A method for judging chelation solidification and stabilization of waste incineration fly ash

Abstract

The application discloses a waste incineration fly ash chelation solidification stabilization determination method, and relates to the technical field of big data analysis.S1, the terahertz absorption spectrum of the fly ash to be determined is obtained, the spectral feature vector is extracted according to 5 occurrence form characteristic intervals, and the combined model of the partial least squares discriminant analysis submodule and the partial least squares regression submodule is inputted in series inversion, and the occurrence form distribution vector is obtained through zero lower limit truncation and summation normalization;S2, the Bayesian network containing 5 occurrence form proportion nodes, 5 process parameter nodes, migration activity index nodes, reaction environment intensity index nodes and normalized leaching risk index nodes is constructed, the distribution vector and the process parameters are taken as evidence to infer, and the sum of the posterior probabilities of the critical, over-limit and serious over-limit states is taken as the leaching risk probability;S3, according to the comparison result of the leaching risk probability and the two thresholds, the chelation solidification product is executed to release, retest or supplement.The application improves the automation and accuracy of the stabilization determination.

Description

Technical Field

[0001] This invention belongs to the field of big data analysis technology and relates to an intelligent sensing system for making judgments through big data analysis, specifically a method for determining the chelation, solidification and stabilization of fly ash from waste incineration. Background Technology

[0002] Fly ash from waste incineration is classified as hazardous waste, containing heavy metals such as lead, cadmium, chromium, and zinc, which pose environmental risks due to leaching and migration during landfill disposal. Industrially, chelating agents are typically used to chelate and solidify fly ash, stabilizing it by converting heavy metals into stable forms with low solubility before landfilling. In continuously operating fly ash disposal lines, the stabilization effect of each batch of chelated solidification products needs to be assessed to determine its safe landfill suitability. Currently, the assessment of stabilization effectiveness mainly relies on laboratory leaching toxicity tests, which involve preparing leachate according to relevant standards, measuring the leaching concentration of heavy metals, and comparing it with limits. This method is time-consuming, cumbersome, and difficult to match with the continuous processing cycle of production lines. Results are often only available after the product has been transported or landfilled, making real-time feedback and closed-loop control of the disposal process impossible. Furthermore, the leaching behavior of heavy metals in fly ash depends not only on their total amount but also on their chemical form and the process conditions of the chelation reaction; complex nonlinear coupling relationships exist among these influencing factors. Existing methods often estimate stabilization effects using a single indicator or simple regression. These methods struggle to characterize the coupling effects between multiple factors and fail to provide robust results when some process parameters are missing or measurement biases exist. Inaccurate assessments can easily lead to excessive heavy metal leaching or overdosing of chelating agents. Therefore, it is necessary to provide a method that can accurately and in real-time determine the chelation and solidification stabilization effect of waste incineration fly ash. Summary of the Invention

[0003] The main objective of this invention is to provide a method for determining the chelation, solidification, and stabilization of fly ash from waste incineration, in order to solve the problems of existing technologies that rely on laboratory leaching toxicity tests, have long testing cycles, are difficult to provide real-time feedback, and are difficult to characterize the coupling effect of multiple factors such as the occurrence form of heavy metals and the conditions of the chelation reaction process, resulting in insufficient robustness in the determination.

[0004] To address the aforementioned technical problems, this invention provides a method for determining the chelation, solidification, and stabilization of fly ash from waste incineration, comprising the following steps: S1: Obtain the terahertz absorption spectrum of the fly ash sample of the batch to be judged, extract spectral feature quantities according to the 5 occurrence morphology feature intervals to obtain spectral feature vector; input the spectral feature vector into the partial least squares discriminant-regression joint model composed of the partial least squares discriminant analysis submodule and the partial least squares regression submodule in sequence for inversion, and then perform zero lower limit truncation and summation normalization to obtain the occurrence morphology distribution vector containing the proportion of the 5 occurrence morphologies. S2: Construct a Bayesian network consisting of 13 discrete nodes: 5 nodes representing the proportion of occurrence morphology, 5 nodes representing process parameters, a migration activity index node, a reaction environment intensity index node, and a normalized leaching risk index node. The normalized leaching risk index node is divided into 5 states: low-risk, qualified, critical, exceeding limits, and severely exceeding limits. The occurrence morphology distribution vector and the collected process parameters are used together as evidence observations of the Bayesian network. Posterior probability inference is performed on the normalized leaching risk index node, and the sum of the posterior probabilities corresponding to the critical, exceeding, and severely exceeding limits states is taken as the leaching risk probability of the batch to be judged. S3: When the stabilization pass probability threshold is less than the stabilization warning probability threshold; when the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released; when the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, the leaching toxicity retest is initiated; when the leaching risk probability is greater than or equal to the stabilization warning probability threshold, the chelating agent is added, and after the addition is completed, S2 is executed again until the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released.

[0005] Furthermore, the step of obtaining the terahertz absorption spectrum of the fly ash sample of the batch to be judged in S1 includes: drying the fly ash sample of the batch to be judged in a constant temperature oven at 105 degrees Celsius for 3 hours, and then ball milling and sieving it to a particle size of less than 75 micrometers; weighing 0.2 grams of the ground fly ash powder and pouring it into a polyethylene circular pressing mold, and holding it under a pressure of 10 MPa for 2 minutes to obtain a circular sheet sample with a diameter of 13 mm and a thickness of 1 mm; and performing terahertz time-domain spectroscopy on the circular sheet sample to obtain the terahertz absorption spectrum.

[0006] Furthermore, the steps for acquiring terahertz time-domain spectrometry include: transmitting terahertz pulses with a frequency range of 0.1 terahertz to 3 terahertz through a circular sheet sample using a terahertz time-domain spectrometer, and sampling the terahertz pulses after they have passed through the circular sheet sample in the time domain to obtain the sample terahertz time-domain signal; acquiring the blank reference terahertz time-domain signal of the air background using the same procedure; performing Fourier transform on the sample terahertz time-domain signal and the blank reference terahertz time-domain signal to obtain the sample terahertz frequency domain signal and the reference terahertz frequency domain signal, respectively; and taking the negative logarithm of the ratio of the frequency domain amplitudes of the sample terahertz frequency domain signal and the reference terahertz frequency domain signal to obtain the terahertz absorption spectrum.

