Multi-parameter coupling intelligent regulation and control method and system in vanadium ore smelting process

Abstract

The invention discloses a multi-parameter coupling intelligent regulation and control method and system in a vanadium ore smelting process, relates to the technical field of non-ferrous metal smelting automatic control, and solves the problems of feeding deviation and resource chain scission caused by slow response of an existing heat balance model to ore impurity fluctuation, online sensing lag and process island. The impurity concentration and the dielectric constant are obtained in real time through an LIBS / XRF combined probe and a microwave dielectric spectrum, rheological disturbance is dynamically corrected through extended Kalman filtering, a dynamic coupling response model is input, a thermodynamic interference factor is calculated, and a leaching and precipitation strategy is generated; performing mother liquor tail gas material flow collaborative optimization in parallel, constructing a multi-dimensional control vector U, performing iterative updating through near-end strategy optimization in a digital twin and issuing the multi-dimensional control vector U to an edge PLC (Programmable Logic Controller) in a distributed manner, and finally correcting a model weight through online XRD (X-Ray Diffraction) feedback; according to the method, the adaptive capacity to complex vanadium ore component disturbance, the response precision of online control and the inter-process resource circulation efficiency are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of non-ferrous smelting automation control, and more particularly to a vanadium ore smelting process multi-parameter coupling intelligent regulation method and system. BACKGROUND

[0002] In the vanadium ore smelting process, multi-parameter coupling regulation is the core technology to ensure stable furnace conditions and efficient resource utilization. Traditional methods rely on empirical operations and are difficult to cope with the synergistic effects of complex ore composition and dynamic working conditions. In recent years, intelligent regulation technology based on data-driven has been gradually introduced, and through dynamic modeling and optimization algorithms, real-time matching of key parameters such as temperature field and load distribution is achieved.

[0003] Current existing vanadium ore smelting intelligent regulation mainly relies on the synergy of thermodynamic models and data-driven algorithms. Taking the blast furnace smelting link as an example, CN112342327B patent proposes a semi-empirical control strategy based on theoretical flame temperature (TFT): by establishing the quantitative relationship between blast temperature, oxygen enrichment rate, coal injection ratio and TFT (such as an increase of 1% in oxygen enrichment rate raises the temperature by 41-50℃), combining historical data to fit a general formula, and achieving rapid suppression of sudden fluctuations in furnace temperature. CN116050085A patent further integrates a heat balance calculation model to construct a heat budget evaluation rule base, and generates regulation instructions based on real-time parameters (blast pressure, coal gas composition, etc.) from the blast furnace data acquisition system to predict furnace condition trends. For the grinding-milling process, CN113762386A patent develops a working condition self-adaptive matching mechanism, based on historical grinding classification data clustering to generate typical working condition labels, dynamically adjusting parameters such as ball mill speed and acid leaching agent flow to solve the problem of fixed control limits and working condition rigidity. At the system coordination level, CN119668109A patent uses a recurrent neural network model, with target control parameters (roasting temperature, leaching pH) as the input layer and control point parameters (waste gas concentration, slag-iron ratio) as the output layer, training a smelting load intelligent regulation model to optimize the efficiency of multi-unit linkage.

[0004] However, the above-mentioned technology has exposed some problems in actual industrial application. First, the heat balance model and the working condition matching mechanism presuppose stable furnace charge composition, but the content of impurities such as silicon, aluminum and chromium in vanadium ore fluctuates sharply. The static model does not embed the thermodynamic coupling response mechanism under the interference of impurities, so that the system cannot correct the key process parameters in real time when the ore composition suddenly changes. For example, the sudden increase of silicon and aluminum impurities causes the silicon colloid to wrap the vanadium oxide in the leaching process, resulting in a significant decrease in dissolution efficiency. Second, the high dependence of intelligent algorithms on real-time data and the lag of the sensing layer form a sharp contradiction. Key parameters such as gas composition and ion valence state depend on offline detection, and the long-period analysis data are out of touch with the model requirements; the filtering delay in multi-source signal transmission further amplifies the execution deviation, and the typical performance is that the oxidation-reduction potential (ORP) electrode in the vanadium precipitation wastewater treatment is disturbed by high ammonia nitrogen, causing the reduction agent dosage to deviate from the preset range, and causing the chromium precipitation rate to fluctuate sharply. More seriously, the independent regulation and control architecture between processes leads to the breakage of the material circulation chain, such as the difficulty of linking the ammonia-rich wastewater from the vanadium precipitation mother liquor to the roasting tail gas purification unit, the ammonia resource recycling efficiency is much lower than the design expectation, and the auxiliary materials are consumed extra and the environmental load is increased. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application discloses a multi-parameter coupling intelligent regulation and control method and system for vanadium ore smelting process, aiming to solve the problems in the background art.

[0006] In order to achieve the above technical effects, the present application adopts the following technical solutions: A multi-parameter coupling intelligent regulation and control method for vanadium ore smelting process, comprising: Step 1, real-time scanning of ore stream by laser-induced breakdown spectroscopy and X-ray fluorescence combined probe to obtain impurity element concentration matrix and valence state distribution data; Step 2, based on the impurity element concentration matrix and the dielectric constant of the ore pulp measured by microwave dielectric spectroscopy in real time, the signal drift caused by the disturbance of the rheological properties of the ore pulp is dynamically corrected by an extended Kalman filter fusion algorithm, and a standardized impurity spectrum vector is output; Step 3, inputting the standardized impurity spectrum vector into a dynamic coupling response model to calculate the impurity thermodynamic interference factor, and generating acid concentration gradient, reaction time optimization instruction and precipitant dosage strategy; Step 4, synchronously collecting the ammonia ion concentration of vanadium precipitation mother liquor and the flow rate of roasting tail gas, and solving the optimal solution of ammonia-chlorine neutralization molar ratio by material flow coordination method to obtain mother liquor spraying amount and tail gas fan coordination regulation instruction; Step 5, coupling the output parameters of step 3 and step 4 to construct a multi-dimensional control vector U, and updating U in the digital twin iteration by a proximal strategy optimization algorithm to output the roasting oxygen partial pressure and the vanadium precipitation pH set value; Step 6, distribute the multi-dimensional control vector to the edge PLC based on the Storm distributed architecture, perform leaching agent dynamic dosing, and variable frequency speed regulation operation of the tail gas fan; Step 7, real-time monitoring of the slag phase composition by online XRD, comparing the twin prediction value of step 5, generating model error and feeding back to step 3 to correct the impurity weight coefficient.

[0007] As a further technical solution of the present application, a vanadium ore smelting process multi-parameter coupling intelligent regulation system comprises: a composition acquisition module for receiving LIBS / XRF probe data by the host computer, extracting impurity concentration matrix and valence state distribution through wavelet packet decomposition and principal component regression, and outputting original spectral data; A Kalman filter module is used to fuse and compensate the slurry rheological disturbance based on the original spectral data and the dielectric constant measured by microwave dielectric spectroscopy through an extended Kalman filter algorithm, and output a standardized impurity spectral vector; A coupling response module is used to calculate the thermal interference factor online by improving the Maxwell-Garnett and non-steady-state energy conservation equation, and to solve the acid concentration gradient, reaction time and dosing strategy; A substance coordination module is used to collect mother liquor Concentration and tail gas HCl flow rate, construct a linear programming model with mass conservation constraints and solve it by Lagrange duality and interior point method Molar ratio, generate spray amount and fan adjustment instruction; A strategy optimization module is used to map the coupling response and the output parameters of the substance coordination module into a multi-dimensional control vector U, and to update it iteratively in the digital twin using the PPO proximal policy optimization algorithm, and to output the oxygen partial pressure and vanadium precipitation pH setting; A distributed issuance module is used to receive the control vector based on the Storm distributed architecture, and to issue U to the edge PLC through routing, OPC-UA to Modbus TCP and back pressure mechanism; An XRD feedback module is used to obtain XRD diffraction data of the slag online, quantitatively analyze it by Rietveld algorithm, compare it with the twin prediction value, generate residual error, and feed back the impurity weight update information to the coupling response module through the gradient correction algorithm.

[0008] Based on the above technical solution, the present application has the following positive effects: This scheme couples online impurity composition analysis and self-correcting thermal dynamics, so that the impurity concentration information of silicon, aluminum, chromium, etc. directly acts on the thermal balance weight and energy conservation equation, and the sudden change of ore composition triggers the real-time reconstruction of dynamic heat flux and reaction parameters, thereby eliminating the blind area of the static model for the steady-state assumption of composition, and making the phenomenon of silica gel wrapping vanadium oxide caused by impurities in complex ores no longer cause a sharp drop in leaching efficiency.

[0009] Through multi-source signal collaborative fusion and time delay compensation mechanism, the dielectric constant, ORP, potential and impurity spectrum are online fused, the time domain complement and prediction compensation are introduced in the extended Kalman filter, the errors of sensor lag and electrode drift are corrected in time, the intelligent algorithm always outputs the dosing instruction based on the latest system state vector, so that the dosing amount of the reducing agent and the precipitant is highly consistent with the preset strategy, and the contradiction between long period offline detection and online control is eliminated.

