Multi-mode intelligent control system for titanium dioxide smelting furnace

Abstract

The invention discloses a multi-mode intelligent control system for a titanium dioxide smelting furnace, which relates to the technical field of metallurgical process control and industrial automation and comprises an acquisition unit, an execution unit and a controller. Collecting a reduction degree fingerprint, a partition heat exchange quantity sequence and an electric parameter / electrode state; and the execution comprises partitioned cooling, waveform power supply, electrode micro displacement and carbon feeding. The controller generates a state derivation key and an orthogonal threshold, opens an identification window and injects a cooling domain / power supply domain code, performs deconvolution and integral separation on furnace shell thermal response based on cross-correlation and read-back records, and outputs a frozen layer thickness field and confidence; gating energy-saving permission is judged through residual convergence, cross-domain consistency and execution controllability, joint optimization is carried out within the limit boundary, and power supply and cooling are cooperatively executed through conjugate coupling constraint. And triggering an emergency degradation back-off conservative track and restarting identification when the limit is out of limit.

Description

Technical Field

[0001] This invention relates to the field of metallurgical process control and industrial automation technology, specifically a multimodal intelligent control system for titanium dioxide smelting furnaces. Background Technology

[0002] Titanium dioxide smelting furnaces operate under conditions of high temperature, strong coupling, and strong disturbance. Operations such as power supply, cooling, and carbon addition collectively affect energy deposition, boundary heat transfer, and changes in reaction heat sources within the furnace, thereby influencing the state of the frozen layer and the degree of reduction. To balance energy conservation and safety, engineering projects typically equip furnaces with data acquisition and execution devices to monitor furnace conditions and adjust key parameters.

[0003] In existing technologies, stable operation and energy consumption reduction are often achieved by collecting relevant furnace condition data and controlling zonal cooling, power supply, electrode position, and carbon feeding. However, certain limitations exist in practical applications: the frozen layer state is difficult to measure directly, often relying on single quantities such as furnace shell temperature or empirical thresholds for judgment, which can easily lead to control drift and misjudgment when there are fluctuations in operating conditions, noise disturbances, or execution link delays; the boundary heat transfer response component and the power supply deposition response component in the furnace shell thermal response are easily mixed, and there may be substitution errors between changes in the heat source term and changes in the frozen layer thickness, resulting in insufficient stability of the frozen layer state inversion, and thus making it difficult to provide verifiable state identification basis for energy-saving control. Energy saving on the power supply side and safety on the cooling side are often solved and executed separately, lacking strong coupling constraints within the same control cycle and a systematic judgment mechanism for entering / maintaining energy-saving operation. This may lead to blind entry into energy-saving operation when the inversion confidence is insufficient or the execution controllability is insufficient, posing a safety risk.

[0004] Therefore, a multimodal intelligent control system for titanium dioxide smelting furnaces is urgently needed to establish energy-saving control based on verifiable identification and inversion. This system should be able to combine reduction fingerprints, zoned heat exchange sequences, and electrical parameters and electrode state indicators to implement verifiable dual-domain excitation of zoned cooling, waveform power supply, and electrode micro-displacement within the identification window, forming a coded readback verification record. Based on the furnace shell thermal response deconvolution separation and reduction fingerprint constraints, the system should output the frozen layer thickness field and inversion confidence. Simultaneously, it should perform permission gating for energy-saving operation through separation residual convergence judgment, cross-domain consistency judgment, and execution controllability judgment. Under control boundary constraints, it should jointly optimize the power supply waveform trajectory, electrode micro-displacement trajectory, zoned cooling flow trajectory, and carbon feeding trajectory, applying conjugate coupling constraints to the power supply waveform trajectory and zoned cooling flow trajectory to improve the safety, interpretability, and reliability of energy-saving control. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and propose a multimodal intelligent control system for titanium dioxide smelting furnaces to solve the above-mentioned problems.

[0006] The objective of this invention is achieved through the following technical solution: a multimodal intelligent control system for a titanium dioxide smelting furnace, comprising a data acquisition unit, an execution unit, and a controller; wherein, the data acquisition unit includes a furnace gas spectrum acquisition component, a furnace shell thermal field acquisition component, and an electrical parameter and electrode state acquisition component; the furnace gas spectrum acquisition component is used to acquire a reduction degree fingerprint, the furnace shell thermal field acquisition component is used to acquire a zoned heat transfer sequence, and the electrical parameter and electrode state acquisition component is used to acquire electrical parameter and electrode state indicators; the execution unit includes a zoned cooling execution component, a waveform power supply execution component, an electrode micro-displacement execution component, and a carbon feeding execution component; the controller includes a state derivation key generation component, an identification window gating component, and a dual-domain... The system comprises an orthogonal coding excitation component, a coding readback verification component, a cross-domain deconvolution assimilation component, a three-key permission domain distribution component, a minimum energy consumption joint optimization control component, and a degradation recovery component. The controller is configured to: generate a state-derived key based on the restoration fingerprint, electrical parameters, and electrode state indicators, and output an orthogonal threshold parameter; open an identification window based on the state-derived key; generate a cooling domain coding sequence and a power supply domain coding sequence by the dual-domain orthogonal coding excitation component within the identification window, and inject the cooling domain coding sequence and the power supply domain coding sequence into the partitioned cooling execution component, the waveform power supply execution component, and the electrode micro-displacement execution component, respectively; and output an execution uncontrollable flag when the cross-correlation evaluation value exceeds the orthogonal threshold parameter. Generate characterization cross-correlation evaluation values, execution uncontrollable flags, and coded readback verification records of command tracking deviations; deconvolve and separate the furnace shell thermal response based on the zoned heat exchange sequence and coded readback verification records, and constrain the heat source term with a reduction degree fingerprint, outputting the frozen layer thickness field and inversion confidence; use the monotonically decreasing and converging component decomposition residuals as the convergence criterion for separation residuals, use the consistency threshold between the direction of change of the frozen layer thickness field and the direction of change of the heat source correction as the cross-domain consistency criterion, and use the absence of the execution uncontrollable flag and the command tracking deviation not exceeding the deviation threshold as the execution controllability criterion, thereby issuing energy-saving permits and forming energy-saving quota control boundaries when all three criteria are met simultaneously; otherwise, rejecting the permits; in the control... Under boundary constraints, the system jointly optimizes the output power supply waveform trajectory, electrode micro-displacement trajectory, zoned cooling flow trajectory, and carbon feeding trajectory. Conjugate coupling constraints are applied to the power supply waveform trajectory and the zoned cooling flow trajectory, ensuring they both satisfy the frozen layer safety constraint and the allowable window constraint within the same control cycle. The frozen layer safety constraint includes the lower limit of the thinnest region thickness, the upper limit of the thinnest region thickness change rate, and the upper limit of the thickness spatial gradient. The power supply waveform trajectory and the zoned cooling flow trajectory are issued only when the conjugate coupling constraints are satisfied, and these two trajectories cannot be solved or executed independently. The system exits energy-saving operation and returns to the identification window when energy-saving permission is rejected or any judgment fails during energy-saving operation.

[0007] The identification window gating component is used to read historical frozen layer thickness field records during the system startup phase and generate an initial frozen layer margin accordingly. When there are no historical frozen layer thickness field records, a conservative initial frozen layer margin is generated based on the furnace shell distributed temperature field and the zoned heat exchange sequence. The identification window gating component only allows the identification window to be opened when the initial frozen layer margin meets the preset safety threshold and the restoration fingerprint, electrical parameters and electrode state indicators meet the joint stability threshold.

[0008] The dual-domain orthogonal coding excitation component is used to perform frequency domain isolation and periodic family isolation on the cooling domain coding sequence and the power supply domain coding sequence based on the state-derived key, and write the isolation flag into the coding readback verification record. The coding readback verification component is used to calculate the cross-correlation evaluation value within the identification window and output the execution uncontrollable flag to the three-key permission domain issuing component when the cross-correlation evaluation value exceeds the orthogonality threshold parameter.

[0009] The furnace gas spectral acquisition component includes a laser absorption spectroscopy measurement unit and a fingerprint generation unit. The laser absorption spectroscopy measurement unit is used to output the volume fraction sequence of carbon monoxide, the volume fraction sequence of carbon dioxide, the volume fraction sequence of water vapor, and the furnace gas temperature sequence. The fingerprint generation unit is used to form a reduction fingerprint based on the volume fraction sequence and the furnace gas temperature sequence and output the reduction deviation direction and reduction change rate for the state derivation key generation component and the cross-domain deconvolution assimilation component.

