091] The trained neural differential equation prediction model is deployed on an edge computing node (such as NVIDIA Jetson AGX Xavier), and the model is converted to an inference optimization version with FP16 precision through the TensorRT engine, with a single prediction time consumption ≤50ms.

[0092] Specifically, after the neural differential equation prediction model is trained, the neural differential equation prediction model is exported to ONNX format, and the dimensions and dynamic axis parameters of the input and output tensors are specified. The ONNX parser of TensorRT is used to load the model, and a TRTBuilder instance is created on the NVIDIA Jetson AGX Xavier edge computing node, and the optimization configuration file is set to define the size range of the input and output tensors. Enable FP16 precision mode and add layer fusion optimization, apply directional optimization for LSTM units to ensure numerical stability. Call the Builder.buildEngine method to generate an optimized inference engine, set the maximum workspace to 1GB and the maximum batch number to 8 to balance memory occupancy and throughput. Verify the calculation time of each layer through TRTProfiler, and the generated serialized engine file is deployed to the edge node. The actual single inference delay is reduced from 120ms of the original model to 43ms, meeting the real-time requirements.

[0093] According to the characteristics of the input feature dimension and the hidden state dimension, a double-buffer queue is pre-allocated in the memory.

[0094] It should be noted that the feature dimension corresponds to the real-time acquisition variable in the standardized input matrix generated in step S1, including: 5 dimensions of gas component data (methane, ethane, propane, carbon monoxide, hydrogen volume percentage concentration); 5 dimensions of boiler operating parameters (steam pressure, flue gas temperature, oxygen content, feedwater flow, load rate); the role in the double-buffer queue: the input data at each time (1 Hz) is composed of these 10 features, the pre-allocated current time input buffer (30 seconds capacity) stores the 10-dimensional feature sequence for 30 seconds, which is used for time-dependent calculation of model inference.

[0095] The hidden state dimension is derived from the construction of the hidden layer of the neural differential equation prediction model in step S2, including: prediction output targets: direct control variables such as thermal efficiency, nitrogen oxide concentration, and pipe wall stress distribution; intermediate state variables: combustion kinetic process features derived during the differential equation solving process (such as instantaneous combustion rate, flue gas residence coefficient). The role in the double-buffer queue: the pre-allocated prediction result buffer (60 seconds capacity) stores the 15-dimensional hidden state sequence of the model output, providing state prediction trajectories in the future time domain for the dynamic optimization controller, ensuring the time sequence coherence of the control command generation.

[0096] Among them, the current time input buffer (storing the latest 30 seconds of standardized data) and the prediction result buffer (storing the future 60 seconds of prediction sequence).

[0097] Furthermore, every time 1 second of new data is received, the neural differential equation prediction model inference process is triggered.

[0098] Specifically, the latest 30 seconds of standardized data is input into the neural differential equation prediction model, and a forward propagation calculation is performed to obtain a future 60 seconds of state prediction sequence (thermal efficiency, NOx concentration, etc.). Physical reasonableness verification is performed on the output sequence: check whether the thermal efficiency is within the interval [20%, 95%] and whether the NOx concentration exceeds the environmental protection limit value (such as ≤50mg / m 3 If not, the neural differential equation prediction model is re-inferred (up to 3 times).

[0099] The prediction sequence is packaged into a binary data packet through a time series compression algorithm (such as Delta encoding + ZigZag compression), and is transmitted to the dynamic optimization controller through the MQTT protocol, with an end-to-end transmission delay ≤100ms.

[0100] A prediction confidence assessment unit is provided at the dynamic optimization controller side, and the prediction time domain length is automatically reduced when the following two conditions are detected: specifically, when the gas composition fluctuation rate exceeds ±10% for 5 seconds in a row (indicating unstable fuel); and when the deviation between the predicted value and the actual value of the steam pressure is greater than 5% for 3 seconds in a row.

[0101] S4. In the dynamic optimization controller, a rolling horizon optimization problem is constructed based on the state prediction sequence, the control instruction set is converted into physical operation quantities by the actuator, the gas flow, the combustion air flow, and the feedwater flow are adjusted, and after the boiler operating state is updated, the actual operating data is fed back to the constraint update process.