[0010] The scheme executes the cross-process closed-loop scheduling and wastewater tail gas linkage resource reconstruction in parallel, so that the ammonia nitrogen-HCl neutralization calculation in the vanadium precipitation mother liquor and the roasting tail gas purification parameters are solved cooperatively, the mother liquor spraying and the tail gas recirculation ratio constitute a complete closed-loop material flow, the single-process island is broken, the interior point method and the Lagrange dual decomposition algorithm are coordinated to control each link, so that the ammonia resource recycling rate and the tail gas purification efficiency are improved at the same time, and the consumption of external auxiliary materials is significantly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor under the premise of not paying creative labor. Figure 1 It is a framework diagram of the vanadium ore smelting process multi-parameter coupled intelligent regulation and control method of the present application; Figure 2 It is a working method flow framework diagram of step 1 of the present application; Figure 3 It is a working principle step diagram of the extended Kalman filter fusion of the present application; Figure 4 It is a working principle framework diagram of the material flow collaborative method of the present application; Figure 5 It is a principle diagram of the vanadium ore smelting process multi-parameter coupled intelligent regulation and control system of the present application. DETAILED DESCRIPTION

[0012] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme of the present application will be described clearly and completely below in combination with the embodiments. Obviously, the described embodiments are some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0013] For the convenience of understanding the present embodiment, first, a kind of vanadium ore smelting process multi-parameter coupling intelligent regulation method disclosed in the present application is introduced in detail, please see Figure 1 The step diagram of the method of a kind of vanadium ore smelting process multi-parameter coupling intelligent regulation method is shown, and the method steps include: Step 1, by laser-induced breakdown spectroscopy and X-ray fluorescence combined probe real-time scanning ore stream, impurity element concentration matrix and valence state distribution data are obtained;Please see Figure 2 The specific working method is: first, the high-energy pulsed laser and the output beam of the multi-target X-ray tube are coaxially aligned by using an optical beam splitter, and the signal acquisition time window of the high-energy pulsed laser and the multi-target X-ray tube is aligned by a time sequence control circuit to obtain the original pulse waveform;Then, based on the ore stream velocity feedback signal obtained by the photoelectric encoder, the probe scanning step and the residence time are adjusted in real time, and the X-ray tube voltage is automatically switched to adapt to the excitation depth requirement of different particle size ores;Finally, the atomic ion emission spectrum generated by laser-induced breakdown spectroscopy and the characteristic X-ray energy spectrum of micro-focal spot X-ray fluorescence are input into a multivariate spectral deconvolution module based on non-negative matrix decomposition, the overlapping peaks are decoupled and the matrix effect is corrected to obtain the impurity concentration matrix and vanadium valence state distribution data containing spatial coordinates. The multivariate spectral deconvolution module based on non-negative matrix decomposition decomposes the original spectral matrix into basis vector matrix and abundance coefficient matrix by alternating least squares iterative optimization method, wherein the basis vector matrix contains the standard spectral library of known elements;In each iteration of the alternating least squares iterative optimization method, a matrix effect correction mechanism is used to dynamically adjust the weight coefficient of the V spectrum according to the quenching factor of the Fe element dominant spectrum area;If Cr-V spectral peak overlap is detected, the overlapping peak decoupling protocol is triggered, and the Gaussian mixture model is used to separate the overlapping peak area to obtain the pure element concentration matrix and vanadium valence state distribution data after matrix correction and peak decoupling.

[0014] It should be noted that the probe end integrates high-energy pulsed laser (wavelength 1064 nm, pulse energy adjustable range 10-100 mJ) and multi-target X-ray tube (anode voltage 20-60 kV, switchable target material Fe / Cr / Al), wherein the "multi-target X-ray tube" refers to switching the output beam between different cathode target materials in the same tube cavity by mechanical switching or electronic switching. The "combination" of "laser-induced breakdown spectroscopy (LIBS)" and "X-ray fluorescence (XRF)" in the present application refers to the spatial and time division multiplexing of the two excitation and collection optical paths in the same measurement probe head by an optical beam splitter and a reflecting prism, ensuring that the same batch of ore particles receive continuous excitation at the same spatial position. The probe head is fixed to the side of the ore conveying belt, and the probe scanning mode is transverse reciprocating scanning, and the scanning speed is coordinated with the ore conveying speed.

[0015] The excitation and collection process is synchronized by a programmable timing control unit (based on FPGA). The key parameters and trigger conditions are as follows: when the photoelectric encoder feedback belt speed exceeds the threshold (typically 0.5 m / s), the "probe residence time" = is calculated by the FPGA in real time according to the beam spot length (selectable range 50-200 μm), specifically: ; wherein and the measurement accuracy is determined by the laser jitter detection module and the encoder resolution, respectively. The value triggers the FPGA to align and adjust the collection window width of the laser pulse and the X-ray tube output voltage , so that the original pulse waveform collection of LIBS and XRF signals at the same particle is completed in sequence. When exceeds the above range, the FPGA triggers the "low-speed mode" or "high-speed mode" and adjusts the excitation energy and residence time accordingly, so that sufficient signal intensity can be obtained each time. The scanning control logic is based on the internal timer of the FPGA, which accurately controls the probe transverse step size by calculating the sampling points and trigger interval of each scanning period , wherein is the transverse speed, is the trigger period, which is determined by the on-site mechanical arm motion accuracy and belt speed fluctuation.

[0016] In terms of timing control and trigger mechanism, field programmable gate array (FPGA) is responsible for the core timing scheduling function. The FPGA triggers the laser to emit a high-energy pulse and synchronously opens the analog front-end collection channel to capture the atomic and ionic emission spectrum formed by laser-induced breakdown when the probe moves to a new sampling point according to the input photoelectric encoder signal. The collection window length is determined by the pulse width and spot decay time, and is usually set to tens of nanoseconds. After the laser collection is completed, the X-ray tube output high-voltage pulse is triggered after a delay (smaller than 10 nanoseconds), and the XRF collection channel is opened to obtain the characteristic X-ray spectrum. This delay mechanism ensures that the two excitations do not interfere with each other and have time correlation. The FPGA simultaneously monitors the laser energy and X-ray tube voltage current signals. If the monitoring value deviates from the calibrated range (such as the laser pulse energy deviation exceeding ±5%), the collection is immediately suspended and a correction warning is issued, and the upper computer or maintenance personnel intervenes to adjust.

[0017] The signal acquisition and preprocessing part is equipped with a multi-channel high-speed analog-to-digital converter (ADC) at the probe end, with a sampling rate of not less than 100 MS / s per channel and a quantization accuracy of 12-16 bits. The raw waveform data collected are first stored in the FPGA ring buffer, and then packaged according to the timestamp, spatial coordinates and trigger ID, and transmitted to the backend server through the PCIe or Ethernet bus. Data preprocessing includes baseline correction, low-pass filtering and power frequency noise suppression, using a finite impulse response filter (FIR). The filter coefficients and cutoff frequency are flexibly configured according to the on-site electromagnetic environment and signal bandwidth characteristics, and are not limited.

[0018] In terms of multivariate spectral deconvolution algorithms, the backend server organizes the processed time-domain waveform matrix into a two-dimensional spectral matrix S, with rows corresponding to discrete wavelength channels (W channels) and columns corresponding to N excitation sampling points. The algorithm uses an improved non-negative matrix factorization (NMF) method, with an additional L1 norm sparsity regularization added to the traditional squared error term to improve the resolution of trace spectral lines. Through alternating least squares (ALS) iteration: first fix the abundance coefficient matrix H, and only optimize the non-negative matrix W, then fix W and update H, each iteration sub-problem can be solved using multiplication update or gradient projection method, and the iteration termination condition can be set to the relative change of decomposition residual below 10⁻ 4 or reach the maximum iteration number of 50-100 times, which can be determined according to the calculation resources and accuracy requirements, and is not limited. The column vectors of the basis vector matrix W correspond to the calibrated element standard spectral line library, including V, Si, Al, Cr, Fe and other main spectral lines and their typical compound spectral regions.

[0019] The matrix effect correction is an internal sub-process of the NMF module, which extracts the average attenuation factor η_Fe of the Fe-related spectral region (such as the range of 640-660 nanometers) in the basis vector matrix W after each ALS iteration, and dynamically corrects the V spectral line corresponding weight using a linear weighting mechanism where κ is an empirical correction coefficient, with a value range of 0.1-0.5, which can be determined according to the on-site sample test, and is not limited. This mechanism compensates for the suppression or enhancement effect of the slag matrix on the V emission peak, making the V concentration calculation more accurate.

[0020] When the ALS iteration process detects that the degree of overlap of a certain row of the abundance coefficient matrix H corresponding to Cr and the V spectral peak exceeds the set threshold value δ (for example, 30%), the system automatically triggers the overlapping peak decoupling protocol. The protocol first extracts the spectral cross-section curve in the overlapping band and calls the Gaussian mixture model (GMM) for multi-peak fitting. The GMM fitting estimates the peak separation parameters (weight π_l, center wavelength μ_l, variance σ_l) by the expectation maximization (EM) algorithm and outputs the pure peak area after parameter convergence, which is used to replace the abundance value of the waveband in the original ALS result. The decoupling iteration order and the number of EM iterations can be determined according to the spectral peak width and signal-to-noise ratio, and are not limited.

[0021] After deconvolution and correction, the system obtains the pure impurity abundance coefficient matrix H_clean, and then constructs the spatial coordinate mapping. The FPGA continuously records the probe horizontal position and trigger number during the scanning process, and transmits it together with the spectral data, and the back-end server can recover the two-dimensional spatial coordinates (x, y) of each sample point according to the trigger number. The valence state distribution is calculated by the peak area ratio of different valence V spectral lines (such as V 2 ⁺ at 480nm, V 3 ⁺ at 492nm, V 4 ⁺ near 504nm), and together with the impurity concentration matrix to form the final output data set.

[0022] Finally, the impurity concentration matrix C_imp(x, y) with spatial coordinates and the valence state distribution vector C_valence are written into the central control unit through shared memory or cache, and the upper computer transmits them to the subsequent Kalman filter fusion and coupled response module through the OPC-UA protocol. The OPC-UA session uses a secure channel and a subscription mode, and the heart rate cycle and the historical data retransmission strategy can be flexibly adjusted according to the network bandwidth and the control cycle, and are not limited. If the heart rate is overdue or the communication is interrupted, the system automatically caches the latest data and reconnects according to the cycle until it is restored.

[0023] The parameters and tools involved in each step of the embodiment include but are not limited to: optical beam splitter insertion loss <1dB, FPGA timing jitter <5ns, ADC quantization precision 12bit, ALS maximum iteration number 100 times, GMM convergence threshold 10⁻ 4 , OPC-UA heart rate cycle 1s, etc. They can be flexibly configured according to engineering needs, and are not limited.