[0010] The furnace shell thermal field acquisition component includes a distributed optical fiber temperature measurement link and a zoned cooling water path acquisition link. The distributed optical fiber temperature measurement link is used to form a continuous temperature sequence along the circumference and height of the furnace shell. The zoned cooling water path acquisition link is used to form the flow rate sequence of the cooling medium in each zone and the inlet and outlet temperature difference sequence of the cooling medium in each zone. The furnace shell thermal field acquisition component is used to generate a zoned heat transfer sequence based on the flow rate sequence and the inlet and outlet temperature difference sequence and output it to the cross-domain deconvolution assimilation component.

[0011] The cross-domain deconvolution assimilation component is used to decompose the distributed temperature field of the furnace shell into components based on the cooling input response corresponding to the cooling domain encoding sequence and the power input response corresponding to the power supply domain encoding sequence to obtain the boundary heat transfer response component and the power supply deposition response component. The component also uses the reduction fingerprint to constrain and correct the reaction heat source term to suppress the mutual substitution error between the change of the reaction heat source term and the change of the frozen layer thickness field. The cross-domain deconvolution assimilation component is used to generate the inversion confidence based on the component decomposition residual and output it to the three-key permission domain distribution component.

[0012] The three-key permission domain issuance component uses the monotonic decrease and stable convergence of the component decomposition residual within a preset time period as the determination of separation residual convergence, the satisfaction of the preset consistency threshold between the change direction of the frozen layer thickness field and the change direction of the heat source correction of the reduction fingerprint as the determination of cross-domain consistency, and the simultaneous satisfaction of the preset deviation threshold between the cooling command tracking deviation, power supply waveform command tracking deviation and electrode micro-displacement command tracking deviation characterized by the encoded readback verification record as the determination of execution controllability. Thus, when all three determinations are met, an energy-saving permission is output, and when any one determination is not met, an energy-saving permission is rejected.

[0013] The minimum energy consumption joint optimization control component is used to write the lower limit of the thinnest region thickness, the upper limit of the thinnest region thickness change rate, and the upper limit of the thickness spatial gradient into the safety constraints, and to write the allowable window corresponding to the reduction fingerprint into the chemical constraints. It also writes the upper limit of the power supply waveform modulation amplitude and the upper limit of the change rate, the upper limit of the partition cooling flow adjustment amplitude and the upper limit of the change rate, and the upper limit of the carbon feeding correction amplitude and the upper limit of the change rate of the energy saving quota into the execution constraints. Thus, under the premise of simultaneously satisfying the safety constraints, chemical constraints, and execution constraints, it outputs the power supply waveform trajectory, electrode micro-displacement trajectory, partition cooling flow trajectory, and carbon feeding trajectory.

[0014] The energy deposition modulation of the power supply waveform trajectory and the boundary heat extraction modulation of the partitioned cooling flow trajectory, which are constrained by conjugate coupling constraints, jointly satisfy the lower limit constraint of the thinnest region thickness, the upper limit constraint of the rate of change of the thinnest region thickness and the upper limit constraint of the thickness space gradient in the same control cycle, and jointly satisfy the allowable window constraint corresponding to the degree of reduction fingerprint.

[0015] The degradation recovery component is used to trigger emergency degradation when the thickness of the thinnest region of the frozen layer thickness field is lower than a preset safety threshold, the rate of change of the frozen layer thickness exceeds a preset rate of change threshold, the deviation of the restoration fingerprint from the allowable window, or the cross-correlation evaluation value represented by the encoded readback verification record exceeds the orthogonality threshold parameter, or the command tracking deviation exceeds a preset deviation threshold. During the emergency degradation, the degradation recovery component is used to revert the power supply waveform command to a conservative power supply waveform, the electrode micro-displacement command to a conservative micro-displacement, the partitioned cooling flow command to a conservative cooling boundary, and the carbon feeding trajectory to a conservative carbon feeding trajectory. This allows the identification window to be reopened to complete the inversion confidence reconstruction while maintaining the stability of the frozen layer and the control of the restoration.

[0016] The beneficial effects of this invention are: The intelligent control system for titanium dioxide smelting furnace provided by this invention includes a data acquisition unit, an execution unit, and a controller. The data acquisition unit acquires the reduction degree fingerprint, the zoned heat exchange sequence, and electrical parameters and electrode state indicators. The execution unit includes at least zoned cooling, waveform power supply, electrode micro-displacement, and carbon feeding. Based on this, the controller forms a closed-loop link of identification—assimilation—permission—joint optimization—execution—re-identification, thereby establishing energy-saving control on the basis of verifiable state recognition and frozen layer constraints, avoiding the control drift and misjudgment risks caused by relying solely on empirical thresholds or single measurements. The controller generates a state-derived key based on the reduction degree fingerprint, electrical parameters, and electrode state indicators and outputs orthogonal threshold parameters. Based on the state-derived key, the identification window is opened, binding the injection and verification of the cooling domain encoding sequence and the power supply domain encoding sequence to the current furnace state. This helps maintain the controllability and consistency of the identification conditions under conditions of fluctuation, execution link delay, or noise disturbance. By injecting cooling domain coding sequences and power supply domain coding sequences into the partitioned cooling, waveform power supply, and electrode micro-displacement respectively within the identification window, and using cross-correlation evaluation values ​​and orthogonal threshold parameters to output an execution uncontrollable flag, and simultaneously generating a coding readback verification record characterizing the cross-correlation evaluation value, the execution uncontrollable flag, and the command tracking deviation, the orthogonality of the dual-domain excitation and the execution tracking status can be self-verified and traced. This facilitates timely identification of abnormal situations such as execution uncontrollability or increased tracking deviation, thereby improving the reliability of system operation.

[0017] This invention deconvolvees and separates the furnace shell thermal response based on partitioned heat transfer sequences and coded readback verification records, and constrains the reaction heat source terms with a reduction degree fingerprint, outputting the frozen layer thickness field and inversion confidence. This distinguishes the boundary heat transfer response components and power supply deposition response components in the distributed temperature field of the furnace shell, reducing the substitution error between changes in reaction heat source terms and changes in the frozen layer thickness field, thus improving the stability and reliability of the frozen layer thickness field inversion. A three-key permission domain issuance component uses residual convergence determination, cross-domain consistency determination, and execution controllability determination as parallel prerequisites for energy-saving permission. Energy-saving permission is issued and an energy-saving quota control boundary is formed when all three determinations are met simultaneously; otherwise, it is rejected and the system returns to the identification window. This explicitly institutionalizes the entry and maintenance conditions for energy-saving operation, avoiding blindly entering energy-saving operation when inversion confidence, cross-domain consistency, or execution controllability are insufficient, thereby improving the security and interpretability of the energy-saving strategy.

[0018] Under the constraint of energy-saving quota control boundary, this invention uses a minimum energy consumption joint optimization control component to jointly optimize the output power supply waveform trajectory, electrode micro-displacement trajectory, zoned cooling flow trajectory, and carbon feeding trajectory. Conjugate coupling constraints are applied to the power supply waveform trajectory and the zoned cooling flow trajectory, so that they jointly satisfy the frozen layer safety constraint and allowable window constraint within the same control cycle. The power supply waveform trajectory and the zoned cooling flow trajectory are issued only when the conjugate coupling constraint is satisfied, and they are restricted from being solved and executed independently. This can suppress the mutual restraint between energy saving on the power supply side and safety on the cooling side, improve the feasibility and consistency of joint control, and realize collaborative energy-saving control with the frozen layer thickness field as the safety boundary, the reduction fingerprint allowable window as the chemical boundary, and the energy-saving quota control boundary as the execution boundary. Furthermore, this invention incorporates a degradation recovery component that triggers emergency degradation when conditions such as the frozen layer thickness field, reduction fingerprint, cross-correlation evaluation value, or command tracking deviation exceed limits. This component reverts the power supply waveform command to a conservative power supply waveform, the electrode micro-displacement command to a conservative micro-displacement, the partitioned cooling flow command to a conservative cooling boundary, and the carbon feeding trajectory to a conservative carbon feeding trajectory. This maintains the stability of the frozen layer and the control of the reduction while simultaneously reopening the identification window to complete the inversion confidence reconstruction, thereby improving the system's safety redundancy and self-recovery capability under abnormal operating conditions. Attached Figure Description