[0007] Furthermore, the five occurrence forms are exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state; the five occurrence form characteristic intervals are the exchangeable state characteristic interval, carbonate-bound state characteristic interval, iron-manganese oxide-bound state characteristic interval, organic matter and sulfide-bound state characteristic interval, and residue state characteristic interval corresponding to the five occurrence forms; in S1, the maximum absorption intensity within each occurrence form characteristic interval is taken as the spectral characteristic quantity of the corresponding occurrence form characteristic interval, and the five spectral characteristic quantities are spliced ​​together in the listed order of the five occurrence form characteristic intervals to obtain the spectral characteristic vector; the occurrence form distribution vector is composed of the proportion of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state in the listed order.

[0008] Furthermore, the partial least squares discriminant analysis submodule uses the spectral feature vector as input to identify the dominant morphological category of the batch to be judged; the partial least squares regression submodule uses the spectral feature vector and the dominant morphological category as input to regress and output the preliminary predicted proportion values ​​of the five morphological categories of the batch to be judged; the zero lower bound truncation process replaces the values ​​less than zero in the preliminary proportion prediction values ​​with zero; the summation and normalization process divides the five values ​​after the zero lower bound truncation process by the sum of the five values; after the zero lower bound truncation and summation and normalization processes, the morphological distribution vector is obtained; the calibration process of the partial least squares discriminant-regression joint model includes: selecting 200 fly ash samples with known morphological distributions to form a calibration fly ash sample set; and performing calibration on each calibration fly ash sample in the calibration fly ash sample set. Tessier continuous extraction experiments were performed on the ash samples to obtain the measured proportions of five occurrence forms, and terahertz absorption spectra were collected. The Tessier extracts corresponding to the five occurrence forms were dried, ground, and compressed into tablets, and then terahertz absorption spectra were collected. The frequency range corresponding to the full width at half maximum (FWHM) of the absorption peaks in the collected terahertz absorption spectra was used as the initial frequency band boundary of the characteristic interval of the corresponding occurrence form. The terahertz absorption spectra and measured proportions of all calibrated fly ash samples were used as training data. The discrimination matrix of the partial least squares discriminant analysis submodule and the regression matrix of the partial least squares regression submodule were obtained by iterative solution using partial least squares discriminant analysis and partial least squares regression algorithms. The initial frequency band boundary was iteratively optimized to obtain the final frequency band boundary of the characteristic interval of the five occurrence forms.

[0009] Furthermore, the five occurrence state percentage nodes are the exchangeable state percentage node, the carbonate-bound state percentage node, the iron-manganese oxide-bound state percentage node, the organic matter and sulfide-bound state percentage node, and the residue state percentage node; the five process parameter nodes are the reaction slurry pH node, the reaction slurry redox potential node, the reaction slurry chloride ion concentration node, the chelating agent dosage node, and the stirring time node; among the 13 discrete nodes, excluding the normalized leaching risk index node, the remaining 12 nodes are arranged at equal intervals according to their respective value ranges. The method is divided into 5 states; the normalized leaching risk index node is divided into four segments: risk qualified segment threshold, risk warning segment threshold, value 1, and risk severe exceedance segment threshold. The four segment values ​​increase sequentially. Value 1 indicates that the normalized leaching quotient of at least one target heavy metal reaches the leaching toxicity limit corresponding to GB16889. The normalized leaching risk index node is divided into five states in sequence: low risk state, qualified state, critical state, exceedance state, and severe exceedance state.

[0010] Furthermore, the normalized leaching risk index is defined as follows: for each batch of chelated solidification product, leachate is obtained according to the HJ / T300 acetic acid buffer solution leaching toxicity test method. The leaching concentration of each target heavy metal in the leachate is divided by the leaching toxicity limit corresponding to GB16889 to obtain the normalized leaching quotient of each target heavy metal limit. The maximum value among all the normalized leaching quotients is taken as the normalized leaching risk index of the corresponding batch. The target heavy metal is at least one of the inorganic pollutants with leaching toxicity limits specified in GB16889. The directed connection relationship between the nodes in the Bayesian network is as follows: the five occurrence form proportion nodes respectively lead to directed edges pointing to the migration activity index node; the reaction slurry pH node, the reaction slurry redox potential node, and the reaction slurry chloride ion concentration node respectively lead to directed edges pointing to the reaction environment intensity index node; the migration activity index node, the reaction environment intensity index node, the chelating agent dosage node, and the stirring time node respectively lead to directed edges pointing to the normalized leaching risk index node.

[0011] Furthermore, both the migration activity index node and the reaction environment intensity index node are deterministic nodes; the conditional probability table of the migration activity index node is assigned by the migration activity mapping table; the conditional probability table of the reaction environment intensity index node is assigned by the reaction environment intensity mapping table; the migration activity mapping table is pre-prepared according to the chemical principle that the migration activity levels of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state decrease sequentially as defined by Tessier continuous extraction; the reaction environment intensity mapping table is pre-prepared according to the chelation reaction chemical principle that the reaction environment intensity level increases when the pH of the reaction slurry increases, the redox potential of the reaction slurry decreases, and the chloride ion concentration of the reaction slurry decreases; the conditional probability tables of the remaining 11 nodes, excluding the migration activity index node and the reaction environment intensity index node, are assigned by calibration experiments; the calibration experiment procedure includes: performing the following on each calibration fly ash sample in the calibration fly ash sample set described in claim 5. Two calibration operations: The first operation involves inverting the morphological distribution vector according to the steps described in claim 5; the second operation involves performing a chelation reaction according to a preset chelating agent dosage and a preset stirring time to obtain a chelated solidified product, recording the pH, redox potential, chloride ion concentration, chelating agent dosage, and stirring time of the chelated solidified product's reaction slurry, and determining the normalized leaching risk index of the chelated solidified product according to the HJ / T300 acetic acid buffer solution leaching toxicity test method; the calibration observations obtained from the two calibration operations are discretized into state values ​​of the corresponding nodes according to the state partitioning method of the corresponding nodes in the Bayesian network, and the conditional probability table of the 11 nodes is assigned values ​​one by one by the maximum a posteriori estimation algorithm with Dirichlet prior smoothing; the morphological distribution vector obtained by inversion according to the steps described in claim 5 is used as input evidence in both the calibration stage and the online judgment stage of the Bayesian network; the posterior probability inference in S2 is completed by the joint tree inference algorithm.