[0024] Step 2, based on the impurity element concentration matrix and the real-time measured dielectric constant of the ore pulp, the signal drift caused by the disturbance of the rheological characteristics of the ore pulp is dynamically corrected by an extended Kalman filter fusion algorithm, and a standardized impurity spectrum vector is output; It needs to be noted that step 2 is applied to the dynamic treatment of vanadium ore smelting slurry (such as the conveying pipeline before leaching, the stirring reaction tank and other slurry flow areas). In this scenario, due to changes in flow rate, particle aggregation / dispersion, temperature fluctuations and other factors, the rheological properties of the slurry (such as viscosity, shear rate, flow rate distribution) continue to change dynamically, which can cause distortion of the laser-induced breakdown spectroscopy (LIBS) spot energy, changes in the electromagnetic field coupling efficiency of the microwave dielectric spectrum, and thus deviations of the impurity concentration matrix and the dielectric constant measurement values from the true values (i.e. "signal drift"). If the drift is not corrected, the dynamic coupling response model in step 3 will generate invalid process instructions based on incorrect input, leading to reduced leaching efficiency and waste of precipitating agent.

[0025] It should be noted that Kalman filtering is a recursive algorithm for state estimation of dynamic systems, and the extended Kalman filter (EKF) fusion algorithm is an extension of its nonlinear scenario, which realizes iterative updating of state and covariance through local linearization (Jacobian matrix); "fusion" refers to the simultaneous use of LIBS impurity concentration and microwave dielectric spectrum dielectric properties to correct each other through the EKF framework. Unexpected changes in the flow state of the slurry (laminar / turbulent flow switching, flow rate gradient changes) or physical properties (viscosity, shear modulus) are the core causes of "signal drift" (such as a jump in pipe flow rate from 1 m / s to 1.5 m / s, a change in viscosity from 10 mPa·s to 15 mPa·s, etc. The specific threshold value depends on the process accuracy requirement and can be determined by the user). The standardized impurity spectrum vector, i.e. the characteristic vector obtained by integrating the impurity concentration matrix and the dielectric properties in a unified dimension (such as concentration normalization and dielectric constant standardization) after EKF correction, has a "standardization" that reflects the elimination of data dimension and drift error, and can be directly input into the subsequent model.

[0026] The core input of step 2 is the impurity concentration matrix at time k and the dielectric constant of the slurry at time k : where The LIBS system scans the slurry flow to obtain: LIBS excites the plasma by laser, collects the emission spectrum and deconvolves (such as polynomial fitting, wavelet denoising, algorithm selection depending on device or process requirements, which can be determined by the user), and obtains the concentration matrix of impurities such as silicon, aluminum, chromium (the matrix dimension is determined by the spatial resolution of LIBS, such as 10x10 pixels, the number of pixels can be adjusted according to the detection accuracy, which can be determined by the user). The microwave dielectric spectrum system measures: the microwave probe (coaxial / waveguide probe) is inserted into the pipeline, emits a signal of a specific frequency (such as 2.45 GHz, 9.15 GHz, selected according to the dielectric properties of the slurry and the penetration depth, which can be determined by the user), and calculates the dielectric constant through the reflection / transmission coefficient (the calculation method is selected according to the probe type, such as transmission line model, resonant cavity model, which can be determined by the user). As shown in Figure 3 The specific execution logic of the extended Kalman filter fusion in the implementation is as follows: s100, at time k, will and Organization as a measurement vector EKF iteration triggering and rheological disturbance linkage: When the slurry flow rate changes by more than 5% / s (e.g., the flow meter shows the flow rate changing from 1.2m / s to 1.3m / s, the change rate is approximately 8.3% / s) or the viscosity fluctuates by more than 10% (e.g., the viscometer reading changes from 12mPa·s to 14mPa·s, the fluctuation is approximately 16.7%) (the trigger threshold is set according to the process tolerance and can be determined by the user), the EKF encryption iteration frequency is increased (e.g., from 1Hz to 5Hz, the frequency strategy is implemented according to PID or rule base and can be determined by the user); if the disturbance is mild (flow rate fluctuation <2% / s, viscosity fluctuation <5%), the base frequency (e.g., 1Hz) is maintained to balance resources and accuracy.

[0027] s200, the prediction phase is based on the state of the previous time step. External disturbances Derivation of prior states With prior covariance: where, state vector Defined as the "impurity concentration-rheological perturbation coupling state", it includes the concentrations of silicon, aluminum, and chromium (cSi, cAl, cCr) and signal drift parameters (drift amplitude a, velocity v), i.e. = (The dimension can be adjusted according to the type of impurities and the complexity of the drift model, and can be determined by yourself).

[0028] External disturbances Measurable disturbances such as the change in slurry pump frequency Δf and the change in agitator speed Δn are collected by sensors (frequency converter, tachometer) and converted into disturbance factors (e.g., for every 1Hz increase in Δf, the predicted value of cSi increases by 0.05%, and the mapping relationship is determined by process modeling and can be determined by oneself).

[0029] The mapping function f(⋅) is used to describe the state evolution law. If it is assumed that the concentration drifts linearly with rheology, then Where A is the state transition matrix (diagonal elements such as 0.99 represent the concentration decay coefficient, and off-diagonal elements such as 0 represent state independence, determined based on historical data, which can be determined manually), and B is the perturbation input matrix (elements represent...). The weights of the influence on the state can be determined by the experiment (based on calibration). The process noise (covariance Qk is determined by LIBS signal wavelet packet decomposition: the spectrum is decomposed into n layers of wavelet packets (e.g., 3 layers, the number of decomposition layers is selected according to the signal bandwidth, which can be determined by yourself), the sub-band energy covariance is calculated, and it is used as the initial value of Qk and updated in real time).

[0030] s300, introduce MG dielectric mixing equation (formula 1: ) for fusion of "impurity concentration" and "dielectric constant", map "impurity concentration prior prediction value " to "dielectric constant prediction value ": wherein, is the dielectric constant of the slag phase matrix (such as the dielectric constant of vanadium slag matrix at 25℃, 1atm, which can be determined by offline test calibration); is the dielectric constant of impurity particles (such as silicon =11.7, aluminum =9.0, chromium =14.0, which can be determined by material database or experimental measurement, and can be determined by oneself); ff is the volume fraction of impurities (converted from mass fraction c, such as mass fraction , (silicon 2.33g / cm 3 , aluminum 2.70g / cm 3 , chromium 7.19g / cm 3 ) and (such as 3.5g / cm 3 ), which can be determined by experimental measurement, and can be determined by oneself).

[0031] The Jacobian matrix is the partial derivative matrix of the MG equation to , that is, . Since is a function of , nonzero elements correspond to the partial derivative of each impurity concentration to (such as represents the change of when the silicon concentration changes by 1%). The Jacobian matrix is realized by numerical differentiation (such as finite difference method, step size 0.01% concentration change, step size can be adjusted according to accuracy, and can be determined by oneself) or analytical derivation. The role is to linearize the MG equation locally and provide a basis for EKF measurement update.

[0032] s400, realize adaptive adjustment of process noise covariance and measurement noise covariance through wavelet packet decomposition and sequence variance analysis: wherein the process noise covariance is the wavelet packet decomposition of the LIBS spectrum signal (such as selecting db4 wavelet basis, decomposing 3 layers, and selecting wavelet basis and number of layers according to time-frequency resolution and calculation amount, which can be determined by oneself), to get the energy sequence of each frequency sub-band; calculate the sub-band energy covariance matrix, and take the diagonal elements as the update value (high-frequency sub-band covariance is large, corresponding to fast-changing noise; low-frequency sub-band covariance is small, corresponding to slow-changing noise). That is, the variance of the microwave dielectric spectrum historical measurement sequence is calculated (such as taking the last 10 sampling points as a sliding window, the window size is adjusted according to the signal stationarity, and can be determined by oneself), and The diagonal element (dielectric constant measurement noise variance) is set as σ ε 2 ; if the LIBS concentration sequence variance σ c 2 changes significantly (such as increasing by 50%), the The middle concentration measurement noise variance element is adjusted synchronously to ensure that the multi-source measurement noise is fully reflected.

[0033] s500, the measurement update stage realizes the optimal fusion of the "a priori prediction value" and the "actual measurement value" through the Kalman gain.

[0034] s600, the standardized impurity spectrum vector is finally obtained through the Kalman update equation.

[0035] In practical application, the LIBS scanning frequency, microwave dielectric spectrum bandwidth, EKF iteration period, etc. need to be adjusted according to the site working condition (such as a certain pilot line debugging for 3 months, increasing the LIBS frequency from 5Hz to 8Hz, and reducing the concentration measurement delay from 200ms to 150ms, the optimization target is determined according to the response speed of the production line, and can be determined by oneself). For special minerals (such as high calcium vanadium ore) or extreme rheological scenes (such as ore slurry containing fibrous impurities), the EKF state equation, mapping function, etc. need to be customized (such as introducing a nonlinear transfer model, increasing a fiber dielectric correction term, and modifying the direction according to the mineral characteristic analysis, which can be determined by oneself).

[0036] Step 3, input the standardized impurity spectrum vector into the dynamic coupling response model, calculate the impurity thermodynamic interference factor, generate the acid concentration gradient, reaction time optimization instruction and precipitant addition strategy; the dynamic coupling response model includes an input preprocessing layer, a feature fusion layer, a thermodynamic factor calculation layer, a dynamic coupling analysis layer, a multi-objective optimization layer and an instruction synthesis layer.

[0037] In implementation, the dynamic coupling response model is deployed in the blast furnace smelting monitoring center, and constitutes a closed loop with the upper EKF fusion module, the lower material coordination and strategy optimization module, and the edge execution layer. The standardized impurity spectrum vector output by the EKF fusion module and the online dielectric constant time sequence are used as the data source of the input preprocessing layer; the "online dielectric constant measurement" in the system uses a coupled electrode array and a network analyzer to realize the dielectric measurement of the impurity-water-solid three-phase mixture, and returns the dielectric constant in complex form, only the real part of which is taken for subsequent processing in this paper. All "frequency" and "sampling rate" parameters can be set according to the on-site device capability, and no limitation is made.