[0102] The rolling horizon optimization objective function is defined based on economic indicators, environmental protection indicators, and safety indicators.

[0103] Among them, the economic indicators: minimize the gas consumption (quadratic correlation with valve opening); environmental protection indicators: limit the rising rate of NOx concentration (linear correlation with air flow adjustment); safety indicators: the absolute value of the tube wall temperature gradient change rate ≤3℃ / s.

[0104] An improved sequential quadratic programming (SQP) algorithm is used to construct a hierarchical solving strategy for control variable dimensions (gas valve opening, fan speed, feedwater pump frequency):

[0105] Specifically, the first layer: calculate the gas flow set value based on the prediction sequence, the constraint conditions are the heat balance equation and the tube wall temperature gradient; the second layer: fix the gas flow, optimize the combustion air flow and the feedwater flow ratio, the constraint conditions are the oxygen content limit and the steam pressure.

[0106] It should be noted that the upper limit of the solving time is set to 200ms, and when the timeout is not converged, a backup scheme is enabled.

[0107] Specifically, when the optimization solving time exceeds 200ms and does not converge, the dynamic optimization controller immediately executes an emergency control strategy: when the load rate is >80%, the steam pressure control target is locked, the preset air-fuel ratio parameter is used, and the pressure closed-loop PID regulation is activated; when the load rate is ≤80%, switch to the efficiency priority mode, generate control instructions based on the optimal historical data of the last 24 hours, and superimpose feedwater flow feedforward compensation. During the transition process, the steam pressure fluctuation is limited within ±0.2MPa, the oxygen content deviation is controlled within ±0.5%, the tube wall temperature gradient does not exceed 3℃ / s, and the NOx concentration is maintained at 50mg / m 3 Below, ensure that the control mode switching is completed within 5 seconds. The preset air-fuel ratio parameter refers to the fuel and air ratio benchmark value stored in advance in the boiler control and verified by historical operating data.

[0108] The optimization solution is converted into actuator instructions.

[0109] Specifically, for the gas regulating valve, a feedforward-feedback compound control is adopted, the feedforward amount comes from the optimization solution, and the feedback amount is based on the pressure deviation PID correction; for the variable frequency fan, the frequency set value is generated based on the air volume-speed characteristic curve interpolation, and the surge prevention logic (avoiding the 35-45 Hz resonance region) is superimposed; for the feed water pump, the flow differential compensation is used to offset the hysteresis effect caused by the boiler water volume.

[0110] A command smoothing unit is deployed in the PLC to apply first-order inertia filtering to the command jump (such as valve opening mutation > 5%) of adjacent control periods to prevent mechanical impact.

[0111] Specifically, the command smoothing unit deployed in the PLC control program uses dynamic inertia filtering technology to process control commands. The command smoothing unit continuously monitors the change amplitude of key control commands such as gas regulating valve opening, fan speed, and feed water pump frequency. When a command jump in adjacent control periods is detected, the smoothing algorithm is automatically activated. The filtering strength is automatically adjusted according to the current operating condition. In the rapid load change stage, the control response speed is emphasized, and in the stable operation stage, the command smoothing degree is emphasized. This adaptive adjustment mechanism not only ensures dynamic response capability, but also controls the action rate of the actuator within a safe range, effectively avoiding mechanical impact and wear of valves, fans, and other equipment caused by sudden changes in commands. All smoothing processes are completed within a single control period, without affecting the overall control timing.

[0112] It should be noted that when referring to the command jump in adjacent control periods. The "command" here refers to the direct operation command generated by the dynamic optimization controller in real time and issued to the actuator (gas regulating valve, combustion air fan, feed water pump), such as: gas regulating valve opening set value (percentage command, such as 50% opening), combustion air fan speed set value (speed command, such as 1200 rpm), feed water pump frequency set value (Hz command, such as 45 Hz). These commands are converted by the PLC into physical execution quantities (such as valve opening signal, frequency converter control signal) to directly drive the regulation of gas flow, combustion air volume, and feed water flow of the boiler. The jump (such as the opening command jumping from 45% to 52% in adjacent control periods) will cause a step change in the actuator's action amplitude, triggering mechanical stress, so a first-order inertia filter is needed to smooth the command sequence, limiting the actuator's action rate within a safe range (such as valve opening changing ≤3% per second) while maintaining control accuracy, thereby avoiding mechanical impact or life loss caused by sudden load changes.