[0012] Furthermore, the process parameter acquisition steps in S2 include: measuring the pH, redox potential, and chloride ion concentration of the reaction slurry of the batch to be judged in situ at the outlet of the chelation reactor corresponding to the batch to be judged using a pH electrode, a redox potential electrode, and a chloride ion selective electrode, respectively; recording the chelating agent dosage of the batch to be judged by a chelating agent metering pump; and recording the stirring time of the batch to be judged by a mixer control system. The evidence observation values ​​of the five process parameter nodes are the pH, redox potential, chloride ion concentration, chelating agent dosage, and stirring time of the batch to be judged.

[0013] Furthermore, in S3: when the leaching risk probability is less than or equal to the stabilization pass probability threshold, the control system outputs a stabilization pass indicator, and the chelated solidified product corresponding to the batch to be judged is released into the landfill conveyor line through the discharge valve of the chelation reactor; when the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, the control system outputs a verification pending indicator and initiates the re-inspection process; the re-inspection process is as follows: an automatic sampling robot takes a 50-gram sample of the chelated solidified product at the discharge port of the chelation reactor, and performs leaching toxicity testing on the sampled chelated solidified product according to the HJ / T300 acetic acid buffer solution leaching toxicity test method. Laboratory leaching toxicity retesting is conducted, and the actual leaching toxicity test results are used as the final judgment basis for the batch to be judged. When the leaching risk probability is greater than or equal to the stabilization warning probability threshold, the control system outputs a leaching risk indicator and initiates the replenishment process. The replenishment process is as follows: close the discharge valve of the chelation reactor, add a preset proportion of chelating agent to the chelation reactor through the replenishment channel solenoid valve, and continue stirring for a preset replenishment stirring time. After the replenishment stirring is completed, S2 is re-executed using the morphological distribution vector of the batch to be judged until the leaching risk probability is less than or equal to the stabilization qualified probability threshold, at which point the control system outputs a stabilization qualified indicator.

[0014] The present invention provides a method for determining the chelation, solidification, and stabilization of fly ash from waste incineration, which has the following beneficial effects: By utilizing terahertz absorption spectroscopy and a partial least squares discriminant-regression joint model, the distribution of heavy metal speciation in fly ash can be rapidly inverted without lengthy chemical extraction. Zero lower limit truncation and summation normalization ensure the inversion results conform to the non-negativity and normalization of proportions, providing objective and repeatable input for subsequent judgments. A Bayesian network is employed to incorporate the speciation distribution along with reaction slurry process parameters and operational parameters into causal inference, characterizing the nonlinear coupling relationships between various factors. Even when some process parameters are missing or measurement biases exist, robust leaching risk probabilities can still be provided, overcoming the limitations of relying on single indicators or... The shortcomings of simple regression in determining robustness are addressed. By introducing two deterministic intermediate nodes—the migration activity index node and the reaction environment intensity index node—the size of the parent state combination of the normalized leaching risk index node conditional probability table is significantly reduced, enabling reliable estimation of the conditional probability table with limited calibration samples. Based on the comparison between the leaching risk probability and the probability threshold, the chelated solidification products are graded and released, retested, and chelating agent is replenished. A replenishment cycle is used to conduct closed-loop adjustment for non-compliant batches, thereby improving the automation, accuracy, and economic efficiency of fly ash disposal while ensuring landfill compliance. Attached Figure Description

[0015] Figure 1 A schematic diagram of the terahertz time-domain signals of the sample and blank reference provided in an embodiment of the present invention; Figure 2A schematic diagram of the terahertz absorption spectrum and five occurrence morphology characteristic regions provided in the embodiments of the present invention; Figure 3 A schematic diagram illustrating the distribution of five dominant endowment morphology categories in the latent variable space of partial least squares discriminant analysis provided in this embodiment of the invention; Figure 4 A calibration comparison diagram of the ratio of partial least squares discriminant-regression joint model inversion and the measured ratio of Tessier continuous extraction provided in this embodiment of the invention; Figure 5 This is a schematic diagram of the posterior probability distribution and leaching risk probability aggregation of the five states of the normalized leaching risk index node provided in an embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the dynamic trajectory of the leaching risk probability during the replenishment cycle of the chelating agent provided in this embodiment of the invention, showing how the probability changes with each replenishment cycle. Detailed Implementation

[0016] The method of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0017] A method for determining the chelation, solidification, and stabilization of fly ash from waste incineration includes the following steps: S1: Obtain the terahertz absorption spectrum of the fly ash sample of the batch to be judged, extract spectral feature quantities according to the 5 occurrence morphology feature intervals to obtain spectral feature vector; input the spectral feature vector into the partial least squares discriminant-regression joint model composed of the partial least squares discriminant analysis submodule and the partial least squares regression submodule in sequence for inversion, and then perform zero lower limit truncation and summation normalization to obtain the occurrence morphology distribution vector containing the proportion of the 5 occurrence morphologies. S2: Construct a Bayesian network consisting of 13 discrete nodes: 5 nodes representing the proportion of occurrence morphology, 5 nodes representing process parameters, a migration activity index node, a reaction environment intensity index node, and a normalized leaching risk index node. The normalized leaching risk index node is divided into 5 states: low-risk, qualified, critical, exceeding limits, and severely exceeding limits. The occurrence morphology distribution vector and the collected process parameters are used together as evidence observations of the Bayesian network. Posterior probability inference is performed on the normalized leaching risk index node, and the sum of the posterior probabilities corresponding to the critical, exceeding, and severely exceeding limits states is taken as the leaching risk probability of the batch to be judged. S3: When the stabilization pass probability threshold is less than the stabilization warning probability threshold; when the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released; when the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, the leaching toxicity retest is initiated; when the leaching risk probability is greater than or equal to the stabilization warning probability threshold, the chelating agent is added, and after the addition is completed, S2 is executed again until the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released.