[0038] The input preprocessing layer first performs a normalization transformation on the raw input, which uses a min-max linear mapping to map the vector components to the interval [0, 1] to eliminate the influence of different physical magnitudes. The minimum and maximum values are automatically read from the field calibration test or historical database, and the system provides a real-time update interface that can supplement new calibration samples as needed. After normalization, the signal is subjected to three levels of wavelet packet denoising in the time series. This process uses the Daubechies wavelet basis and automatically sets the threshold after each layer of decomposition based on noise estimation indicators, removes the detail coefficients above the threshold, and reconstructs the approximate original signal through energy conservation. The threshold calculation uses a robust extreme value estimation method: that is, the median of the absolute values of the high-frequency coefficients of the first layer is multiplied by an empirical factor of 1.4826 to obtain an estimate of the noise standard deviation, and then multiplied by the threshold multiple (default 3), which can be adjusted through the background configuration interface. The trigger condition for this layer is to execute immediately after the input data collection is complete, and the action is automatically scheduled by the preprocessing scheduler, and the output is accompanied by a noise confidence index indicating the reliability of the denoised signal.

[0039] The preprocessed time series feature vector is written to the high-speed shared memory and notifies the feature fusion layer to load. The feature fusion layer processes the multi-modal features within the time window in parallel based on the multi-head attention mechanism. The window length L is by default equal to five times the number of samples corresponding to the control period, and can be set to between 5 and 20 on the parameter interface, and can be determined by the system delay and real-time requirements, which is not limited. In each attention head, the system calculates the query-key-value matrix based on the fusion features and historical window data, and the dimensions are determined by the number of heads h set during model training and the hidden dimension of each head determination, with typical values of h = 8, = 64. The attention weight calculation uses the scaled dot-product mechanism and is normalized by the softmax function. If the maximum attention weight calculated by any head is lower than the preset threshold τ = 0.1, the system automatically calls the adaptive re-calibration mechanism - this mechanism will collect the original normalized data of the last Nr = 10 windows, re-estimate the projection matrix parameters of the head according to the least squares method with regularization, and verify whether the attention peak value has risen within three fine-tuning iterations. The re-calibration process is executed in parallel on the GPU, and the duration is usually no more than 50 milliseconds to avoid blocking the real-time control flow. All head outputs are spliced and passed through a linear mapping layer to generate fusion features , and the attention heat map is stored in the performance monitoring library for subsequent debugging and visualization.

[0040] The fusion features are passed to the thermodynamic factor calculation layer, which calls the built-in thermochemical data interface to calculate the thermodynamic factors based on the current operating parameters, mainly including the real-time measured furnace temperature and pressure and impurity concentration vector [c1, …, c5] - online retrieve corresponding standard enthalpy change and standard entropy change Data. Enthalpy and entropy database is obtained by experimental calibration or literature calibration, and the data accuracy can reach 0.1%. The index in the table is the key value of the impurity chemical formula and the working condition interval, which can be determined according to the actual situation.

[0041] After retrieval, the thermodynamic factor calculation layer performs temperature correction to convert the standard state enthalpy and entropy to the current working condition. Enthalpy correction is approximated by linear temperature difference: Wherein is the specific heat capacity of the th impurity at constant pressure, is the standard reference temperature; the entropy correction adopts the constant pressure-constant temperature approximation: The two-step calculation is realized by vectorization in the numerical calculation module, and the delay control is within tens of milliseconds, which meets the second-level update requirement. After the enthalpy and entropy values are corrected, the system maps the contribution value of each impurity purification according to the Gibbs free energy disturbance calculation formula: The formula is executed in the realization layer with high-precision floating-point operation, and the output vector [ΔG1, …, ΔG5] represents the thermodynamic interference factor of each impurity on the reaction system. The thermal disturbance gradient is calculated simultaneously in the layer, and the sensitivity matrix of adjacent concentration disturbance to ΔG is obtained through numerical difference, which is used for local grid mapping in the subsequent dynamic coupling analysis layer. The trigger condition is that whenever the fusion feature vector or the online temperature is updated, the calculation is performed immediately; the action scheduling is completed by the thermal factor scheduling unit, which broadcasts the calculation start signal through the event bus and writes the results to the shared memory of the coupling analysis layer after the calculation is completed.

[0042] The dynamic coupling analysis layer couples the thermal disturbance factor vector with the pre-established three-dimensional reactor geometry mesh in the finite element method. The geometry model is generated offline by CAD and CFD tools and imported into the finite element solver, which includes mesh node coordinates, fluid and solid domain boundary conditions, and fluid dynamics parameters such as viscosity and density. In the layer, first, interpolation is performed according to the distribution of thermal disturbance factors on the mesh nodes to assign local energy disturbance values to each node; then, under the constraint of boundary conditions including flow velocity field and pressure field, the built-in weak coupling solver for non-steady-state heat transfer and mass transfer in the software is used to discretely calculate the partial differential equation according to the Galerkin method. During the solving process, the system monitors the residual convergence, and if the iteration residual does not decrease continuously for multiple generations or reaches the maximum iteration number of 100, the degradation mechanism is triggered, which automatically switches to a one-dimensional pipe model simulation to quickly obtain an approximate result. After the simulation is completed, the local acid concentration gradient ∇[H+] distribution map and the mass transfer resistance coefficient Φmix distribution map are saved as a vector and a matrix format and written to the cache area of the multi-objective optimization layer.

[0043] The multi-objective optimization layer links the coupling analysis results with the precipitator model to build a three-objective optimization problem: minimum acid concentration gradient variance, shortest reaction time, and least precipitator consumption. The algorithm uses NSGA II, which is implemented in the optimization engine through the following process: First, generate an initial population centered on the previous period instruction Uk−1 with a small perturbation. The population size is usually set to 50 to 150 individuals, and the individual dimension corresponds to the acid concentration rate, reaction time, and dosage, which can be automatically adjusted according to system load during runtime. The "perturbation amplitude" is determined by the adaptive noise scheduler based on system stability indicators, and the noise is generated based on Gaussian distribution. During each generation evolution process, the algorithm stratifies the population by non-dominated sorting, dividing individuals into levels according to their dominance relationship in the target space; then, use the crowding distance to select diversity in the same level; the crossover operation uses simulated binary crossover (SBX) to cross the control variables of parent individuals according to probability; the mutation operation uses a polynomial mutation method, and automatically increases the mutation probability when the population convergence degree is lower than the threshold value of 0.02, to prevent premature convergence. The entire iteration process is executed under a parallel computing framework, and the crossover and mutation operations can be completed concurrently on multi-core CPUs or GPUs.

[0044] There are two iteration termination conditions: reaching the maximum number of generations (e.g., 100 generations) or the main body crowding degree of the Pareto frontier changing by less than a preset threshold in consecutive g generations. After termination, the algorithm outputs the Pareto frontier solution set to the strategy synthesis unit and selects a representative solution according to the current production priority (such as prioritizing yield or minimizing reagent consumption). During optimization, the system continuously monitors the algorithm convergence curve and performance indicators and triggers a global restart mechanism to regenerate the population to restore the search range when the population diversity decreases too quickly.

[0045] The instruction synthesis layer selects a representative solution from the Pareto solution set according to production priority rules, and performs a weighted linear combination of three control variables: acid concentration adjustment rate, reaction residence time, and precipitant dosage. The combination weights are issued by the production management module through the interface and can be set to different modes such as output priority, energy consumption priority, or resource recovery priority. The synthesized control sequence is framed according to the control cycle. Each frame is encapsulated by the OPC UA protocol and sent to the edge PLC for parallel execution with the mother liquor spraying and tail gas regulation instructions generated in step 4. The execution layer interfaces with the corresponding dosing pump station and valve actuator. After receiving the instruction, it drives the hardware in closed-loop PID mode. The execution feedback signal is uploaded in real time through the fieldbus and incorporated into the EKF fusion input of the next cycle.

[0046] Step 4: Simultaneously collect the ammonia ion concentration in the vanadium precipitation mother liquor and the hydrogen chloride flow rate in the roasting tail gas. Solve for the optimal ammonia-chlorine neutralization molar ratio using a material flow coordination method to obtain the coordinated adjustment command for the mother liquor spray rate and the tail gas blower. Please refer to [link to relevant documentation]. Figure 4 Specifically, the material flow coordinated method monitors the ammonia ion concentration in the vanadium precipitation mother liquor online using an ion chromatograph, simultaneously measures the hydrogen chloride concentration in the roasting tail gas in real time using a tunable diode laser absorption spectroscopy, and collects the waste gas volumetric velocity using a turbine flow meter to construct a gas-liquid two-phase flow field model. Based on the Reynolds-Navier-Stokes equations, the trajectory of the spray droplets and the diffusion path of the waste gas are calculated. Then, a sequential quadratic programming algorithm is used to solve for the optimal ammonia-chlorine molar ratio, with the objective function set at a molar ratio of 1.05 ± 0.02. If the ammonia ion concentration fluctuates beyond ± 5%, the diaphragm pump frequency is adjusted in real time using an incremental PID control protocol. When the waste gas flow rate change rate exceeds a preset threshold, the induced draft fan speed is pre-adjusted through a feedforward compensation mechanism. It should be noted that "online monitoring" refers to continuous data collection and real-time output during the vanadium ore smelting process, which differs from periodic offline sampling; "ammonia ion concentration" specifically refers to the concentration of ammonia ions in the vanadium precipitation mother liquor. The content was obtained by quantitative analysis using ion chromatography; "Diode Laser Absorption Spectroscopy (TDLAS)" achieves quantitative monitoring by utilizing the relationship between absorption intensity and concentration at a specific wavelength. This scheme uses a tunable semiconductor laser to measure the HCl absorption spectrum in real time at the 1.72µm absorption band; "Turbine Flow Meter" is used to determine the exhaust gas volumetric flow rate, and its output is Nm³ at standard temperature and pressure. 3 / h; "Two-phase flow field model" refers to a numerical simulation environment that couples the motion of the gas phase (exhaust gas) and the liquid phase (spray droplets); "Sequential quadratic programming (SQP)" is an iterative algorithm for constrained nonlinear optimization problems, used to solve the optimal dosing strategy within the target molar ratio range; "Incremental PID control protocol" refers to adjusting the controller output according to the deviation increment to correct the pump frequency when the deviation exceeds the limit; "Feedforward compensation" refers to adjusting the induced draft fan speed in advance according to the measured rate of change of exhaust gas flow to reduce system response time delay.