[0019] Figure 1 The process of this invention Figure 1 ; Figure 2 The process of this invention Figure 2 ; Figure 3 The process of this invention Figure 3 . Detailed Implementation

[0020] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1 like Figure 1As shown in this embodiment, a feasible scheme for a multimodal intelligent control system for a titanium dioxide smelting furnace is provided. The system includes an acquisition unit, an execution unit, and a controller. The acquisition unit is used to acquire the reduction fingerprint, the heat exchange sequence of the partition, and the electrical parameters and electrode state indicators. The execution unit includes partition cooling, waveform power supply and electrode micro-displacement, and carbon feeding. The controller is used to complete the generation of state derivation keys and output of orthogonal threshold parameters, opening of the identification window, dual-domain encoding injection and cross-correlation evaluation, generation of encoding back-read verification records, deconvolution separation of furnace shell thermal response and output of frozen layer thickness field and inversion confidence, three-item judgment and issuance or rejection of energy-saving permits, joint optimization of minimum energy consumption within the energy-saving quota control boundary and output of power supply waveform trajectory, electrode micro-displacement trajectory, partition cooling flow trajectory and carbon feeding trajectory, and apply conjugate coupling constraints to the power supply waveform trajectory and partition cooling flow trajectory so that they jointly satisfy the frozen layer safety constraint and the allowable window constraint in the same control cycle, and execute only when the conjugate coupling constraint is satisfied; when the energy-saving permit is rejected or any judgment fails during energy-saving operation, the system exits energy-saving operation and returns to the identification window.

[0022] The furnace gas spectrum section of the acquisition unit uses a tunable semiconductor laser absorption spectroscopy measurement channel to scan the absorption spectra of carbon monoxide, carbon dioxide, and water vapor respectively, and simultaneously outputs the furnace gas temperature; in engineering implementation, the furnace gas spectrum sampling frequency is set to... to In each sampling period, Volume fraction sequence , Volume fraction sequence , Volume fraction sequence With furnace gas temperature sequence The reproducibility fingerprint is constructed by a fingerprint generation algorithm at the same timestamp and output as a continuous time series; the reproducibility fingerprint includes at least the reconstruction index. With the rate of change of reduction and temperature characteristics Furthermore, it can include deviations in the reduction direction. (For example, output by the fingerprint generation unit;) Desirable These represent negative deviation, zero deviation, and positive deviation, respectively. Preferably, the reduction fingerprint is defined as a vector composed of the reduction index, rate of change, and temperature characteristics, and optionally includes the deviation direction: Restoration Index It can be defined as the proportion of carbon monoxide volume fraction in the sum of the volume fractions of carbon monoxide and carbon dioxide, i.e. .

[0023] Fingerprint fidelity It can be taken as a reason , and The vector formed, and optionally further includes the direction of reduction deviation. .

[0024] in Calculated using first-order differences, the difference window is... Or it may be consistent with the sampling period. The controller will As a fingerprint input for accuracy restoration, it simultaneously acquires electrical parameter and electrode state indicators from the electrical parameter and electrode state acquisition channels. The electrical parameter and electrode state indicators include at least: the root mean square of the three-phase voltage. Three-phase current root mean square Active power reactive power Electrode displacement Electrode displacement velocity and waveform power supply modulation state To ensure timing consistency, the sampling frequency is set to... to The time alignment module then downsamples the data to the same level as the furnace gas spectrum or an integer multiple thereof.

[0025] The furnace shell thermal field data acquisition unit includes the distributed temperature field of the furnace shell and the data from the zoned cooling water channels. The distributed temperature field of the furnace shell is obtained through a distributed fiber optic temperature measurement link, with a spatial sampling interval of [missing information]. to The time sampling frequency is to ; Collect cooling medium flow rate data for each zone of the cooling water circuit. Inlet water temperature Outlet water temperature The number of partitions Preferred to The heat exchange sequence for each zone is calculated and output as a vector sequence: In this embodiment, the zoned cooling flow rate can be characterized by either volumetric flow rate or mass flow rate; to be consistent with the calculation method of zoned heat exchange, mass flow rate will be used below. Indicates partition The zone cooling flow rate, and denoted as .

[0026] The heat exchange sequence can be ordered by the cooling medium flow rate of each zone. Temperature difference between inlet and outlet Calculated heat exchange of the zones (For example, using) (energy conservation calculation method), and each partition Composition of vector sequence Output.

[0027] in For the isobaric specific heat capacity of the cooling medium, when water cooling is used, take... The controller has the following functions: Perform timestamp alignment, with After the electrical parameters are aligned with the electrode state indicators, the process proceeds to the next stage.

[0028] The controller in each control cycle Execution state derived key generation. To ensure the key is bound to the furnace condition and is reproducible, the controller first quantizes and concatenates the fidelity fingerprint, electrical parameters, and electrode state indicators, with the quantization step size set to [value missing]. , , , It can optionally introduce periodic random numbers. The input string is formed; the state-derived key can be obtained by using a hash function to digest the input string (e.g., SHA-256). Among them, when not introduced At that time, the controller can It is set to a deterministic sequence derived from the window start and end timestamps, control cycle counter, or fixed seed. The controller is based on... Output orthogonal threshold parameters ,in In this embodiment, it is set to to The adaptive range, the default value is... When furnace condition fluctuations increase, follow Adjustment, To identify the window Inside The standard deviation.

[0029] The controller opens the identification window based on the state-derived key. Identification window length. Set as to In this embodiment, The conditions for opening the recognition window include: the fingerprint with high fidelity is in a discernible steady state ( ). continued (above) and no mutations in the zonal heat exchange sequence ( continued The above, of which (This represents the maximum absolute value of each component of the vector). When the condition is met, the controller sets the identification window to enable and enters the dual-domain orthogonal encoding excitation.

[0030] Within the identification window, the controller injects cooling domain encoding sequences and power supply domain encoding sequences into the zoned cooling, waveform power supply, and electrode micro-displacement domains, respectively. The cooling domain encoding sequence is of length... Multi-partition encoded vector sequence The power supply domain encoding sequence is of length scalar or low-dimensional vector sequences Both are from A defined frequency band and periodic family are generated and guaranteed to be non-aliased; in this embodiment, the frequency band and periodic family are selected. Sampling interval Encoding cycle The identification window length is... The number of sampling points within the corresponding window Point (of which) (This indicates rounding down to obtain the integer number of sample points). The controller takes consecutive values ​​of the encoded sequence within the recognition window. Points are injected and subsequent calculations are performed, and the orthogonality of the dual-domain excitation is verified online using cross-correlation evaluation and orthogonal threshold parameters.

[0031] The injection method involves superimposing a small excitation on the base value of the execution command, with the upper limit of the superposition amplitude in the cooling domain being a fraction of the base flow rate of each zone. The upper limit of the superposition amplitude of the power supply waveform modulation is the base modulation amount. The upper limit of the superposition amplitude of electrode micro-displacements is In order to avoid affecting the production schedule.

[0032] The controller calculates the cross-correlation evaluation value in real time within the identification window and determines whether to output an uncontrollable execution flag based on orthogonal threshold parameters. To ensure that the cross-correlation evaluation value can still be directly calculated when the cooling domain is encoded as a multi-partition vector sequence, this embodiment uses a scalar convergence sequence of two-domain encoding for normalized cross-correlation calculation: Among them, the cooling zone convergence sequence The result is obtained by weighted summation of the cooling domain codes of each partition according to the partition weights: ; Power supply domain convergence sequence It is obtained by linearly combining the power supply waveform modulation code and the electrode micro-displacement code according to a preset normalization coefficient: ; in Cooling domain encoding sequence The One portion, For power supply waveform modulation encoding, Encoding electrode micro-displacement; weighting It can be obtained from the zone heat transfer sensitivity calibration or by taking the uniform weight. Normalization coefficient For dimensional standardization, this embodiment takes... , To make It is a dimensionless quantity.

[0033] Cross-correlation evaluation value The result was calculated using the normalized cross-correlation peak values ​​of the two convergent sequences within the allowable time delay range. Let the allowable time delay be... Pick to In this embodiment, The corresponding maximum lag point number Lags for any integer Define the normalized cross-correlation coefficient: ; in and Each is in the identification window and The mean; when the index When the value exceeds the window limit, zero padding or summation using the available index set can be used; this invention is not limited to these methods. The cross-correlation evaluation value is taken as: like The controller then outputs an uncontrollable execution flag. And terminate further injection within this window; if Then place And complete the code injection for this window.