[0018] For a batch of approximately 5 to 10 grams of fly ash from waste incineration to be assessed, it was first poured into a clean borosilicate glass weighing bottle and placed in a constant-temperature oven set at 105 degrees Celsius for 3 hours. After drying, the fly ash was transferred to a planetary ball mill jar and ball-milled using zirconia grinding balls as the grinding medium until all samples could pass through a 200-mesh stainless steel sieve, meaning the particle size did not exceed 75 micrometers.

[0019] After ball milling and sieving, 0.2 g of the ground fly ash powder was weighed using an electronic analytical balance and slowly and evenly poured into the circular cavity of a polyethylene tableting mold. The inner diameter of the mold cavity was set to 13 mm. The mold containing the fly ash was placed into a manual tableting machine, and the pressure was slowly increased to 10 MPa and held for 2 minutes. After releasing the mold, the resulting circular sheet sample had a diameter of 13 mm and a thickness of 1 mm.

[0020] like Figure 1 As shown, a circular sheet sample is vertically placed into the sample holder of a terahertz time-domain spectrometer, with the sample surface perpendicular to the terahertz pulse propagation direction and the sample center coaxial with the optical path. The terahertz time-domain spectrometer transmits terahertz pulses covering frequencies from 0.1 terahertz to 3 terahertz through the circular sheet sample, scanning the optical time delay line in steps of approximately 5 femtoseconds to completely cover a 60 picosecond time window, recording one sample terahertz time-domain signal. .in, Indicates the scan delay time. Indicates at time The measured transient terahertz electric field amplitude is expressed as instrument readings and is given in dimensionless form.

[0021] After removing the circular sheet sample, a second scan was performed at the unloaded position (i.e., the terahertz pulse passes through the air background) with the same step size and time window length as described above, to obtain the blank reference terahertz time-domain signal. .in, Indicates at time The measured transient electric field amplitude of the terahertz pulse passing through the air background. See also Figure 1The sample terahertz time-domain signal exhibits a time delay due to the sample's refractive index and an amplitude attenuation due to the sample's absorption, compared to the blank reference terahertz time-domain signal. Furthermore, it exhibits free induction attenuation of the sample after the main pulse.

[0022] After obtaining the sample and reference two-channel time-domain signals, a frequency-domain transformation is performed to obtain the terahertz absorption spectrum. Before performing the fast Fourier transform, the signals are first processed separately... and Apply a Hanning or Hamming window; then perform a Fast Fourier Transform of the same length on the two windowed time-domain signals to obtain the sample terahertz frequency domain signal. With reference terahertz frequency domain signal Both are related to frequency. The complex-valued function. For each discrete frequency point. The amplitude ratio of the sample frequency domain signal to the reference frequency domain signal is calculated, and the natural logarithm of this ratio is taken as its negative value as the absorbance of the sample at that frequency. Mathematically, let the absorbance be denoted as . Then there is .in, Represents frequency Absorbance at [location] (dimensionless) This indicates that the sample terahertz frequency domain signal is at a frequency of The modulus of the complex amplitude at that point This indicates the reference terahertz frequency domain signal at frequency. The modulus of the complex amplitude at that point Represents the natural logarithm operation. Representing frequency. The frequency corresponding to each discrete frequency point. according to Arrange the samples in ascending order to obtain the terahertz absorption spectrum of that batch of fly ash.

[0023] like Figure 2 As shown, the five occurrence forms in waste incineration fly ash exhibit distinct vibrational fingerprints in a frequency range below approximately 5 terahertz, each concentrated within its respective frequency window: the exchangeable state is typically concentrated in the range of 0.20 terahertz to 0.55 terahertz, the carbonate-bound state is typically concentrated in the range of 0.55 terahertz to 0.95 terahertz, the iron-manganese oxide-bound state is typically concentrated in the range of 0.95 terahertz to 1.45 terahertz, the organic matter and sulfide-bound state is typically concentrated in the range of 1.45 terahertz to 2.05 terahertz, and the residue state is typically concentrated in the range of 2.05 terahertz to 2.95 terahertz. These five frequency windows represent the characteristic intervals of the five occurrence forms.

[0024] The above five frequency band boundaries are merely exemplary initial boundaries. The final frequency band boundaries of the five stored morphological feature intervals are determined by the pre-calibration process described later through iterative optimization.

[0025] After determining the five characteristic intervals, the following operation was performed on the terahertz absorption spectra of one batch: Within the frequency range covered by each characteristic interval, the maximum absorbance was searched and this maximum value was taken as the spectral characteristic quantity of that characteristic interval. The five extracted spectral characteristic quantities were concatenated into a column vector in the order of exchangeable state characteristic interval, carbonate-bound state characteristic interval, iron-manganese oxide-bound state characteristic interval, organic matter and sulfide-bound state characteristic interval, and residue state characteristic interval, and denoted as the spectral characteristic vector of this batch. .in, This represents the spectral feature vector of this batch. to These represent the maximum absorbance values ​​within the characteristic ranges of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state, respectively. This represents the matrix transpose operation.