[0047] In practical implementation, the system first continuously extracts vanadium precipitation mother liquor samples using an ion chromatograph, and then dilutes the mother liquor to a suitable concentration range for chromatographic detection using an automatic online diluent within the sampling flow path (common dilution ratios are 1:10 to 1:100, depending on the ion concentration range). The chromatograph employs gradient elution and a high-efficiency ion exchange column, operating at a temperature of 40 ℃ and a flow rate of 1.0 mL / min. A suppressor and conductivity detector are connected post-column, enabling one analysis and output per minute. Concentration C_NH4 (mg / L or mmol / L). When the change in C_NH4 exceeds ±5% relative to the previous sampling period, the system triggers the deviation monitoring submodule of the PID control protocol to adjust the diaphragm pump frequency in real time.

[0048] Simultaneously, a tunable diode laser absorption spectrometer was installed in the calcination exhaust gas pipeline. This instrument swept the frequency in the wavelength range of 1.712–1.716µm, and measured the light intensity. and Calculate HCl concentration by ratio (ppm or mg / m 3 The fundamental relationship follows the Beer-Lambert law: In the formula and These represent the incident and transmitted light intensities, respectively. This represents the absorption cross-section of HCl at that wavelength, typically taken as 1.2 × 10⁻⁻⁻⁶. 20 cm 2 / molecule, The optical path length is specified, and the probe is installed along the diameter of the exhaust pipe. The optical path length is selectable from 25 to 100 mm depending on the pipe diameter. The measurement is continuously updated at millisecond-level speeds, and the data is sent to the main control system via serial port or Ethernet.

[0049] Exhaust gas volumetric flow rate Qgas (Nm 3 The acquisition of the flow rate ( / h) is accomplished by a turbine flow meter. The impeller speed signal of the flow meter is converted into frequency by a photoelectric sensor, and then immediately converted into volumetric flow rate through the calibrated frequency-flow relationship. To ensure measurement accuracy, the flow meter is pre-installed with a heat exchanger and a straight pipe section interface. A straight pipe section of 10 times the pipe diameter is reserved at both the fluid inlet and outlet to reduce turbulence interference. The operating temperature and pressure compensation module converts the raw data to standard conditions.

[0050] The three data streams mentioned above are centralized in the main control system and merged into a synchronous sampling group via a high-speed data bus. The data set first enters the two-phase flow field model simulator. The model is based on the Reynolds-averaged Navier-Stokes equations, which describe the gas phase flow field in the following form: and coupled with the droplet motion through Lagrangian trajectory tracking, whose equation is as follows: where is the gas phase density, is the viscosity, is the gas phase velocity field, is the droplet velocity, is the droplet mass, is the drag coefficient, includes the body force and the reaction force term. The present embodiment discretizes the reactor geometry into a three-dimensional grid and solves the gas phase velocity field on the grid nodes using the finite volume method, while generating the initial position and diameter distribution of the droplets (common diameter 50-200 pm, distribution can be set according to the atomization characteristics of the sprayer) through random sampling, and then updating the droplet trajectory along the time step. The model needs to be run in parallel in a GPU or multi-core CPU cluster to achieve 1 second level update.

[0051] After completing the two-phase flow field simulation, the system extracts the current The spray droplet-waste gas coupling field indicators are constructed based on the current The above indicators are used as the input of the sequential quadratic programming (SQP) algorithm. The SQP formulates the neutral mole ratio optimization problem in the following constraint and objective function form: The mole ratio target should fall within the interval 1.05±0.02. When solving the SQP, the Lagrangian function is constructed and the constraints are linearized and the objective function is quadratic approximated in each iteration, the search direction is updated using the Newton method, and the new solution is obtained through step search. The iteration termination condition is that the change of the objective function is less than 10⁻6 or the number of iterations exceeds 50.

[0052] When the SQP algorithm is completed, the system obtains the optimal coefficients α∗, β∗. The mother liquor spray quantity Qliq is calculated as: ; the tail gas fan suction speed Qfan is calculated as: ; and both are issued as coordinated regulation instructions to the diaphragm pump controller and the fan frequency converter.

[0053] If the NH4⁺ concentration fluctuation exceeds ±5% (compared to the previous sampling value) in the above online monitoring, the system immediately starts the incremental PID control protocol to adjust the frequency of the primary and secondary liquid diaphragm pumps with the incremental correction defined as: wherein is the concentration incremental deviation, Kp, Ki, Kd are PID parameters. The PID controller is implemented in a differential manner, directly outputting incremental instructions to the pump frequency converter. At the same time, when the rate of change of the exhaust gas volume flow rate Qgas exceeds ±3%, the system adds a feedforward compensation mechanism in the fan control protocol: according to the rate of change, the next cycle flow rate is estimated in advance, and the required fan speed adjustment amount is calculated , which is combined with the basic instruction Qfan obtained by SQP to eliminate system response lag, specifically: wherein is the feedforward gain coefficient, usually set between 0.5 and 1.5, which can be determined according to the fan characteristics and pipe network resistance; this action is executed immediately by the feedforward control unit after receiving the flow rate change signal, and is combined with the incremental PID instruction before being issued.

[0054] Step 5, coupling the output parameters of step 3 and step 4 to construct a multi-dimensional control vector U, updating U through near-end strategy optimization algorithm in digital twin, and outputting roasting oxygen partial pressure and vanadium precipitation pH set value; in implementation, the multi-dimensional control vector U includes furnace condition control vector, leaching parameter vector, precipitation adjustment vector, tail gas purification vector, wastewater reuse vector and model correction vector, wherein the furnace condition control vector includes roasting section blast flow, furnace oxygen partial pressure set value and fuel injection rate; the leaching parameter vector includes acid concentration gradient and reaction time of each stage in the leaching section; the precipitation adjustment vector includes precipitation section pH set, precipitation agent addition rate of each stage and stage switching trigger threshold; the tail gas purification vector includes tail gas extraction rate and recycling ratio; the wastewater reuse vector includes mother liquor spraying amount and ammonia nitrogen recovery flow rate; the model correction vector is the impurity weight coefficient update rate based on online XRD feedback, the adaptive gain of heat balance model and the rheological disturbance compensation factor. The above sub-vectors realize data mapping and flow through the unified digital twin state interface, and the outputs of the furnace condition control vector, the leaching parameter vector and the tail gas purification vector are used as the inputs of the precipitation adjustment vector, the wastewater reuse vector and the model correction vector, and finally iterated in the near-end strategy optimization algorithm to drive the PLC edge to perform multi-linkage control such as leaching addition, tail gas speed regulation and weight self-correction.

[0055] It should be noted that in this scheme, "near-end policy optimization algorithm" specifically refers to a reinforcement learning algorithm that iteratively updates policy network parameters based on the trust region concept in offline or online digital twin environments; "importance sampling mechanism" refers to weighting samples according to the probability ratio of actions before and after policy update to reduce variance when collecting policy gradients; "probability ratio exceeding confidence interval threshold" means that if the probability ratio of the same action output by the new and old policies deviates from the interval [1–ε, 1+ε], it needs to be pruned; "generalized dominance function" refers to smoothing the state-action value deviation through a temporal difference estimation algorithm; "Pareto optimal" in this scheme refers to the optimal solution between the dual objectives of maximizing vanadium recovery rate and minimizing energy consumption, where one cannot be improved without harming the other.

[0056] In implementation, the algorithm execution flow is triggered by the central control unit in offline or servo online mode: when the output parameters of step 3 or step 4 change significantly (i.e., the change in the impurity thermal disturbance factor vector ΔG exceeds 10% or the neutralization molar ratio deviates from the optimal ±0.02), the system triggers the strategy update sub-flow. First, data from k control cycles within the window are collected. Cached, of which This represents the current state of the digital twin (including simulation outputs such as ΔG, gradient field, and flow rate). This is the action vector (U component) executed at that time. The reward function value is given immediately. The combination of reward functions is defined as follows: in , λ1 and λ2 are weighting coefficients set by the process management system to balance the dual objectives, serving as reference recovery rate and reference energy consumption (the specific weighting coefficients can be determined according to actual process requirements and are not limited). During the strategy improvement phase, the system follows the current strategy network. Perform importance sampling on the cached data and calculate the probability ratio for each trajectory: like If the value exceeds the confidence interval [1−ϵ,1+ϵ] (typically ϵ=0.2), it is mapped back to this interval through a pruning operation. This pruning mechanism limits the magnitude of a single update, enhancing update stability. The policy loss function is... in The generalized advantage estimate (GAE) is calculated as follows: ,in As a discount factor, The GAE attenuation coefficient. Value function. trained in parallel with the value network for estimating state values and reducing the sampling variance of the policy network. The loss is combined with the value loss and entropy regularization to form the total loss, and the policy network and value network parameters θ, The learning rate can be set between 10-4and 10-3(specifically, it can be determined by the model convergence, which is not limited), and the batch size can be 32 to 128 trajectory samples.

[0057] In the back propagation phase, the system calculates the gradient and updates the network weight in parallel in the GPU cluster. After each update, the new policy network for a new round of simulation in the digital twin and generate the control vector U, the simulation output Rv and Ec are fed back to the main controller. When the preset number of iterations is completed or when the policy convergence curve shows that the average reward change is less than a threshold value (such as 0.01), the optimization process is terminated, and the final policy network parameters and the optimal control vector are output.

[0058] The roasting oxygen partial pressure and vanadium precipitation pH set value in the optimization result are mapped to specific physical control quantities: the oxygen partial pressure instruction is issued to the blower and oxygen valve execution unit through the control system to accurately adjust the composition of the oxygen-rich gas blown into the blast furnace or leaching tank; the vanadium precipitation pH set value instruction is issued to the dosing system to drive the alkaline precipitant or acid regulator pump station to add according to the given ratio and flow rate to achieve the target pH. The instruction issuance follows the OPC UA communication specification, and execution logs and feedback deviations are recorded. When the actual system feedback pH or oxygen partial pressure deviates more than ±0.02 units, the internal controller automatically calls the fine-tuning PID submodule to quickly compensate for the local execution unit, ensuring the overall stable operation of the system.