[0034] The controller generates a code readback verification record at the end of the identification window. The code readback verification record must include at least: the window start and end times, the identifiers of the cooling domain code sequence and the power supply domain code sequence, and the cross-correlation evaluation value. Uncontrollable flags And the instruction tracking deviation obtained from the execution readback. In this embodiment, the instruction tracking deviation is measured as the mean square deviation of the executed quantity, calculated for the partition cooling flow, power supply waveform modulation, and electrode micro-displacement, and stored in the record as a vector: ; in Corresponding to cooling flow rate, power supply waveform modulation, or electrode micro-displacement, For the first The value of the instruction sent at each readback moment This corresponds to the actual readback value. This represents the number of readback points within the window.

[0035] Subsequently, the controller performs cross-domain deconvolution assimilation to deconvolve and separate the furnace shell thermal response, and outputs the frozen layer thickness field and inversion confidence. The controller spatially discretizes the distributed temperature field of the furnace shell into... Construct the furnace shell thermal response vector at one measurement point. This is represented as the superposition of the cooling domain input response, the power supply domain input response, and the reaction heat source term, where the reaction heat source term is constrained by the degree of reduction fingerprint: The input for cross-domain deconvolution assimilation includes at least: a partitioned heat transfer sequence. Distributed temperature field of furnace shell (discrete) ), and the code readback verification record (which at least includes the cooling zone / power supply zone code sequence identifier, isolation identifier, and cross-correlation evaluation value). Uncontrollable flags (Deviation from instruction tracking). The controller reproduces the content within this identification window based on the encoded sequence identifier and isolation identifier in the encoded readback verification record. and and in If the instruction tracking deviation exceeds the limit, the window will be marked as an invalid window and will not be entered into the assimilation solution.

[0036] Within the effective window, the controller represents the furnace shell thermal response as the superposition of the cooling domain input response, the power supply domain input response, and the equivalent contribution of the reaction heat source term, while also considering noise disturbances: in and These are the convergence sequences of the aforementioned two domains, and The equivalent impulse responses for the cooling and power supply domains, respectively (which can be implemented as lengths of...) The discrete FIR response kernel, in this embodiment, is taken as... to Preferred selection ), The spatial distribution weights of the reaction heat source term at the furnace shell measuring points (which can be obtained by normalizing the spatial location of the furnace shell measuring points, making...) ), For the scalar characterization of the heat source term of the reaction, For noise terms, It is a discrete convolution operator.

[0037] The controller will partition the heat exchange sequence. Resampling according to the coded sampling interval And use it to perform consistency checks and weight settings on the cooling domain response: for example, when a certain partition When the quality marker is invalid or mutates, the partition's performance is reduced. Weights in Alternatively, the current assimilation solution can be rejected outright, thus ensuring that the input response of the cooling domain corresponds to the heat transfer sequence of the partition in a data sense.

[0038] To suppress the substitution error between the change in the reaction heat source term and the change in the frozen layer thickness field, the controller uses a reduction degree fingerprint. Apply constraint correction. This embodiment presents a linear constraint model from fingerprint to heat source that can be directly implemented: ; in For amplitude limiting operators, and The reference value for the stable segment, coefficient and It can be obtained from historical stable segment regression calibration; upper limit Pick to .

[0039] The controller estimates the value using regularized least squares within the identification window. and (Or estimate its output components), and calculate the component decomposition residuals as the basis for convergence and confidence. The component decomposition residuals are defined as: In this embodiment, the output of the frozen layer thickness field is achieved by mapping the thermal resistance field to the thickness field. To enable direct implementation by those skilled in the art, the controller can use the following feasible computational chain to obtain the equivalent thermal resistance field and generate the frozen layer thickness field: at the end of the identification window, calculate the window average value of the heat transfer of each zone. and combined with the zoned heat exchange area Obtain the heat flux at the partition boundary Simultaneously, the representative temperature of each zone of the furnace shell is obtained from the distributed temperature field of the furnace shell. The equivalent temperature on the hot side is obtained by process settings. (For example, the equivalent temperature of the slag bath). Then the equivalent thermal resistance of the zone can be taken as... and will Mapping to discrete grid locations by partition Form an equivalent thermal resistance field The controller will thermal conductivity of the frozen layer Correlation, obtaining the frozen layer thickness field according to the discrete grid. And output it to the downstream module. For ease of implementation, this embodiment will... Set as to And it can be obtained through calibration.

[0040] The inversion confidence score can be obtained by normalizing the energy of the component decomposition residuals within the identification window: the smaller the residual energy and the better the convergence, the higher the inversion confidence score; residual scaling parameter The average residual energy of the historical stable period can be taken. times.

[0041] in For the first A recognition window, The residual scaling parameter is the average residual energy of the historical stable period. times.

[0042] The controller then performs three checks and decides whether to issue an energy-saving permit. The residual convergence check is achieved by comparing the monotonic decrease of residual energy within a window with convergence: calculations are performed at fixed step sizes within the identification window. ,in For the first A recognition window, To be The first, divided according to time sequence Each sub-window Represent the Euclidean norm; require And in the final segment, it satisfies This embodiment takes , Cross-domain consistency determination is achieved by ensuring that the direction of change in the frozen layer thickness field and the direction of change in the heat source correction meet a consistency threshold: the controller extracts the thickness of the thinnest region of the frozen layer thickness field. And calculate the direction of change within the same preset time period. And extract the heat source to correct the direction of change. ,in and This represents the difference between the final value and the initial value within the preset time period. Output the sign function (positive, zero, and negative respectively). Considering that an increase in the equivalent contribution of the reaction heat source term usually tends to lead to a thinner frozen layer, this embodiment defines consistency as having opposite directions and a correlation not exceeding a negative threshold: Requirements And consistency score In this embodiment, .

[0043] To avoid accidental satisfaction due to the use of symbols alone, this embodiment uses consistency scoring. Defined as the correlation coefficient between the thinnest region thickness difference sequence and the heat source correction difference sequence within a preset time period: Set within a preset time period with sampling intervals Obtain the sequence and ,in Preferred control period Or assimilate the output cycle (preferred). (so as to synchronize with the permission decision); then the difference sequence , The consistency score is defined as follows: ; in and These are the mean of the difference sequences, This represents the number of difference points within a preset time period.

[0044] The controllability determination is based on a combination of the uncontrollability flag and the instruction tracking deviation: Requirements and ,in Take cooling flow rate Power supply waveform modulation amount Electrode micro-displacement The controller issues an energy-saving permit and establishes an energy-saving limit control boundary only when all three conditions are met; otherwise, it outputs a rejection and returns to the identification window.

[0045] When an energy-saving permit is issued, the controller performs minimum energy consumption joint optimization control under the constraint of the energy-saving quota control boundary, outputting the power supply waveform trajectory, electrode micro-displacement trajectory, zoned cooling flow trajectory, and carbon feeding trajectory. Control cycle Set as to In this embodiment, Optimize time domain length Pick to In this embodiment, The zoned cooling flow trajectory The flow rate setting for the partitioned cooling execution unit is consistent with that of the acquisition link; when the acquisition link uses mass flow rate... During characterization, it can be The mass flow rate setting for the corresponding partition is directly taken. The controller uses a weighted sum of energy consumption and execution cost as the objective function, and the optimization variables include the power supply waveform modulation. Electrode micro-displacement Zoned cooling flow With carbonized feed : The controller uses a weighted sum of energy consumption and execution cost as its objective function: for example, using the power input as the energy consumption term, while applying smoothing penalties to the variations in zone cooling flow, electrode micro-displacement, and carbon feeding to suppress abrupt changes in execution quantity; the weight parameters can be chosen as follows. , , , As a set of implementable examples.

[0046] in For power input, As the weighting parameter, this embodiment takes... , , , .

[0047] The safety constraint of the frozen layer is incorporated into the hard constraints of the joint optimization: the thickness of the thinnest region, the rate of change of the thinnest region thickness, and the spatial gradient of the thickness are extracted from the thickness field of the frozen layer, and constrained to be no less than the lower limit of the thinnest region thickness during the prediction period. Not exceeding the upper limit of the rate of change Not exceeding the upper limit of the thickness space gradient ;in Pick to (This embodiment takes) ), Pick to (This embodiment takes) ), Pick to (This embodiment takes) The thickness change rate of the thinnest region can be controlled according to the cycle. A first-order difference approximation is performed, where the thickness spatial gradient is obtained by difference calculation on the discrete mesh and its infinity norm is taken. .