[0026] The spectral feature vector is fed into a pre-calibrated partial least squares discriminant-regression joint model to invert the proportion of different storage forms. This joint model consists of two cascaded parts: the first part (partial least squares discriminant analysis submodule) is used to identify the dominant storage form category of this batch of fly ash, and the second part (partial least squares regression submodule) is used to regress the preliminary proportion prediction values ​​of the five storage forms by combining the spectral feature vector and the dominant storage form category output from the first stage.

[0027] like Figure 3 As shown, the partial least squares discriminant analysis submodule divides the calibrated fly ash samples with different dominant morphological categories into five mutually separate categories in the latent variable space. The specific execution method of the first stage is as follows: In the pre-calibration stage, a discriminant matrix is ​​obtained. In the online determination stage, spectral feature vectors are used as the basis for the determination. Calculate the discriminant score vector for the input. ,Right now ;by The category index corresponding to the largest component is used as the dominant memory type category for this batch of fly ash. .in, This represents the discriminant matrix obtained from partial least squares discriminant analysis. This represents the spectral feature vector of this batch. This represents the discriminant score vector. The subscript representing the dominant memory type category has the following value set: These correspond to five categories: exchangeable state-dominated, carbonate-bound state-dominated, iron-manganese oxide-bound state-dominated, organic matter and sulfide-bound state-dominated, and residue-bound state-dominated, respectively.

[0028] The specific execution method of the second phase is as follows: In the pre-calibration phase, for each dominant memory type category... Each obtained one set of regression matrices There are 5 groups in total; during the online judgment phase, based on the output of the first phase... Select the corresponding And for the spectral eigenvectors Performing linear regression yields a preliminary proportion prediction vector of length 5. ,Right now .in, Indicates the dominant endowment form category The corresponding regression matrix, This represents the initial proportion prediction vector directly output by the partial least squares regression submodule, with a length of 5; ( The numbers represent the preliminary predicted proportions of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state, respectively.

[0029] Directly obtained from partial least squares regression It cannot be used directly as the final percentage for two reasons: First, as a physical quantity, the percentage must satisfy certain conditions. That is, there is no negative proportion; however, partial least squares regression is essentially a linear regression. When the true proportion of a certain form is close to zero and the spectral noise is high, the regression output may show a slightly negative predicted value, which is inconsistent with the physical meaning. Secondly, the five occurrence forms constitute the entire occurrence state of this heavy metal in fly ash, and the sum of the five proportions should strictly be 1; however, the regression output itself does not guarantee this normalization constraint, and the five proportions... The sum usually deviates slightly from 1. To satisfy both physical constraints simultaneously, a two-step sequential post-processing step is introduced after the regression output.

[0030] Step 1: Zero Lower Bound Truncation: For each preliminary proportion prediction value output by the partial least squares regression submodule, if the value is less than zero, replace it with zero; otherwise, retain the original value. Let the proportion vector after zero lower bound truncation be... ,but .in, Indicates the first The percentage of each memory allocation after zero lower bound truncation. This indicates the preliminary predicted proportion of this pattern. This indicates an operation that takes the larger of two input values. All values ​​are truncated after processing. In the range Within this range, the non-negativity constraint is satisfied. Step 2 involves summation and normalization: after truncating to the zero lower bound, five values ​​are obtained. Divide each value by the sum of these five values ​​to obtain the final allocation proportion. Let the final assignment morphology distribution vector be... ,but .in, This indicates the summation and normalization process for the th... The final percentage of each allocation form, This indicates the percentage of the form after truncation at the zero lower bound. This represents the sum of the percentages after truncating the five zero lower bounds. The summation index takes values ​​from 1 to 5. After normalization, the 5 values ​​are... Strictly meet and This aligns with the physical meaning of the proportion of occurrence forms. The five normalized values, arranged in the order of exchangeable state proportion, carbonate-bound state proportion, iron-manganese oxide-bound state proportion, organic matter and sulfide-bound state proportion, and residual state proportion, are concatenated to form the occurrence form distribution vector of this batch of fly ash. Subsequently, these observations serve as evidence for the proportion of the five occurrence forms in the subsequent causal inference stage.

[0031] Partial least squares discriminant-regression joint models must be pre-calibrated before being used for online judgment. The core objectives of calibration are threefold: first, to determine the final frequency band boundaries of the five morphological feature intervals; second, to determine the discriminant matrix for the first stage. Third, determine the regression matrix for each of the five branches in the second stage. , , , and .

[0032] To ensure the calibration results cover the distribution combinations of the five common occurrence forms in industrial scenarios, a calibration fly ash sample set of 200 fly ash samples with known occurrence form distributions was selected. These samples should cover different furnace types, feed calorific values, chlorine content, and flue gas treatment processes, and should be pre-screened to ensure that the proportion of each occurrence form in the set covers the specified range. This avoids an overabundance or underabundance of samples dominated by a particular form.

[0033] Tessier sequential extraction experiments were performed on each of the calibrated fly ash samples in the calibration sample set. This experiment is a standard chemical method for five-stage fractional extraction of heavy metal occurrence forms in fly ash. Five extracts were obtained sequentially, corresponding to exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state. The mass concentration of the target heavy metal in each of the five extracts was determined by inductively coupled plasma mass spectrometry. The mass concentration of each of the five extracts was divided by the sum of the mass concentrations of the five extracts to obtain the measured proportion of the five occurrence forms in the sample.

[0034] Two preparatory operations were performed in parallel for each calibrated fly ash sample. First, the sample was prepared into a specimen according to the circular sheet specimen preparation process described above, and its terahertz absorption spectrum was acquired. Second, the solid precipitate obtained in each of the five stages of Tessier's continuous extraction and the residue corresponding to the fifth stage were taken, dried at 105 degrees Celsius for 3 hours, and then pressed into five pure-form circular sheet specimens according to the same pressing method described above. The terahertz absorption spectrum of each pure-form specimen was acquired. In the pure-form terahertz absorption spectrum, the characteristic absorption peak corresponding to the morphology can be identified as a single main peak. The frequency range covered by the full width at half maximum (FWHM) of this main peak is used as the initial frequency band boundary of the characteristic interval corresponding to the morphology.