[0059] The execution process of the entire step 5 is closely linked to steps 3 and 4: the acid concentration gradient provided by step 3 and the precipitation strategy, as well as the spraying amount and fan instruction of step 4, together constitute the state observation vector, which drives the policy network to generate a new U; the output of step 5 is fed back to the execution layer and forms a closed loop through online monitoring. This strategy optimization process is repeated at the control center at a minute level or shorter, combined with digital twin simulation to ensure that each control vector update can be verified in the digital environment, and the verified instruction is applied to actual production, thereby realizing multi-parameter coupled intelligent regulation and control of the entire vanadium ore smelting process.

[0060] Step 6, distribute the multi-dimensional control vector to the edge PLC based on the Storm distributed architecture, perform dynamic dosing of leaching agents and variable frequency speed regulation of tail gas blowers; the working principle of the Storm distributed architecture is: The central control node encapsulates the multi-dimensional control vector as a data stream through a message queue and pushes it to the input component, which receives it and injects it into the processing topology in the form of micro-batches or unit tuples; The routing component maps the data stream based on the PLC identification field; The delivery component converts the partitioned control vector into Modbus TCP messages through the OPC-UA protocol and sends them to the mapping address of the edge controller. If no feedback is received within the specified confirmation time limit, the retry mechanism is triggered to resend the messages with an exponential backoff strategy. The state monitoring component subscribes to the PLC state feedback stream and automatically adjusts the data injection rate of the input component based on backpressure feedback; The index aggregation component collects execution feedback and delay information from each edge node and writes them into a real-time database for the scheduling unit to pull and analyze.

[0061] It is noted that the "Storm distributed architecture" in this embodiment refers to a real-time control delivery platform based on the streaming computing paradigm. It uses micro-batch or unit tuple level data injection and multi-level flow control mechanisms to achieve high throughput, low latency, and high reliability in the distribution of industrial control instructions. "Micro-batch injection" refers to assembling data accumulated in the message queue according to the time window and injecting it into the topology at once to reduce network overhead. "Unit tuple injection" sends each tuple one by one to achieve the lowest latency. "Partition mapping" means routing the control vector to the corresponding sub-topology based on the PLC unique identification field. "Exponential backoff" is a retry mechanism that increases the retry interval by a factor of two to prevent network congestion. "Backpressure feedback" is used to dynamically control the upstream transmission rate based on the downstream processing capacity. "Real-time index database" is a database that stores execution delay and feedback results to allow the scheduling unit to further analyze and optimize.

[0062] In specific implementation, the central control node first obtains the latest multi-dimensional control vector U from the strategy optimization module. This vector contains the roasting oxygen partial pressure setting, vanadium precipitation pH setting, and all sub-vectors such as leaching, precipitation, tail gas purification, and wastewater reuse. The vector is encapsulated into a Storm message structure by the message packaging component, including the message header (containing timestamp, message ID, and destination PLC list), message body (hexadecimal serialized representation of U), and checksum field. The message size is typically 512 bytes to 2KB, with the message header length being about 64 bytes. The specific message size can be determined based on the delivery strategy and network MTU, and is not limited.

[0063] Message queues (e.g. Kafka or RabbitMQ) act as the source of Storm Spouts. Spouts pull messages in micro-batches through a consumer client interface, with batch size and pull interval configurable between 100ms and 1s, or alternatively in unit tuple mode, where each U is injected individually. Pulled messages are converted by the Spout into Storm Tuples, with metadata fields attached, and emitted to the topology's routing Bolts through internal RPC calls.

[0064] Routing components (Fields Grouping Bolts) distribute Streams to corresponding sub-topologies according to the PLC identification field in Tuples, either by hash partitioning or range partitioning. The PLC identification field is a string or integer ID (e.g. 0x01, 0x02), which maps to the IP address and Modbus address of the physical PLC, stored in a central configuration library and updated by field operations through an HMI interface. Partition mapping ensures that all control instructions for the same PLC arrive in the same sub-topology instance in order, avoiding out-of-order execution.

[0065] Instruction Dispatcher Bolts convert control vectors U in message bodies into Modbus TCP packets according to the OPC UA client protocol stack after reading Tuples from their partitioned streams. The packets include function codes 0x10 (write multiple holding registers) or 0x06 (write single holding register), with register addresses specified by a mapping table. The conversion rules are as follows: U components are encoded as floating-point numbers or integers, occupying one or more registers, with encoding methods supporting IEEE 754, 32-bit floating-point, or 16-bit integers. After sending the packet, the Dispatcher starts a timer to monitor ACK confirmation, with a timeout configurable between 100-500ms. If no response is received for the corresponding identifier, the system enters an exponential backoff retry: the initial delay is 250ms, and each failure increases the delay by 1.5, with a maximum of 5 retries. If the retry still fails, the PLC is marked as "offline", and event information is written to the operation log.

[0066] Status Monitor Bolts subscribe to status update Topics reported by OPC UA servers or PLCs, including command execution results, I / O channel states, alarm information, and heartbeat signals. The component calculates downstream processing delays and generates backpressure feedback signals based on comparisons between current delays and configurable maximum delay thresholds (e.g. 200ms). Backpressure feedback notifies upstream Spouts through a flow control API call, instructing them to adjust the sending rate r_send: When Δt exceeds the threshold, r_send is compressed to prevent overloading the PLC; when Δt falls back, the rate is proportionally recovered.

[0067] The Metrics Aggregator Bolt collects the execution feedback and latency statistics from the issuing components and state monitoring components at the end of the topology. It maintains the success rate, average latency and maximum latency data in a sliding window for each PLC, and writes them into a real-time database (e.g. InfluxDB or TimescaleDB). The database time series key is composed of PLC ID, control instruction type and timestamp, and the value is multi-dimensional execution metrics. The scheduling unit can pull these metrics through SQL or RESTful API for fault diagnosis or rescheduling logic decision.

[0068] The whole Storm topology is deployed in cluster mode, containing multiple Supervisor nodes and Nimbus nodes. Nimbus is responsible for topology submission and monitoring, and Supervisor is responsible for running Bolt / Spout processes. Resource allocation refers to Apache Mesos or Kubernetes for containerized scheduling to ensure that CPU / memory resources can be dynamically scaled according to load. The topology parallelism hint is determined by the production management system when submitting, which can be adjusted between 1-10 according to the number of PLCs and control frequency.

[0069] Step 7, real-time monitoring of slag phase composition by online XRD, comparing the twin prediction value of step 5, generating model error and feeding back to step 3 to correct the impurity weight coefficient.

[0070] Note that online XRD refers to a "high-temperature online X-ray diffractometer", which is different from the laboratory offline XRD. It needs to withstand harsh environments such as high-temperature production lines (such as slag roasting temperatures often reaching 600-800°C, the specific temperature threshold is determined by the process, which can be determined by oneself), dust, etc. Non-contact real-time detection is realized by opening a detection window (the window diameter is usually Φ50-100 mm, which can be determined by oneself) in the conveying pipeline. The sampling frequency is generally set to 5-15 min / time (adjustable according to the production line rhythm and model update requirements, which can be determined by oneself). The Rietveld full spectrum fitting algorithm refers to a nonlinear least squares method based on the full spectrum of powder diffraction (the 2θ range is usually taken as 10°-90°, the specific range is determined according to the diffraction characteristics of the phase, which can be determined by oneself). By fitting the residual error between the observed diffraction spectrum and the theoretically calculated spectrum, the phase type, content and crystal structure parameters (such as lattice constant, microstrain) are analyzed. Compared with the traditional "single peak integration method", its quantitative precision for low content phases (such as vanadium iron spinel content <5%) is improved by more than 30% (data is an example, actual verification depends on equipment and process). The error vector ΔP refers to the element-by-element difference between the actual slag phase composition vector Pactual and the digital twin prediction vector Ppred (i.e. ΔP=Pactual-Ppred), which has the same dimension as the number of phases (such as 3D, corresponding to the volume fraction deviation of φFeV2O4, φCaSiO3, φCr2O3), which is used to quantify the direction and amplitude of the model prediction deviation.

[0071] The impurity weight coefficient matrix is the contribution weight set of each impurity (silicon, aluminum, chromium) to the thermodynamic interference factor (entropy, enthalpy, free energy disturbance) in step 3 "dynamic coupling response model". For example, the weight of silicon to silicon colloid wSi, the weight of chromium to redox potential wCr, the matrix dimension is determined by the number of impurities and the dimension of the interference (commonly 33 (impurities) x 3 x 3 (interference dimension), which can be determined by oneself), which needs to be dynamically updated with slag phase feedback to match the actual process.

[0072] In implementation, this step is composed of five modules: diffraction acquisition subsystem, full spectrum quantitative analysis subsystem, error calculation subsystem, weight distribution subsystem and feedback updating subsystem. Its role scenario is to compare the actual crystal phase composition with the digital twin prediction composition during the blast furnace roasting to slag stage, and dynamically correct the impurity weight parameters of the dynamic coupling response model through real-time diffraction measurement of the slag flow or tray surface.

[0073] In the diffraction acquisition subsystem, a high-temperature window type X-ray diffraction probe is installed on the slag discharge pipeline or the tray conveying belt, the probe window temperature is maintained between 350 and 600°C through a water cooling or air cooling system, the detector array covers a 2θ angle range of 10 to 90 degrees, and the powder diffraction spectrum is continuously acquired with a step of 0.02 degrees. The geometric parameters between the probe and the detector, the incident light intensity, and the exposure time can be adjusted in real time according to the slag emission intensity to ensure that the signal-to-noise ratio always meets the quantitative analysis requirements. After the detection is completed, the original angle intensity data stream is packaged and transmitted to the analysis server through real-time Ethernet.

[0074] After receiving the diffraction data, the full-spectrum quantitative analysis subsystem first performs background subtraction, removes the wide base and scattering background by polynomial curve fitting, and then applies the Rietveld full-spectrum fitting algorithm to cooperatively fit the remaining diffraction peaks. The peak shape is selected as a mixed pseudo-Voigt function to accommodate the peak broadening and offset effects caused by high temperature, and the fitting process is iterated by nonlinear least squares until the residual error converges. The software loads crystal structure models and relative atomic coordinates of three phases including spinel phase (FeV2O4), wollastonite phase (CaSiO3), and chromite phase (Cr2O3), and directly outputs the volume fraction of each phase after fitting. Each analysis is accompanied by a fitting residual index and R factor to evaluate the quantitative accuracy, and the threshold can be set by the system administrator, usually not more than 5%.