[0048] in Pick to In this embodiment, ; Pick to In this embodiment, ; Pick to In this embodiment, The window constraint is determined by the reproducibility fingerprint and is constrained during the prediction period. Located in the Allow window Within this embodiment, , And according to raw material fluctuations every Updated once. The energy-saving limit control boundary restricts the amplitude and rate of change of the executed variable: power supply waveform modulation amplitude. and Zoned cooling flow rate adjustment range And the rate of change Correction range for carbon feed And the rate of change ,in As the reference zone cooling flow vector, , These represent the relative values ​​to the current period's benchmark value. and The offset.

[0049] In this embodiment, conjugate coupling constraints are implemented by jointly satisfying them within the same control cycle: during the solution process, the controller uses both the power supply waveform trajectory and the partitioned cooling flow trajectory as inputs to the prediction model. The feasibility judgment of the frozen layer safety constraint and the allowable window constraint depends on their joint input. The controller does not solve the power supply problem and the cooling sub-problem separately; it only solves the problem when the joint optimization yields... The controller sends the power supply waveform trajectory and the zoned cooling flow trajectory to the corresponding execution unit only when both the frozen layer safety constraint and the allowable window constraint are met during the prediction period. Otherwise, the solution for that period is deemed invalid and the energy-saving permission rejection process is entered, thus ensuring that the power supply waveform trajectory and the zoned cooling flow trajectory cannot be solved and executed independently.

[0050] During execution, the controller sends the power supply waveform trajectory to the waveform power supply execution channel in the form of a waveform modulation parameter sequence, sends the electrode micro-displacement trajectory to the electrode servo execution channel in the form of a displacement setting sequence, sends the zone cooling flow trajectory to the zone cooling execution channel in the form of a sequence of valve opening or flow rate settings for each zone, and sends the carbon feeding trajectory to the carbon feeding execution channel in the form of a sequence of screw feeder speed or instantaneous mass flow rate settings; the execution unit uses... The actual execution volume is periodically read back and sent back to the controller to calculate the tracking deviation for subsequent instructions.

[0051] During energy-saving operation, the controller updates the cross-correlation evaluation value, execution uncontrollable flag, command tracking deviation, frozen layer thickness field, and inversion confidence in a rolling manner every control cycle, and re-evaluates the three judgments. When the energy-saving permission is rejected or any judgment fails during energy-saving operation, the controller exits energy-saving operation and returns to the identification window. During the exit process, the power supply waveform, electrode micro-displacement, zoned cooling flow, and carbon feeding are reverted to the pre-stored conservative trajectory set. The conservative trajectory set is calibrated by the steady-state production section and ensures that the frozen layer safety constraint and the allowable window constraint are met. The controller then reopens the identification window to rebuild the inversion confidence and restore subsequent energy-saving permission judgments, thereby completing the closed-loop degradation recovery.

[0052] Example 2 like Figure 1 and Figure 2 As shown, based on the intelligent control system for titanium dioxide smelting furnace described in Example 1, this embodiment further provides a specific feasible scheme for the acquisition unit and the identification window gating component, so that the controller can obtain a continuous, aligned data stream that can be used for subsequent state derivation key generation and cross-domain deconvolution assimilation during the system startup phase, and only allows the identification window to be opened when the security threshold and joint stability threshold are met.

[0053] The furnace gas spectral acquisition component includes a laser absorption spectroscopy measurement unit and a fingerprint generation unit. The laser absorption spectroscopy measurement unit is positioned in the furnace gas channel or flue, at a location representative of the reducing atmosphere within the furnace. It employs a tunable semiconductor laser to cover the target absorption lines of carbon monoxide, carbon dioxide, and water vapor, and simultaneously acquires the furnace gas temperature along the measurement path. In engineering implementation, the measurement unit outputs a carbon monoxide volume fraction sequence at a fixed scanning cycle. Carbon dioxide volume fraction sequence Water vapor volume fraction sequence With furnace gas temperature sequence The timestamps are generated from the same clock source and written into the acquisition frame header, and the sampling frequency is set to [value missing]. to In this embodiment, To ensure that the output volume fraction sequence can be directly used for controller calculations, the laser absorption spectroscopy measurement unit can convert the spectral scan data into a volume fraction sequence using existing spectral line fitting, integral inversion, or table lookup conversion methods. This invention is not limited to these inversion methods. The measurement unit completes calibration work such as optical path length, spectral line parameters, or table lookup configuration during the factory or installation and commissioning phases, and during operation... , , and Output the data to the controller in a uniform data format.

[0054] The fingerprint generation unit performs time alignment and denoising on the aforementioned volume fraction sequence and furnace gas temperature sequence to form a reproducibility fingerprint, and outputs the direction of reproducibility deviation and the rate of change of reproducibility. To ensure continuous and uninterrupted data processing, the fingerprint generation unit adopts a sliding window processing method: for each time step... Take the length as sliding window ,in Pick to In this embodiment, Median filtering is used within the window to suppress transient spikes, and first-order difference is used to estimate the rate of change. The reproducibility fingerprint includes at least the reproducibility exponent. Rate of change in reduction Dependence on direction Temperature characteristics The reduction index is constructed from the volume fractions of carbon monoxide and carbon dioxide: ; ; in The sampling period for the furnace gas spectrum is specified. The reduction deviation direction characterizes the trend of the current reduction state relative to the center of the allowable window, which is set by the process or given by the controller. In this embodiment, ;in and The lower and upper limits of the allowable window for reproducibility can be set by the process and can be updated by the controller according to raw material fluctuations and risk levels. The output will be based on the following relationship: Deviation of reproduction direction Used to characterize the current restoration state relative to the center of the allowed window. Deviation direction: when When the output deviates positively, When the two are equal, the output shows a negative deviation; when they are equal, the output shows a zero deviation.

[0055] For ease of implementation, The output can be obtained as a piecewise function: The fingerprint generation unit will As a fingerprint reproduction output, and will and The deviation direction of the degree of reduction and the rate of change of the degree of reduction are output to the state-derived key generation module and the cross-domain deconvolution assimilation module, respectively, for subsequent threshold generation and reaction heat source term constraints.

[0056] The furnace shell thermal field acquisition component includes a distributed fiber optic temperature measurement link and a zoned cooling water path acquisition link. The distributed fiber optic temperature measurement link is laid along the circumference and height of the furnace shell to form a continuous temperature measurement channel, outputting a continuous temperature sequence of the distributed temperature field of the furnace shell. ,in This represents the arc length coordinates along the laying path, with a spatial sampling interval of [missing information]. to The time sampling frequency is to To facilitate alignment with the partition model, the controller will Mapped to representative temperature sequences for each partition according to predefined partition boundaries. ,in , This represents the number of partitions.

[0057] The zoned cooling water circuit acquisition link synchronously acquires the cooling medium flow rate and inlet / outlet temperature difference of each zone, and outputs the cooling medium flow rate sequence of each zone. Inlet water temperature sequence With water temperature sequence The flow sampling frequency is to Temperature sampling frequency is to It is aligned with the fiber optic temperature measurement link using the same time reference. Based on this, the furnace shell thermal field acquisition component generates a zoned heat transfer sequence and outputs it to the controller and the cross-domain deconvolution assimilation module. The core relationship for generating the zoned heat transfer sequence is: ; ; ; in For the isobaric specific heat capacity of the cooling medium, when water cooling is used, take... To ensure the continuous effectiveness of the zoned heat exchange sequence, the acquisition link retains the most recent valid value for missing flow and temperature measurements and marks them with a quality tag. In the subsequent gating and assimilation stages, the controller refuses to open the identification window for data segments with invalid quality tags.