[0035] Using the terahertz absorption spectra of all 200 calibrated fly ash samples and their corresponding measured proportions as training data, a joint model was trained and iteratively optimized for frequency band boundaries. Specifically, within the current five feature intervals, a spectral feature vector of length 5 was extracted from the terahertz absorption spectrum of each calibrated fly ash sample according to the principle of "maximum absorbance within the interval." These 200 spectral feature vectors from the 200 calibrated fly ash samples were then stacked row-wise to form a spectral feature matrix. (Scale is 200 rows and 5 columns); stack the 5 measured proportions of each calibrated fly ash sample in the same order row to form a measured proportion matrix. (Scale is 200 rows and 5 columns); for each calibrated fly ash sample, the dominant occurrence morphology category is determined based on the morphology corresponding to the largest of its 5 measured proportions, resulting in a category vector composed of 200 dominant occurrence morphology categories. Then with and As input, the partial least squares discriminant analysis algorithm is used to... Perform latent variable decomposition and iterative solution to obtain the discriminant matrix for the current round. Based on this, for each dominant memory type category From 200 calibrated fly ash samples, the dominant occurrence morphology was selected as follows: A subset of samples is used, with the spectral feature vectors of each sample in the subset and their corresponding measured proportions forming the training set. A partial least squares regression algorithm is then used to iteratively solve for the regression matrix of that category. .in, This represents a spectral feature matrix obtained by stacking 200 spectral feature vectors row by row. This represents the measured proportion matrix obtained by stacking the corresponding measured proportions row by row. This represents the dominant attributed morphology category vector. Represents the discriminant matrix, Representation and Category The corresponding regression matrix.

[0036] Get the current round With 5 groups Then, a complete inversion and two-step post-processing were performed on each calibrated fly ash sample to obtain the predicted proportion vector of that sample; the mean square error between the predicted proportion vectors and the corresponding measured proportion vectors of all 200 calibrated fly ash samples was used as the objective function. ,Right now .in, This represents the mean squared error objective function based on the proportions of the five memory configurations during the calibration phase. Indicates the first The first calibrated fly ash sample obtained from the inversion The final percentage of each allocation form, Indicates the first The first calibrated fly ash sample was measured by Tessier's continuous extraction experiment. The measured percentage of each memory type The sample index ranges from 1 to 200. The value of the assigned morphological index is 1 to 5.

[0037] The lower and upper boundaries of the five characteristic intervals are each subjected to small-amplitude perturbations with a preset step size of 0.005 terahertz. From all perturbation directions, a direction is selected that... The direction of the fastest descent is used as the boundary update direction for this round. After updating the boundary, the spectral feature vector is re-extracted and the solution is re-solved based on the updated boundary. With 5 groups and recalculate Repeat this process until boundary disturbances in all directions no longer cause... Until the frequency drops further, the frequency band boundary at this point is the final frequency band boundary of the five memory morphology feature intervals. With 5 groups This is the final set of discriminant and regression matrices, completing the pre-calibration.

[0038] The partial least squares discriminant-regression joint model, after the above calibration, shows good consistency between its inversion proportion and the measured proportion extracted by Tessier continuous extraction, such as... Figure 4 As shown. For each batch to be judged, the sample is prepared according to the above procedure, the terahertz absorption spectrum is collected and the spectral feature vector is extracted. The preliminary proportion prediction vector is obtained by the category identification of the partial least squares discriminant analysis submodule and the proportion regression of the partial least squares regression submodule. Then, the distribution vector of the occurrence form of fly ash in this batch is obtained by the zero lower limit truncation and summation normalization. This vector serves as the evidence observation value of the proportion of the five occurrence forms in the subsequent causal inference.

[0039] Based on the above considerations, a Bayesian network consisting of 13 discrete nodes is deployed in S2. These 13 discrete nodes are functionally divided into three groups: Group 1 consists of five nodes representing the proportion of different states of matter: exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state; Group 2 consists of five nodes representing five process parameters: reaction slurry pH, reaction slurry redox potential, reaction slurry chloride ion concentration, chelating agent dosage, and stirring time; Group 3 consists of three nodes: migration activity index, reaction environment intensity index, and normalized leaching risk index.

[0040] In the third group of nodes, two intermediate nodes, the migration activity index node and the reaction environment intensity index node, are arranged instead of all five occurrence mode proportion nodes and five process parameter nodes directly pointing to the normalized leaching risk index node. This is to address the scalability bottleneck of the conditional probability table. If all 10 nodes directly point to the normalized leaching risk index node, the number of parent state combinations for that node will reach [a certain threshold]. In 200 calibrated fly ash samples, most parent state combinations could not be observed, making reliable estimation difficult. After introducing two intermediate nodes, the migration activity index node receives all inputs from five occurrence morphology proportion nodes and determines its state value according to a pre-made migration activity mapping table. The reaction environment intensity index node receives inputs from three reaction slurry process parameter nodes: reaction slurry pH, reaction slurry redox potential, and reaction slurry chloride ion concentration, and determines its state value according to a pre-made reaction environment intensity mapping table. The direct parent nodes of the normalized leaching risk index node are reduced to four nodes: migration activity index node, reaction environment intensity index node, chelating agent dosage node, and stirring time node. The number of its parent state combinations is... The average coverage of 200 calibrated fly ash samples was approximately per combination. After several observations, a statistically meaningful conditional probability table can be obtained by superimposing Dirichlet prior smoothing.