[0075] The error calculation subsystem synchronously aligns and differentiates the actual crystal phase composition vector obtained by analysis and the phase composition vector predicted by the digital twin in step 5 to obtain a phase composition deviation vector. The alignment is based on the shared time stamp and sample number, and the system ensures that the actual and predicted data at the same event point are synchronized. If the time difference exceeds the preset tolerance (for example, 200 ms), the alignment is marked as abnormal and skipped this time.

[0076] The weight allocation subsystem allocates weights to the deviation vector based on the three-phase covariance matrix calculated in advance from historical samples. The covariance matrix reflects the statistical correlation between the crystal phases under normal working conditions, for example, an increase in FeV2O4 often accompanies a decrease in CaSiO3. This subsystem generates an error weight vector by matrix operation of the deviation vector and the covariance matrix, and each component is the adjustment value of the corresponding impurity weight of the three phases, which is normalized and passed to the next module after normalization. The update of the covariance matrix can be re-calculated based on monthly or batch production samples to adapt to changes in raw ore batches, and the related period is determined by process engineers according to the stability of the production line.

[0077] The feedback updating subsystem converts the error weight vector into the incremental update of the impurity weight parameters in the dynamic coupled response model according to a predefined mapping relationship after receiving the error weight vector. The mapping relationship is defined by a set of weight mapping coefficient matrices, which are usually obtained through sensitivity analysis during the model development phase and describe the influence intensity of each crystal phase deviation on different impurity weights. The updating process adopts a gradient descent strategy to control the updating amplitude with a learning rate parameter and records the weight change trajectory and convergence status after each iteration. After the weight is updated once, the system writes the updated weight into the response model and triggers the model reconstruction, thereby affecting the calculation of the thermodynamic factors in step 3 and the generation of subsequent control instructions.

[0078] The entire online XRD feedback process is automatically driven by the event bus of the central control system. When any link appears timeout or fails to resolve (such as excessive fitting residual, data alignment anomaly or covariance calculation error), the system will mark the current feedback period as abnormal, and according to the redundancy configuration, it can choose to use the weight of the previous period or enter the manual intervention mode, while recording the abnormal details in the log for subsequent analysis.

[0079] Please refer to Figure 5 A multi-parameter coupled intelligent control system for vanadium ore smelting process, comprising: a composition acquisition module for receiving LIBS / XRF probe data by a host computer, extracting impurity concentration matrix and valence state distribution through wavelet packet decomposition and principal component regression, and outputting original spectral data; A Kalman filter module for fusing and compensating the rheological disturbance of the slurry based on the original spectral data and the dielectric constant measured by microwave dielectric spectroscopy through an extended Kalman filter algorithm, and outputting a standardized impurity spectral vector; A coupled response module for calculating the thermal interference factor online through improved Maxwell-Garnett and non-steady-state energy conservation equation, solving the acid concentration gradient, reaction time and dosing strategy; A substance coordination module for parallelly collecting mother liquor Concentration and tail gas HCl flow rate, constructing a linear programming model with mass conservation constraints and solving it through Lagrange duality and interior point method Molar ratio, spray amount and fan adjustment instruction; A strategy optimization module for mapping the coupled response and the output parameters of the substance coordination module into a multi-dimensional control vector U, iteratively updating it in the digital twin through PPO proximal policy optimization algorithm, and outputting the oxygen partial pressure and vanadium precipitation pH setting; A distributed issuance module for receiving the control vector based on the Storm distributed architecture, issuing U to the edge PLC through routing, OPC-UA to Modbus TCP and back pressure mechanism; An XRD feedback module is used to obtain XRD diffraction data of the slag online, which is quantitatively analyzed by a Rietveld algorithm and compared with the predicted value of the twin, to generate a residual and feed back the impurity weight update information to the coupling response module through the gradient correction algorithm.

[0080] In one implementation, the system automatically associates the vanadium-titanium magnetite batch information of the day, loads the corresponding ore composition calibration parameters and historical working condition model. After the ore is sent to the drying and preheating section by the bucket elevator, it flows into the LIBS / XRF combined probe scanning area through the conveying belt. The composition acquisition module first triggers the laser and X-ray excitation in parallel at a frequency of 20 Hz. The probe collects spectra in real time and maps the signals into a five-item impurity concentration matrix of SiO2, Al2O3, Cr2O3, Fe2O3 and V2O5 and vanadium valence ratio through wavelet packet decomposition and principal component regression model. The results are pushed to the Kalman filter module within 500 ms.

[0081] The Kalman filter module fuses the original spectral data with the pulp dielectric constant obtained from the online microwave dielectric spectrometer. In the first 1s control cycle, EKF predicts the prior state according to the previous cycle control disturbance input, then corrects the drift term using the online dielectric constant, generates a standardized impurity spectrum vector, and writes it to the coupled response module.

[0082] The coupled response module receives the temperature field, air volume and the standardized vector, and based on the improved Maxwell-Garnett equation and real-time thermal measurement data of the furnace condition, it calculates the thermal interference factor online, and then calculates the local acid concentration gradient and mass transfer resistance of each tank area in the leaching section in the partition finite element model, to generate the preliminary acid concentration distribution and reaction time suggestion.

[0083] The substance coordination module is started in parallel in the same control cycle: the ion chromatograph extracts vanadium precipitation mother liquor samples every minute and outputs the NH4⁺ concentration, while the TDLAS instrument measures the tail gas HCl concentration at a rate of 10 Hz, and the turbine flow meter provides Nm 3 / h level of volume flow rate. The system constructs a mass conservation constraint linear programming with the three, and solves the NH4⁺-HCl optimal molar ratio and corresponding spraying amount and fan speed instruction within 2s through the Lagrange dual and interior point method.

[0084] The strategy optimization module then integrates the acid concentration gradient given by the coupled response and the spraying amount, wind speed and other parameters generated by the substance coordination into an 18-dimensional control vector U, and uses the PPO algorithm in the digital twin to perform three generations of iterative optimization. The simulation environment completely reproduces the dynamic of the blast furnace top temperature, the leaching tank stirring rate and the precipitation tank pH, evaluates the vanadium recovery rate and energy consumption ratio for each control strategy, and outputs the final roasting zone oxygen partial pressure set value and precipitation section pH target after the tenth generation of optimization.

[0085] The distributed delivery module utilizes the Storm framework to fluidize the multi-dimensional vector, injects it into the routing topology with 200 ms micro-batches through Kafka Spout, and directs the partitioning according to the PLC ID with Fields Grouping. The delivery component encapsulates the control data into OPC UA to Modbus TCP messages, pushes them to the edge PLC, and completes the parameter loading of the diaphragm pump frequency and the fan frequency converter within 300 ms. The state monitoring component continuously subscribes to the PLC execution feedback and adjusts the transmission rate in the reverse direction according to the delay, ensuring stable delivery pressure and execution.

[0086] After the high-temperature XRD probe installed at the tapping section collects the first batch of slag diffraction patterns in real time, the XRD feedback module immediately performs Rietveld full spectrum fitting to obtain the difference vector between the actual crystal phase composition and the digital twin prediction value. The error calculation subsystem performs weighted distribution on the difference through the pre-set phase composition covariance matrix, and the generated weight increment is then applied to the impurity weight coefficient in the coupled response module by the gradient update algorithm, completing the first online correction.

[0087] The overall process runs at a cycle of 10 s, and the system completes the first closed loop within 100 s. Subsequently, all modules continue to iterate at this pace, allowing the control strategy and model parameters to continuously adapt to the field conditions. During the entire production, the system successfully increases the vanadium recovery rate by about 3%, reduces energy consumption by about 2%, stabilizes the HCl emission concentration of tail gas below 50 ppm, and increases the reuse rate of mother liquor NH4⁺ by 15%. The above data is only an example, and the specific effect can be determined according to the actual batch of ore and the size of the equipment.

[0088] Although the specific embodiments of the present application are described above, those skilled in the art should understand that these specific embodiments are only illustrative, and those skilled in the art can make various omissions, substitutions and changes to the details of the above method and system without departing from the principles and essence of the present application. For example, combining the above method steps, performing substantially the same function in substantially the same manner to achieve substantially the same result is within the scope of the present application. Therefore, the scope of the present application is only limited by the appended claims.

Claims

1. A multi-parameter coupling intelligent regulation method for a vanadium ore smelting process, characterized in that: The application comprises the following steps: Step 1, probe real-time scanning of ore stream by laser-induced breakdown spectroscopy combined with X-ray fluorescence, to obtain impurity element concentration matrix and valence state distribution data; Step 2, based on the impurity element concentration matrix and the dielectric constant of the ore slurry measured by microwave dielectric spectroscopy, the signal drift caused by the disturbance of the rheological properties of the ore slurry is dynamically corrected by an extended Kalman filter fusion algorithm, and a normalized impurity spectrum vector is output; Step 3, inputting the normalized impurity spectrum vector into a dynamic coupling response model to calculate the impurity thermodynamic interference factor, and generating acid concentration gradient, reaction time optimization instruction and precipitant dosing strategy; Step 4, synchronously collecting the ammonia ion concentration of the vanadium precipitation mother liquor and the hydrogen chloride flow rate of the roasting tail gas, and solving the optimal solution of the ammonia-chlorine neutralization molar ratio by a material flow coordination method, to obtain the mother liquor spraying amount and tail gas fan coordinated regulation instruction; Step 5, coupling the output parameters of steps 3 and 4 to construct a multi-dimensional control vector U, and updating U in the digital twin through a proximal policy optimization algorithm to output the roasting oxygen partial pressure and the vanadium precipitation pH set value; Step 6, distributing the multi-dimensional control vector to the edge PLC based on the Storm distributed architecture, and performing leaching agent dynamic dosing and tail gas fan variable frequency speed regulation operation; Step 7, real-time monitoring of the slag phase composition by online XRD, comparing the predicted value of the twin in step 5, generating a model error and feeding back to step 3 to correct the impurity weight coefficient.