[0058] The identification window gating component generates an initial freeze layer margin during the system startup phase. It allows the identification window to open only when the initial freeze layer margin meets a preset safety threshold and the restoration fingerprint, electrical parameters, and electrode state indicators meet a joint stability threshold. The startup phase is defined as the preheating and stabilization period from system power-on to energy-saving control enabling, with a duration of [duration missing]. to In this embodiment, The identification window gating component first attempts to read the historical frozen layer thickness field record. The historical frozen layer thickness field record is the frozen layer thickness field output and persistently stored by the cross-domain deconvolution assimilation module at the end of the last stable production segment or the last energy-saving permit valid segment. and its timestamp, among which This refers to the discrete grid positions on the inner surface of the furnace shell. If a historical frozen layer thickness field record exists and its timestamp is no more than the preset validity period from the current startup time... (This embodiment takes) Then, the identification window gating component calculates the initial freeze margin using the historical freeze layer thickness field. The margin of the thinnest region is used as the basis for safety judgment: ; ; in This represents the thickness of the thinnest region in the historical frozen layer thickness field. Define the domain for the thickness field of the frozen layer. This is the lower limit of the thinnest zone thickness in the safety constraint of the frozen layer. The preset safety threshold is... In this embodiment, , It also allows for adjustments based on furnace lining thickness and operating risk level.

[0059] If no historical frozen layer thickness field record exists or the record has expired, the identification window gating component generates a conservative initial frozen layer margin based on the furnace shell distributed temperature field and the zoned heat transfer sequence. To ensure that those skilled in the art can directly implement this, the conservative margin is generated using a zoned one-dimensional steady-state heat conduction approximation combined with the zoned heat transfer sequence to calculate the zoned heat flux; for each zone... A stable window is taken at the end of the startup phase. (in The end time of the selected stable segment at the end of the startup phase is taken in this embodiment. Calculate the average heat exchange rate in each zone. Represented by average furnace shell temperature And calculate the average heat flux of the zone. : ; in For partitioning The corresponding heat exchange area can be determined and pre-calculated based on the furnace shell structure dimensions and partition boundaries. For example, the corresponding inner surface area or equivalent heat exchange area can be calculated based on the coverage of each partition in the circumferential and height directions of the furnace shell, and this area can be entered into the controller parameter table during the installation and commissioning phase. The conservative estimate of the frozen layer thickness is obtained based on the thermal conductivity and thermal resistance relationship. ; in The equivalent thermal conductivity of the frozen layer. This refers to the equivalent temperature of the slag bath inside the furnace or the equivalent temperature of the hot side inside the furnace; in this embodiment, Pick , Pick It also allows online updates based on process points. The conservative thinnest zone thickness is obtained from the conservative thickness estimates of each zone. Based on this, the conservative initial freezing layer margin is calculated. ;like If the value is below the safety threshold, the identification window gating component keeps the identification window closed and outputs a "not allowed to open" message to the controller.

[0060] Provided that the initial freeze layer margin meets the preset safety threshold, the identification window gating component further performs a joint stability threshold judgment. The joint stability threshold ensures that the data segment during the startup phase can be used for subsequent state derivation key generation and encoding injection, preventing the identification window from opening when furnace conditions fluctuate rapidly, the execution channel is unstable, or measurement quality is insufficient. The joint stability threshold includes two parts: the stability of the reproducibility fingerprint and the stability of electrical parameters and electrode state indicators; within the stability window... (In this embodiment, the inner part is taken as) ), Identify the peak value of the rate of change of the fidelity of the window gating component. With electrical parameter fluctuation index And compare it with the threshold: In the stability window Inside, the identification window gating component calculates the peak value of the rate of change in fidelity. (For example, take the window) (maximum value) and electrical parameter fluctuation index (For example, take active power) relative to its window mean (the maximum relative deviation), and compared with the threshold.

[0061] in This refers to the active power in electrical parameters and electrode state indices. The joint stability threshold condition is: and And the electrode displacement velocity satisfies In this embodiment, , , The identification window gating component sets the identification window opening permission signal and outputs it to the controller only when the initial freeze layer margin meets the preset security threshold and the joint stability judgment threshold is met simultaneously; the controller opens the identification window accordingly and enters the key and threshold generation, dual-domain encoding injection, readback verification record generation and subsequent assimilation and licensing process described in Example 1.

[0062] Example 3 like Figures 1 to 3 As shown, based on the intelligent control system for titanium dioxide smelting furnace described in Examples 1 and 2, this embodiment further provides specific feasible schemes for the dual-domain orthogonal coding excitation component, coding readback verification component, three-key permission domain issuance component, and degradation recovery component. This enables the system to complete the frequency domain isolation and periodic family isolation of the cooling domain coding sequence and the power supply domain coding sequence within the identification window and write the isolation flag into the coding readback verification record. It can calculate the cross-correlation evaluation value and output the execution uncontrollable flag when the cross-correlation evaluation value exceeds the orthogonal threshold parameter. It can refine the energy-saving permission output logic with three criteria: component decomposition residual convergence, cross-domain consistency, and execution controllability. It can trigger emergency degradation and revert to conservative power supply waveform, conservative micro-displacement, conservative cooling boundary, and conservative carbon feeding trajectory when any of the frozen layer, degree of restoration, or execution controllability exceeds the limit. This allows the identification window to be reopened to complete the inversion confidence reconstruction while maintaining the stability of the frozen layer and the degree of restoration.

[0063] The dual-domain orthogonal coding excitation component operates after the controller opens the recognition window. Its input includes the state-derived key output by the state-derived key generation component. Orthogonal threshold parameters Identify window length The current reference instruction of the execution unit is also output, consisting of a cooling domain encoding sequence, a power supply domain encoding sequence, and an isolation flag used for encoding readback verification records. This embodiment uses a discrete encoding method to generate the two-domain excitation, simultaneously satisfying frequency domain isolation and periodic family isolation. Frequency domain isolation: Within the identification window, non-overlapping discrete frequency point sets are assigned to the cooling domain coding sequence and the power supply domain coding sequence (i.e., the cooling domain uses only one set of frequency points, and the power supply domain uses only another set of frequency points, and the two sets of frequency points do not intersect); the two sets of frequency points are determined by the key. The quantities are determined in this embodiment. to Each frequency point, preferably selected One frequency point.

[0064] Periodic family isolation: Different cyclic shift numbers and phase sets are set for the two domain codes, so that the two domain codes within the same identification window maintain verifiable low cross-correlation characteristics in the time domain; the cyclic shift number and phase set are determined by the key. The derivation is determined, and the cyclic shift numbers of the two fields are guaranteed to be different.

[0065] The cooling domain encoding sequence is a partition dimension of vector sequence The power supply domain coding sequence is a vector sequence containing power supply waveform modulation coding and electrode micro-displacement coding. (in (This refers to the number of partitions, consistent with Examples 1 and 2). In this example, we take... , The injection amplitude threshold is as follows: for each partition The superimposed amplitude of the cooling zone code does not exceed the zone's baseline cooling flow rate. The amplitude of the power supply waveform modulation encoding superposition does not exceed the modulation amount of the reference power supply waveform. The superimposed amplitude of the electrode micro-displacement encoding does not exceed The above isolation method ensures that the two-domain codes do not overlap in the frequency domain and have verifiable low cross-correlation characteristics in the time domain.

[0066] The dual-domain orthogonal coding excitation component writes the isolation flag into the generation field of the coding readback verification record. The isolation flag includes at least a frequency domain isolation flag and a periodic family isolation flag. In this embodiment, the frequency domain isolation flag is defined as follows: ,in For the set of discrete frequency points assigned to the cooling domain, The set of discrete frequency points assigned to the power supply domain, and the two sets do not intersect; the periodic family isolation identifier is defined as... ,in and These are the cyclic shift numbers (or equivalent periodic family numbers) for the cooling domain and the power supply domain, respectively. For the reason The derived pseudo-random seed number; the isolation identifier and the identification window timestamp are written together into the encoding readback verification record, so that subsequent assimilation and permission determination can reproduce the encoding input in this window and perform consistency verification.

[0067] The encoding readback verification component continuously calculates the cross-correlation evaluation value within the identification window, and outputs an execution uncontrollable flag to the three-key permission domain issuing component when the cross-correlation evaluation value exceeds the orthogonality threshold parameter. In the discrete implementation, the cross-correlation evaluation value can be calculated using the normalized cross-correlation peak value of the two-domain encoding, taking into account the allowable delay range to cover the execution link latency.

[0068] To ensure that cross-correlation can still be directly calculated when the cooling domain encoding is a multi-partition vector sequence, the encoding readback verification component constructs scalar convergence sequences for the two domain encodings: the power supply domain convergence sequence is obtained by linearly combining the power supply waveform modulation code and the electrode micro-displacement code according to a preset normalization coefficient; the cooling domain convergence sequence is obtained by weighted summation of the cooling domain codes of each partition according to the partition weights, where the partition weights are... It can be obtained by zonal heat exchange sensitivity calibration (e.g., by regression identification through historical stable production section data, or by applying a small perturbation and identifying it based on the response of the zonal heat exchange sequence, provided that a safety threshold is met).