[0041] The directed edges of the Bayesian network are connected according to the above causal grouping: five nodes—the commutative state percentage node, the carbonate-bound state percentage node, the iron-manganese oxide-bound state percentage node, the organic matter and sulfide-bound state percentage node, and the residue state percentage node—each lead one directed edge to the migration activity index node; three nodes—the reaction slurry pH node, the reaction slurry redox potential node, and the reaction slurry chloride ion concentration node—each lead one directed edge to the reaction environment intensity index node; and four nodes—the migration activity index node, the reaction environment intensity index node, the chelating agent dosage node, and the stirring time node—each lead one directed edge to the normalized leaching risk index node. Besides these 10 directed edges, there are no other directed edges, reverse edges, or self-loops, ensuring that the Bayesian network meets the topological requirements of a directed acyclic graph.

[0042] Of the 13 discrete nodes, excluding the normalized leaching risk index node, the remaining 12 nodes are divided into 5 states at equal intervals according to their respective value ranges. The value ranges for each node are predetermined by the treatment plant based on historical data. A typical example of the value ranges and equal-interval state divisions is as follows: The value ranges for the five occurrence morphology percentage nodes are closed intervals from 0 to 1, with the corresponding sub-intervals being [0, 0.2), [0.2, 0.4), [0.4, 0.6), [0.6, 0.8), and [0.8, 1.0]; the value range for the pH node of the reaction slurry is 7.0 to 13.0, with the corresponding sub-intervals being [7.0, 8.2), [8.2, 9.4), [9.4, 10.6), [10.6, 11.8), and [11.8, 13.0]; the value range for the oxidation-reduction potential node of the reaction slurry is -300 mV to +500 mV, with the corresponding sub-intervals being [-300, -140), [-140, 20), [20, 180), [180, 300), [140, 200), [20, 180), [180, 300], [140, 200], [20, 180], ... 40) and [340,500], all in millivolts; the value range of the chloride ion concentration node in the reaction slurry is 0 to 50 g / L, and the sub-intervals corresponding to the 5 states are [0,10), [10,20), [20,30), [30,40) and [40,50], all in g / L; the value range of the chelating agent dosage node is 0% to 10% of the dry weight of fly ash, and the sub-intervals corresponding to the 5 states are [0,2)%, [2,4)%, [4,6)%, [6,8)% and [8,10]%; the value range of the stirring time node is 0 to 60 minutes, and the sub-intervals corresponding to the 5 states are [0,12), [12,24), [24,36), [36,48) and [48,60] minutes; the values ​​of the migration activity index node and the reaction environment intensity index node are discrete integers from 1 to 5, and the 5 integers are directly used as the 5 states.

[0043] The normalized leaching risk index nodes are classified in a different way than the 12 nodes mentioned above: the value range of each node is divided into 5 states by four segment values: the risk compliance segment threshold, the risk warning segment threshold, the value 1, and the risk severe exceedance segment threshold. The four segment values ​​are arranged in ascending order. The value 1 indicates that the leaching concentration of at least one target heavy metal is exactly equal to the leaching toxicity limit corresponding to GB16889, which is the dividing point between compliance and non-compliance. The risk compliance segment threshold and the risk warning segment threshold are below the value 1, and the risk severe exceedance segment threshold is above the value 1. In the implementation of the project, one feasible combination of segmented values ​​is as follows: the threshold for qualified risk segment is 0.5, the threshold for risk warning segment is 0.8, the value of 1 is 1.0, and the threshold for severe risk exceeding the limit segment is 2.0. The four segmented values ​​divide the normalized leaching risk index nodes into the following states: low risk (values ​​falling within [0, 0.5)), qualified (values ​​falling within [0.5, 0.8)), critical (values ​​falling within [0.8, 1.0)), exceeding the limit (values ​​falling within [1.0, 2.0)), and severe exceeding the limit (values ​​falling within [2.0, ∞)).

[0044] The precise definition of the normalized leaching risk index is given below. For one batch of chelated solidification product, the leachate is obtained according to the leaching toxicity test method for acetic acid buffer solution specified in HJ / T300; the leaching concentration of each target heavy metal in the leachate is determined by inductively coupled plasma mass spectrometry or inductively coupled plasma emission spectroscopy. Here, the target heavy metals are at least one of the inorganic pollutants whose leaching toxicity limits are specified in GB16889, and the specific list is selected by each treatment plant based on the type of waste received and regulatory requirements. For each target heavy metal... Based on its leaching concentration Divide by the leaching toxicity limit corresponding to GB16889 The limit normalized leaching quotient of the target heavy metal was obtained. A total of The normalized leaching risk index for this batch is [value missing]. Take this The maximum value among the normalized leaching quotients of the limit values, i.e. .in, Indicates the first The leaching concentration of the target heavy metal in the leachate, expressed in milligrams per liter; Indicates the first The leaching toxicity limits for the target heavy metals in GB16889, in milligrams per liter; Indicates the first The normalized leaching quotient of the limit for the target heavy metal is dimensionless; This represents the normalized leaching risk index of the chelated solidified product in this batch, which is dimensionless. This indicates the number of target heavy metal species selected. ; For the target heavy metal index, the value ranges from 1 to Integers; This indicates the operation of taking the maximum value among the parameters.

[0045] The aggregation method using the maximum value as the normalized leaching risk index aligns with the compliance judgment rule of GB16889, which requires compliance with each item individually: if the normalized leaching dilution factor for any single target heavy metal exceeds 1.0, the batch is considered non-compliant. Maximum value aggregation reflects this bottleneck effect, preventing high leaching dilution from other low leaching dilution factors and subsequent misjudgment. In scenarios where only one key target heavy metal needs monitoring... A value of 1 can be taken, and the normalized leaching risk index degenerates into the limit normalized leaching quotient of the single heavy metal.