[0113] Actual operating data (steam pressure actual value, NOx concentration measurement value, etc.) after execution are collected to calculate the prediction-actual deviation index.

[0114] Specifically, short-term deviation (<10 seconds): online parameter self-tuning for dynamic optimization controller; long-term deviation (> 30 minutes): trigger constraint update process.

[0115] When any of the following conditions is detected, the constraint equation dynamic update of S2 step is started:

[0116] Specifically, the same constraint equation continuously predicts the violation amount for 10 times > thermal balance deviation threshold (such as thermal balance deviation > 8 kW); the gas composition changes in quality (such as hydrogen gas proportion suddenly increases by > 5% and lasts for 1 minute).

[0117] The feedback data is collected and returned to the edge computing node through an encrypted channel, triggering lightweight genetic programming retraining (only updating mutation rate and selection pressure parameters), and the constraint condition iteration is completed in ≤ 2 minutes.

[0118] The embodiment also provides a gas-steam boiler energy supply energy efficiency optimization system, comprising a data acquisition module, a constraint modeling module, a real-time prediction module and an optimization control module; the data acquisition module is used for acquiring gas composition data in real time, synchronously acquiring boiler operation parameters, and generating a standardized input matrix from the gas composition data and the boiler operation parameters through data preprocessing; the constraint modeling module is used for mining physical constraint conditions through a genetic programming algorithm based on historical operation data and the standardized input matrix, generating a constraint equation set in a differential algebraic form, encoding the constraint equation as a Lagrange multiplier item to inject into a neural differential equation, and constructing a neural differential equation prediction model with physical constraints; the real-time prediction module is used for acquiring a boiler real-time operation data stream, inputting the boiler real-time operation data stream into the trained neural differential equation prediction model, calculating a state prediction sequence in a future time domain, and transmitting the state prediction sequence to a dynamic optimization controller; and the optimization control module is used for constructing a rolling horizon optimization problem based on the state prediction sequence in the dynamic optimization controller, converting a control instruction set into a physical operation quantity through an actuator, adjusting gas flow, combustion air volume and feed water flow, and feeding back actual operation data to a constraint update process after completing boiler operation state updating.

[0119] The embodiment also provides a computer device suitable for the gas-steam boiler energy supply energy efficiency optimization method, comprising a memory and a processor; the memory is used for storing computer executable instructions, and the processor is used for executing the computer executable instructions to realize the gas-steam boiler energy supply energy efficiency optimization method proposed in the above embodiment.

[0120] The computer device can be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved by WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0121] The embodiment also provides a storage medium having a computer program stored thereon, the program being executed by a processor to implement the method for optimizing energy efficiency of a gas-steam boiler as described in the above embodiment; and the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk.

[0122] To sum up, the present application achieves deep coupling of physical laws and data-driven models by dynamically mining differential algebraic constraint equations based on genetic programming; significantly reduces constraint violation detection delay by dynamically injecting constraint gradient tensors through Lagrange multiplier weighting layers combined with edge computing acceleration and buffering mechanisms; improves the response speed of control instructions by implementing fast rolling optimization with dynamic constraints through an improved hierarchical solving strategy; ensures the physical reasonableness of the prediction results through reverse sign deduction verification of the constraint equations; and maintains prediction accuracy and control stability by adaptively updating the constraint equations when the gas composition suddenly changes.

[0123] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for optimizing energy efficiency of a gas-steam boiler, characterized in that: Comprising, Real-time gas composition data is collected, boiler operating parameters are synchronously acquired, and the gas composition data and the boiler operating parameters are generated into a standardized input matrix through data preprocessing; Based on the standardized input matrix, a set of constraint equations in the form of differential algebra is generated by mining physical constraint conditions through a genetic programming algorithm, the constraint equations are coded into Lagrange multiplier terms, the neural differential equation is injected, and a neural differential equation prediction model with physical constraints is constructed; Real-time boiler operating data streams are acquired and input into the trained neural differential equation prediction model, the state prediction sequence in the future time domain is calculated, and the state prediction sequence is transmitted to a dynamic optimization controller; In the dynamic optimization controller, a rolling horizon optimization problem is constructed based on the state prediction sequence, the control instruction set is converted into physical operation quantities through an actuator, the gas flow, combustion air flow, and feed water flow are adjusted, the boiler operating state is updated, and the actual operating data is fed back to a constraint update process.