[0069] The encoding readback verification component operates within the allowable latency range (in this embodiment, the latency is taken as...). The normalized cross-correlation of the two convergent sequences is calculated within the range, and the peak value is taken as the cross-correlation evaluation value. .when At that time, the encoding readback verification component sets the execution uncontrollable flag. It is then output to the three-key permission domain issuance component, and the flag is matched with the corresponding Write the encoded readback verification record; when Time setting .

[0070] The code readback verification component simultaneously generates and maintains an instruction tracking deviation field to support execution controllability determination. The execution unit periodically reads back the actual execution volume during the identification window and energy-saving execution period, including at least the actual cooling flow rate of the partition. Actual quantity of power supply waveform modulation Actual quantity of electrode micro-displacement and the corresponding instructions issued by the controller. , , Alignment is performed on the same time base. For the readback sequence within the identification window, the encoded readback verification component calculates the cooling command tracking deviation. Power supply waveform command tracking deviation Deviation from electrode micro-displacement command tracking All measurements use the root mean square deviation (RMS) as the metric. ; ; in This represents the number of readback points within the window. The encoded readback verification record must contain at least [number of points]. , Isolation signs and as well as The record is then output to the cross-domain deconvolution assimilation component and the three-key permission domain distribution component.

[0071] The three-key permission domain issuing component receives the component decomposition residual and frozen layer thickness field output by the cross-domain deconvolution assimilation component, the restoration fingerprint and its thermal source correction change direction information output by the fingerprint generation unit, and the execution uncontrollable flag and three types of instruction tracking deviation in the encoded readback verification record. It sequentially forms the separation residual convergence judgment, cross-domain consistency judgment and execution controllability judgment, and outputs energy-saving permission when all three judgments are true, and outputs energy-saving permission rejection when any judgment is false.

[0072] The convergence criterion for the separated residuals is based on the condition that the component decomposition residuals monotonically decrease and converge stably within a preset time period. Assume the cross-domain deconvolution assimilation component outputs a discrete residual vector within the identification window. ,in The number of furnace shell temperature measurement points; the identification window is divided into... There are three equal-length sub-segments, each with a length of [length missing]. In the Segment calculation of residual energy : ; in The above expression represents the Euclidean norm. In the discrete implementation, the upper and lower limits of the summation can be rounded down by integer indices (e.g., rounding down), and the last segment can be included with the remaining sampling points to ensure that the residual energy can be directly calculated.

[0073] Let the monotonically decreasing threshold be... The convergence fluctuation threshold is In this embodiment, , , and require to meet right Established and meets the requirements and This leads to the conclusion that the component decomposition residuals monotonically decrease and converge stably.

[0074] Cross-domain consistency is determined based on the condition that the direction of change in the frozen layer thickness field and the direction of change in the thermal source correction of the reconstructed fingerprint meet a preset consistency threshold. The cross-domain deconvolution assimilation component outputs the frozen layer thickness field at the end of the recognition window. The three-key permission domain issuance component extracts the thinnest area thickness. And calculate the direction of change within a preset time period; let the preset time period be the second half of the identification window. The direction of thickness variation in the thinnest region is defined as: ; in Define the domain for the thickness field of the frozen layer. Output the sign function (positive, zero, and negative respectively). ).

[0075] The direction of thermal source correction variation for the fidelity fingerprint constraint is determined by the thermal source correction amount output by the cross-domain deconvolution assimilation component within the same preset time period. The calculation, and its direction of change are defined as: ; The consistency threshold is achieved using a combination of conditions where the directions are opposite and the correlation is no higher than the negative threshold. Considering that an increase in the equivalent contribution of the reaction heat source term usually tends to lead to a thinner frozen layer, this embodiment defines consistency as: within the same preset time period, the direction of change in the thickness of the thinnest region is opposite to the direction of change in the heat source correction amount (i.e.,...). Furthermore, the difference sequences of the two are significantly negatively correlated.

[0076] This embodiment calculates the consistency score. As a measure of correlation of differential sequences (e.g., using correlation coefficient calculation), and taking a consistency threshold. ;when and At that time, it is determined that the direction of change of the frozen layer thickness field and the direction of change of the heat source correction meet the preset consistency threshold.

[0077] The controllability determination is based on the condition that the tracking deviations of cooling commands, power supply waveform commands, and electrode micro-displacement commands, as represented by the coded readback verification records, simultaneously meet preset deviation thresholds. Simultaneously, an uncontrollable execution flag is used as a strong rejection condition. Let the preset deviation thresholds be as follows: , and In this embodiment, , , ,in and To identify the reference cooling flow rate and reference power supply waveform modulation within the window, This represents the maximum absolute value of each component of the vector. The controllability criterion is: Among the symbols This indicates a logical AND operation.

[0078] When the residual convergence determination, cross-domain consistency determination, and execution controllability determination are all met simultaneously, the three-key permission domain issuance component outputs an energy-saving license. The license timestamp, judgment result, and key indicators are written into the license record; if any judgment is invalid, the energy-saving license is rejected. It outputs a rejection reason flag to the controller, which then either maintains or reverts to the conservative trajectory and re-enters the identification window.

[0079] The degradation recovery component triggers emergency degradation when any of the following conditions are exceeded: freezing layer safety, reduction allowable window, or controllability. During emergency degradation, it reverts the power supply waveform command to a conservative power supply waveform, the electrode micro-displacement command to a conservative micro-displacement, the zoned cooling flow command to a conservative cooling boundary, and the carbon feeding trajectory to a conservative carbon feeding trajectory. This maintains freezing layer stability and controlled reduction while reopening the identification window to complete the inversion confidence reconstruction. The degradation trigger condition uses an OR logic combination: the thickness of the thinnest region of the freezing layer is lower than a preset safety threshold. Triggered by time, the criterion is The rate of change of the frozen layer thickness exceeds the preset rate of change threshold. Time-triggered, where the rate of change is approximated by difference. The criterion is (in for The sampling interval can be taken as the control period. (or assimilation output cycle); triggered when the fidelity fingerprint deviates from the allowable window, where the allowable window is... The criterion is Triggered when the cross-correlation evaluation value exceeds the orthogonality threshold parameter, the criterion is... Triggered when any instruction tracking deviation exceeds a preset deviation threshold, the criterion is as follows: or or This embodiment takes , And it is consistent with the safety constraints of the frozen layer described in Example 1.

[0080] When the degradation recovery component triggers emergency degradation, the controller operates within a degradation switching cycle. Complete the rollback within the specified time. Pick to In this embodiment, To avoid sudden changes in execution volume, rollback is implemented using linear ramp interpolation: Let the current instruction be... The conservative trajectory instruction is During the degradation period, the instructions issued are as follows: ; in The degradation trigger moment, These correspond to power supply waveform commands, electrode micro-displacement commands, zoned cooling flow commands, and carbon feeding trajectory commands, respectively. The conservative power supply waveform, conservative micro-displacement, conservative cooling boundary, and conservative carbon feeding trajectory are calibrated and stored from historical steady-state production sections, and satisfy the frozen layer safety constraints and allowable window constraints. The degradation recovery component continuously monitors during degradation. , Electrical parameters, electrode state indices, and furnace shell distributed temperature field quality markers, when the joint stability threshold described in Example 2 is met and the aforementioned degradation triggering conditions are released and maintained for a certain duration. (This embodiment takes) After that, the controller reopens the identification window and re-runs the dual-domain orthogonal coding excitation and coding readback verification process. The cross-domain deconvolution assimilation component reconstructs the component decomposition residual and inversion confidence based on this. The three-key permission domain issuance component decides whether to restore the energy-saving permission based on the judgment result of the new window, thereby completing the closed-loop degradation recovery and self-certification restart.