2. The method according to claim 1, wherein the method is characterized in that: The working method of step 1 is as follows: first, the output beams of high-energy pulsed laser and multi-target X-ray tube are coaxially aligned by using an optical beam splitter, and the signal acquisition time window of high-energy pulsed laser and multi-target X-ray tube is aligned by a time sequence control circuit to obtain the original pulse waveform; then, the probe scanning step and residence time are adjusted in real time based on the ore stream speed feedback signal obtained by an optical encoder, and the X-ray tube voltage is automatically switched to adapt to the excitation depth requirement of different particle size ores; finally, the atomic ion emission spectrum generated by laser-induced breakdown spectroscopy and the characteristic X-ray energy spectrum of micro-focal spot X-ray fluorescence are input into a multivariate spectral deconvolution module based on non-negative matrix decomposition, and the impurity concentration matrix and vanadium valence state distribution data containing spatial coordinates are obtained by decoupling overlapping peaks and correcting matrix effects.

3. The method according to claim 2, wherein the method is characterized in that: The multivariate spectral deconvolution module based on non-negative matrix decomposition decomposes the original spectral matrix into a basis vector matrix and an abundance coefficient matrix by an alternating least squares iterative optimization method, wherein the basis vector matrix contains a standard spectral line library of known elements; in each iteration of the alternating least squares iterative optimization method, a matrix effect correction mechanism is adopted, and the weight coefficient of the V spectrum is dynamically adjusted according to the quenching factor of the Fe element dominant spectrum area; if Cr-V spectral peak overlap is detected, an overlapping peak decoupling protocol is triggered, and the Gaussian mixture model is used to separate the overlapping peak area, to obtain the pure element concentration matrix and vanadium valence state distribution data after matrix correction and peak decoupling.

4. The method according to claim 1, wherein the method is characterized in that: The working principle of the extended Kalman filter fusion algorithm is as follows: s100, the first Impurity concentration matrix derived from laser-induced breakdown spectrum at any time Dielectric constant measured by microwave dielectric spectroscopy Organization as a measurement vector Online fusion is achieved using an EKF iterative structure; s200, in a prediction phase, a previous impurity concentration combined to generate a current concentration and drift state combined to generate a current concentration and drift state and update the prediction covariance with the Jacobian matrix and process noise covariance; s300, mapping the predicted concentration into a dielectric constant prediction value based on a measurement mapping function MG dielectric mixing equation, expressed as: （1） In equation (1), is the dielectric constant of the slag phase matrix, is the dielectric constant of the impurity particles, is the volume fraction of the impurities, used to map the predicted concentration to a dielectric constant prediction, thus constructing the Jacobian matrix; s400, obtaining process noise covariance based on energy component covariance of wavelet packet decomposition of the LIBS signal, adjusting measurement noise covariance based on variance of permittivity sequence; s500, introducing a forgetting factor in the Kalman gain to strengthen the response to new data, and when the measurement residual exceeds a preset threshold, dynamically recalibrating the covariance matrix through a recalibration mechanism; s600, obtaining a normalized impurity pedigree vector through a Kalman update equation.

5. The vanadium ore smelting process multi-parameter coupling intelligent regulation method according to claim 1, characterized in that: The dynamic coupling response model comprises an input preprocessing layer, a feature fusion layer, a thermodynamic factor calculation layer, a dynamic coupling analysis layer, a multi-objective optimization layer, and an instruction synthesis layer; the input preprocessing layer pre-processes the normalized impurity pedigree vector through normalization transformation and wavelet packet denoising; the feature fusion layer integrates impurity concentration and permittivity time series in parallel through a multi-head attention mechanism, and triggers an adaptive recalibration mechanism to compensate for time domain coupling information when the attention weight is lower than a preset threshold; the thermodynamic factor calculation layer generates a thermodynamic interference factor matrix for each impurity based on a nonlinear mapping function of Gibbs free energy calculation, for reflecting the contribution of different species to the entropy and enthalpy disturbance of the reaction system; The dynamic coupling analysis layer couples the thermodynamic interference factor matrix with the reactor geometry model using the finite element method to solve the local acid concentration gradient and mass transfer resistance; the multi-objective optimization layer performs Pareto optimal search among the acid concentration gradient, reaction time, and precipitant dosage through the NSGA-II algorithm, and updates the crossover and mutation operators when the population convergence degree is lower than a preset threshold; the instruction synthesis layer translates the optimization results into acid concentration adjustment rate, reaction residence time, and precipitant dosage control sequence through weighted linear combination.

6. The method of claim 1, wherein the method is characterized by: The material flow coordination method online monitors the ammonia ion concentration of the vanadium precipitation mother liquor by an ion chromatograph, simultaneously measures the hydrogen chloride concentration of the roasting tail gas in real time by a tunable diode laser absorption spectrum, and collects the exhaust gas volume flow rate by a turbine flowmeter, constructs a gas-liquid two-phase flow field model, and calculates the spray droplet trajectory and exhaust gas diffusion path based on the Reynolds time-averaged N-S equation; then, the optimal solution of the ammonia-chlorine molar ratio is solved by a sequential quadratic programming algorithm, and the objective function is set to 1.05±0.02; if the ammonia ion concentration fluctuation exceeds ±5%, the diaphragm pump frequency is adjusted in real time by an incremental PID control protocol; when the exhaust gas flow rate changes rate exceeds a preset threshold, the material flow coordination method pre-adjusts the fan speed through a feedforward compensation mechanism.

7. The method according to claim 1, characterized in that: The step 5 constructs a multi-dimensional control vector state space based on the output parameters of the step 3 and the step 4, and iteratively updates the multi-dimensional control vector state space through a proximal policy optimization algorithm; the multi-dimensional control vector state space comprises a furnace condition control vector, a leaching parameter vector, a precipitation adjustment vector, a tail gas purification vector, a wastewater reuse vector, and a model correction vector; the proximal policy optimization algorithm simulates the roasting-vanadium precipitation coupling process based on a digital twin, and sets dual objective functions of maximizing vanadium recovery rate and minimizing energy consumption; The importance sampling mechanism is used to collect policy gradient data, the probability ratio of new and old policies is calculated, and the policy update step is restricted by the gradient clipping mechanism when the probability ratio exceeds the preset confidence interval threshold; then, the proximal policy optimization algorithm estimates the temporal difference error through the generalized advantage function, and updates the parameters of the policy network in the backward propagation; the Adam optimizer is used to iterate the weights of the policy network, and the roasting oxygen partial pressure set value and the vanadium precipitation pH set value that meet the Pareto optimality are output.

8. The method of claim 1, wherein the method is characterized by: The working principle of the Storm distributed architecture is as follows: The central control node encapsulates the multi-dimensional control vector as a data stream through the message queue and pushes it to the input component, which receives it and injects it into the processing topology in the form of micro-batches or unit tuples; The routing component maps the data stream based on the PLC identification field; The delivery component converts the partitioned control vector into a Modbus TCP message through the OPC-UA protocol and sends it to the mapping address of the edge controller, and if no feedback is received within the set confirmation time limit, the retry mechanism is triggered to resend with an exponential backoff strategy; The state monitoring component subscribes to the PLC state feedback stream and automatically adjusts the data injection rate of the input component based on back pressure feedback; The index aggregation component collects the execution feedback and delay information of each edge node and writes it into a real-time database for the scheduling unit to pull and analyze.

9. The method according to claim 1, wherein the method is characterized in that: The working principle of the step 7 is as follows: The slag phase diffraction pattern is collected in real time by a high temperature online X-ray diffractometer, and the crystal phase composition vector is obtained by Rietveld full spectrum fitting algorithm analysis wherein is a volume fraction of vanadium iron spinel phase, is a volume fraction of wollastonite phase, is a volume fraction of chromite phase; Synchronously call the digital twin prediction phase composition vector generated in step 5 ; Performed through the error vector calculation module ;like If the L2 norm exceeds the preset tolerance threshold, an error weighting mechanism based on Mahalanobis distance is triggered, as shown in the formula: （2） In formula (2), ∑ is the phase composition covariance matrix, is the inverse covariance matrix, which is used to eliminate the interference of inter-phase correlation; then, The input feedback correction channel updates the impurity weight coefficient matrix in the dynamic coupling response model in step 3 by the gradient descent method, and finally generates the corrected impurity thermodynamic interference factor calculation parameters.

10. A multi-parameter coupling intelligent regulation system for a vanadium ore smelting process, characterized in that: The vanadium ore smelting process multi-parameter coupling intelligent control method applied to any one of claims 1-9 comprises: The composition acquisition module is used to receive LIBS / XRF probe data on the host computer, extract impurity concentration matrix and valence state distribution through wavelet packet decomposition and principal component regression, and output original spectral data; The Kalman filter module is used to fuse and compensate the rheological disturbance of the slurry based on the original spectral data and the dielectric constant measured by microwave dielectric spectroscopy through the extended Kalman filter algorithm, and output the standardized impurity spectral vector; The coupling response module is used to calculate the thermal interference factor online through the improved Maxwell-Garnett and non-steady-state energy conservation equation, and solve the acid concentration gradient, reaction time and dosing strategy; A substance coordination module is used to collect mother liquor in parallel Concentration vs. exhaust HCl flow rate, constructing a linear programming model with mass conservation constraints and solving it by Lagrangian duality and interior point method Mole ratio, generating spray amount and fan adjustment instruction The strategy optimization module is used to map the coupling response and the output parameters of the substance coordination module into a multi-dimensional control vector U, and apply the PPO proximal policy optimization algorithm in the digital twin to iteratively update and output the oxygen partial pressure and vanadium precipitation pH settings; The distributed delivery module is used to receive the control vector based on the Storm distributed architecture, and deliver U to the edge PLC through routing, OPC-UA to Modbus TCP and back pressure mechanism; The XRD feedback module is used to obtain XRD diffraction data of the slag online, quantitatively analyze it through the Rietveld algorithm, and compare it with the predicted value of the twin to generate a residual error and feed back the impurity weight update information to the coupling response module through the gradient correction algorithm.

Images