2. The method for optimizing the efficiency of a gas-steam boiler energy supply according to claim 1, characterized in that: The gas composition data includes the volume percentage concentrations of methane, ethane, propane, carbon monoxide, and hydrogen; and the boiler operating parameters include steam pressure, flue gas temperature, oxygen content, feed water flow, and load rate.

3. The method for optimizing the efficiency of a gas-steam boiler energy supply of claim 1, characterized by the fact that: The set of constraint equations includes a heat balance equation, an oxygen content limit, and a tube wall temperature gradient constraint; The constraint equations are coded into Lagrange multiplier terms and injected into the neural differential equation, and the specific steps are as follows, The heat balance equation, the oxygen content limit, and the tube wall temperature gradient constraint are input into a differential algebraic encoder, the partial derivatives of the constraint equations with respect to the hidden state are calculated through a gradient field generation layer, and a constraint gradient tensor is output; The constraint gradient tensor is dynamically coupled with the forward propagation result of the neural differential equation, a constraint-enhanced gradient flow is generated through a Lagrange multiplier weighting layer; The coupled gradient flow is input into an adjoint optimizer, joint backpropagation calculation is performed, and the neural network parameters and the Lagrange multiplier are synchronously updated; The optimized gradient flow is input into a prediction corrector, and the heat balance equation, the oxygen content limit, and the tube wall temperature gradient constraint are simultaneously satisfied through an interior point method solver.

4. The method of claim 3, wherein: The input layer of the neural differential equation prediction model receives real-time sensor data streams, the hidden layer describes the combustion dynamics process through a neural differential equation, the output layer generates state prediction values in the future time period, the state prediction values include thermal efficiency, nitrogen oxide concentration, and tube wall stress distribution, the neural differential equation prediction model parameters are optimized based on the adjoint method during the training process, and the loss function fuses the prediction error and the constraint violation penalty.

5. The method of claim 4, wherein: The control variables of the rolling horizon optimization problem include the gas valve opening, the fan speed, and the feed water pump frequency, the control instruction set is generated in real time through an optimization algorithm, and the control instruction set is issued to the actuator.

6. The method of claim 5, wherein: The actual operating data refers to real-time operating condition verification data after the gas-steam boiler executes the control instruction; The real-time operating condition verification data includes control response data, operating condition verification parameters, and energy efficiency indicators.

7. The method of claim 6, wherein: The constraint update process refers to a closed-loop optimization process of dynamically correcting physical constraint conditions based on real-time boiler operating data streams.

8. A gas-steam boiler energy supply efficiency optimization system based on the gas-steam boiler energy supply efficiency optimization method according to any one of claims 1-7, characterized in that: Comprising a data acquisition module, a constraint modeling module, a real-time prediction module, and an optimization control module; The data acquisition module is configured to collect gas composition data in real time, acquire boiler operation parameters synchronously, and generate a standardized input matrix from the gas composition data and the boiler operation parameters through data preprocessing; The constraint modeling module is configured to mine physical constraint conditions through a genetic programming algorithm based on historical operation data and the standardized input matrix, generate a set of constraint equations in a differential algebraic form, encode the constraint equations as Lagrange multiplier terms to inject into a neural differential equation, and construct a neural differential equation prediction model with physical constraints; The real-time prediction module is configured to acquire a boiler real-time operation data stream, input the data stream into the trained neural differential equation prediction model, calculate a state prediction sequence in a future time domain, and transmit the state prediction sequence to a dynamic optimization controller; The optimization control module is configured to construct a rolling horizon optimization problem based on the state prediction sequence in the dynamic optimization controller, convert a control instruction set into physical operation quantities through an actuator, adjust gas flow, combustion air flow, and feed water flow, and complete boiler operation state updating, and then feed actual operation data back to a constraint updating process. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the gas-steam boiler energy supply efficiency optimization method of any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the gas-steam boiler energy supply efficiency optimization method of any one of claims 1-7.

Images