[0081] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A multimodal intelligent control system for a titanium dioxide smelting furnace, characterized in that, The application relates to a furnace energy-saving control system, which comprises a collecting unit, an executing unit and a controller; wherein the collecting unit comprises a furnace gas spectrum collecting assembly, a furnace shell thermal field collecting assembly and an electric parameter and electrode state collecting assembly; the furnace gas spectrum collecting assembly is used for acquiring a reduction degree fingerprint; the furnace shell thermal field collecting assembly is used for acquiring a partition heat exchange amount sequence; the electric parameter and electrode state collecting assembly is used for acquiring electric parameter and electrode state indexes; the executing unit comprises a partition cooling executing assembly, a waveform power supply executing assembly, an electrode micro-displacement executing assembly and a carbon addition feeding executing assembly; the controller comprises a state derived key generation assembly, an identification window gating assembly, a dual-domain orthogonal coding excitation assembly, a coding read-back checking assembly, a cross-domain deconvolution assimilation assembly, a three-key permission domain issuing assembly, a minimum energy consumption joint optimization control assembly and a degradation recovery assembly; the controller is configured to generate a state derived key and output an orthogonal threshold parameter according to the reduction degree fingerprint and the electric parameter and electrode state indexes, and to open an identification window based on the state derived key; the dual-domain orthogonal coding excitation assembly generates a cooling domain coding sequence and a power supply domain coding sequence in the identification window, and injects the cooling domain coding sequence and the power supply domain coding sequence into the partition cooling executing assembly, the waveform power supply executing assembly and the electrode micro-displacement executing assembly respectively, and outputs an execution uncontrollable flag when a cross-correlation evaluation value exceeds the orthogonal threshold parameter; the coding read-back checking assembly generates a coding read-back checking record representing the cross-correlation evaluation value, the execution uncontrollable flag and an instruction tracking deviation; the furnace shell thermal response is deconvolved and separated based on the partition heat exchange amount sequence and the coding read-back checking record, and the reduction degree fingerprint is used to constrain a reaction heat source term, so as to output a frozen layer thickness field and an inversion confidence; the component decomposition residual error is monotonously decreased and converged as a separation residual error convergence judgment, the change direction of the frozen layer thickness field and the change direction of the heat source correction meet a consistency threshold as a cross-domain consistency judgment, and the execution uncontrollable flag is not and the instruction tracking deviation does not exceed a deviation threshold as an execution controllability judgment, so that the energy-saving permission is issued and an energy-saving quota control boundary is formed when the three judgments are simultaneously established, otherwise the energy-saving permission is refused; under the control boundary constraint, a power supply waveform trajectory, an electrode micro-displacement trajectory, a partition cooling flow trajectory and a carbon addition feeding trajectory are output by joint optimization, and a conjugate coupling constraint is applied to the power supply waveform trajectory and the partition cooling flow trajectory, so that the two trajectories meet the frozen layer safety constraint and the allowable window constraint in the same control period; only when the conjugate coupling constraint is met, the power supply waveform trajectory and the partition cooling flow trajectory are issued, and the power supply waveform trajectory and the partition cooling flow trajectory cannot be independently solved and independently executed; when the energy-saving permission is refused or any judgment fails during energy-saving operation, the energy-saving operation is exited and the identification window is returned.

2. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The recognition window gating component is configured to read historical frozen layer thickness field records at system startup stage and generate initial frozen layer margin therefrom, generate conservative initial frozen layer margin based on furnace shell distributed temperature field and partition heat exchange amount sequence when the historical frozen layer thickness field records are absent, and only allow opening of the recognition window when the initial frozen layer margin meets a preset safety threshold and the reduction degree fingerprint and the electrical parameter and electrode state indicators meet joint stability judgment threshold.

3. The intelligent control system for a titanium dioxide smelting furnace as claimed in claim 1, wherein, The dual-domain quadrature coded excitation component is configured to perform frequency domain isolation and period family isolation on the cooling domain coded sequence and the power supply domain coded sequence based on the state derived key, and write isolation identification into the coded back read verification record, and the coded back read verification component is configured to calculate cross-correlation evaluation value within the recognition window and output an uncontrollable flag to the three-key permission domain issuance component when the cross-correlation evaluation value exceeds the quadrature threshold parameter.

4. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The furnace gas spectrum acquisition component includes a laser absorption spectrum measurement unit and a fingerprint generation unit, the laser absorption spectrum measurement unit is configured to output carbon monoxide volume fraction sequence, carbon dioxide volume fraction sequence, water vapor volume fraction sequence and furnace gas temperature sequence, and the fingerprint generation unit is configured to form a reduction degree fingerprint based on the volume fraction sequence and the furnace gas temperature sequence, and output reduction degree deviation direction and reduction degree change rate for the state derived key generation component and the cross-domain deconvolution assimilation component.

5. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The furnace shell thermal field acquisition component includes a distributed optical fiber temperature measurement link and a partition cooling waterway acquisition link, the distributed optical fiber temperature measurement link is configured to form continuous temperature sequence along the circumferential direction and height direction of the furnace shell, the partition cooling waterway acquisition link is configured to form partition cooling medium flow sequence and partition cooling medium inlet and outlet temperature difference sequence, and the furnace shell thermal field acquisition component is configured to generate the partition heat exchange amount sequence based on the flow sequence and the inlet and outlet temperature difference sequence and output to the cross-domain deconvolution assimilation component.

6. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The cross-domain deconvolution assimilation component is configured to perform component decomposition on the furnace shell distributed temperature field based on the corresponding cooling input response of the cooling domain coded sequence and the corresponding power supply input response of the power supply domain coded sequence to obtain boundary heat exchange response component and power supply deposition response component, and constrain and correct the reaction heat source term based on the reduction degree fingerprint to suppress the mutual replacement error between the reaction heat source term change and the frozen layer thickness field change, and the cross-domain deconvolution assimilation component is configured to generate the inversion confidence based on the component decomposition residual and output to the three-key permission domain issuance component.

7. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The three-key license domain issuing component is configured to take monotonic decrease and stable convergence of the component-decomposed residual error within a preset time period as a separate residual error convergence judgment, take satisfaction of a preset consistency threshold for consistency between a change direction of the frozen layer thickness field and a heat source correction change direction of the reduction degree fingerprint as a cross-domain consistency judgment, and take simultaneous satisfaction of preset deviation thresholds for tracking deviations of cooling instructions represented by an encoded readback check record, tracking deviations of power supply waveforms, and tracking deviations of electrode micro-displacements as an execution controllability judgment, so as to output an energy-saving license when the three judgments are all met and output a refusal to save energy when any judgment is not met.

8. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The minimum energy consumption joint optimization control component is configured to write a minimum thickness of a thinnest region of the frozen layer thickness field, an upper limit of a thickness change rate of the thinnest region, and an upper limit of a thickness spatial gradient into safety constraints, write an allowed window corresponding to the reduction degree fingerprint into chemical constraints, and write upper limits of a modulation amplitude and a change rate of the power supply waveform, upper limits of a modulation amplitude and a change rate of the zoned cooling flow, and upper limits of a modulation amplitude and a change rate of the carbon addition feedstock into execution constraints, so as to output the power supply waveform trajectory, the electrode micro-displacement trajectory, the zoned cooling flow trajectory, and the carbon addition feedstock trajectory under the premise of simultaneously satisfying the safety constraints, the chemical constraints, and the execution constraints.

9. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The conjugate coupling constraint defines that energy deposition modulation of the power supply waveform trajectory and boundary heat extraction modulation of the zoned cooling flow trajectory jointly satisfy the minimum thickness of the thinnest region of the frozen layer thickness field, the upper limit of the thickness change rate of the thinnest region, and the upper limit of the thickness spatial gradient, and jointly satisfy the allowed window corresponding to the reduction degree fingerprint, in the same control period.

10. The titanium dioxide smelting furnace intelligent control system of claim 1, wherein, The degradation recovery component is configured to trigger emergency degradation when the minimum thickness of the thinnest region of the frozen layer thickness field is lower than a preset safety threshold, or the frozen layer thickness change rate exceeds a preset change rate threshold, or the reduction degree fingerprint deviates from the allowed window, or a cross-correlation evaluation value represented by the encoded readback check record exceeds the orthogonality threshold parameter, or the instruction tracking deviation exceeds a preset deviation threshold, and the degradation recovery component is configured to, during emergency degradation, roll back the power supply waveform instruction to a conservative power supply waveform, roll back the electrode micro-displacement instruction to a conservative micro-displacement, roll back the zoned cooling flow instruction to a conservative cooling boundary, and roll back the carbon addition feedstock trajectory to a conservative carbon addition feedstock trajectory, so as to reopen the identification window to complete inversion confidence reconstruction while maintaining frozen layer stability and controlled reduction degree.

Images