[0046] The conditional probability table of the migration activity index nodes is assigned values ​​by the migration activity mapping table. The migration activity mapping table is constructed according to a chemical sequence where the migration activity decreases in the order of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic and sulfide-bound state, and residue state. One possible construction method is to assign migration activity weights to the five occurrence forms, for example, exchangeable state weights. Carbonate binding state weights weight of iron-manganese oxide binding states Weights of organic matter and sulfide binding states Residual state weights The weights decrease sequentially in the order of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state; for each state value of the node representing the proportion of the five occurrence forms. (Round to the nearest integer from 1 to 5), press Convert to the midpoint of the sub-interval corresponding to this state; calculate the migration activity score. ;Will Its value range (after the weights are selected, the theoretical range is the closed interval 0 to...) The region is divided into 5 equally spaced sub-intervals to obtain the state values ​​of the migration activity index nodes. Indicates the first The migration activity weights of various occurrence forms are dimensionless; Indicates the first The state value of each assigned type node is an integer from 1 to 5. Represents state value The midpoint of the corresponding subinterval; The migration activity score is dimensionless. For morphological indexing, integers from 1 to 5 are used in the following order: exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic and sulfide-bound state, and residual state. The migration activity mapping table covers all... Each parent state combination specifies a unique migration activity index node state value, and the conditional probability table of this node is a deterministic mapping.

[0047] The reaction environment intensity mapping table is constructed based on the chemical law of chelation reactions, which states that the reaction environment intensity increases as the pH of the reaction slurry increases, the redox potential of the reaction slurry decreases, and the chloride ion concentration of the reaction slurry decreases. One possible construction method is to convert the state values ​​of the pH node, the redox potential node, and the chloride ion concentration node of the reaction slurry into normalized values ​​within their respective ranges. The normalized value of the pH of the reaction slurry is denoted as... The normalized value of the redox potential of the reaction slurry is: The normalized value of chloride ion concentration in the reaction slurry is: (All values ​​are dimensionless and range from 0 to 1 in a closed interval); Define the intensity score of the reaction environment. ,Require and ;Will The value range (closed interval 0 to 1) is divided into 5 equally spaced sub-intervals to obtain the state values ​​of the nodes reflecting the environmental intensity index. This represents the normalized value of the pH of the reaction slurry; This represents the normalized value of the redox potential of the reaction slurry; The normalized value representing the chloride ion concentration in the reaction slurry; This indicates the weight of the pH value of the reaction slurry in the intensity score of the reaction environment; it is dimensionless. This indicates the weight of the redox potential term in the reaction slurry, and is dimensionless. The weight of the chloride ion concentration term in the reaction slurry is dimensionless. This represents the intensity score of the reaction environment, dimensionless. A reasonable value for Group 1 in industrial practice is... , , This reflects the dominant influence of pH on chelation reactions, the secondary influence of redox potential, and the auxiliary influence of chloride ions in industrial experience. The reaction environment intensity mapping table covers all... Each parent state combination specifies a unique response environment intensity index node state value, and the conditional probability table for this node is a deterministic mapping. Once selected, the two mapping tables serve as fixed components of the Bayesian network structure to ensure the homology between the calibration and online decision-making stages.

[0048] In the Bayesian network, the conditional probability tables of the remaining 11 nodes (excluding the migration activity index node and the reaction environment intensity index node) are assigned values ​​by calibration experiments. Of these 11 nodes, 10 nodes (5 nodes representing the proportion of occurrence forms and 5 nodes representing process parameters) have no parent nodes (referred to as root nodes in the Bayesian network), and their conditional probability tables degenerate into prior probability distributions of the node's own states, i.e., probability vectors of length 5. The normalized leaching risk index node has 4 parent nodes, and its conditional probability table consists of 5 state probability distributions defined for each combination of parent states. The combination of parent states, total There are several probability terms. After deducting the constraint that the sum of the probabilities of the five states under each parent state combination is 1, the free parameters are: indivual.

[0049] For each of the 200 calibrated fly ash samples, two calibration operations were performed. The first calibration operation involved inverting the morphological distribution vector of the sample using a pre-calibrated partial least squares discriminant-regression joint model, along with its associated zero-lower-bound truncation and summation normalization processes. This vector served as the calibration observation for five morphological proportion nodes. The observations for the five morphological proportion nodes in the calibration phase must be from the same source as those in the online determination phase, i.e., they must all originate from the inversion results of the aforementioned joint model, rather than directly using the measured proportions extracted by Tessier, to avoid the conditional probability table inheriting the discrepancy between measured and inverted values. The second calibration procedure: In a chelation reactor consistent with the on-site treatment process, a chelation reaction was performed on the sample with a preset chelating agent dosage and a preset stirring time. At the outlet of the chelation reactor, the pH, redox potential, and chloride ion concentration of the reaction slurry were measured in situ using a pH electrode, a redox potential electrode, and a chloride ion selective electrode. The dosage of chelating agent and the stirring time were recorded, resulting in a total of 5 process parameter observations. The leachate of the obtained chelated solidified product was obtained according to the HJ / T300 acetic acid buffer solution leaching toxicity test method, and the normalized leaching risk index corresponding to the sample was calculated according to the above definition of normalized leaching risk index.

[0050] All observations obtained from the above two calibration operations are discretized into integer state values ​​according to the state division method of the corresponding nodes in the Bayesian network: the proportion of the five occurrence forms is mapped to the corresponding state value according to the sub-interval of the five occurrence form proportion nodes; the five process parameters of the reaction slurry pH, reaction slurry redox potential, reaction slurry chloride ion concentration, chelating agent dosage, and stirring time are mapped to the corresponding state values ​​according to the sub-interval of their respective process parameter nodes; the normalized leaching risk index is mapped to the corresponding state value according to the four segmented values; the state value of the migration activity index node is obtained by looking up the state values ​​of the five occurrence form proportion nodes through the migration activity mapping table; the state value of the reaction environment intensity index node is obtained by looking up the state values ​​of the three reaction slurry process parameter nodes through the reaction environment intensity mapping table. Each calibrated fly ash sample thus provides a complete 13-node state value record, resulting in a total of 200 records.

[0051] The conditional probability table for the 10 root nodes is assigned as follows: For each root node, the sample count of each state value taken by the node is counted from the 200 state value records; the sample count of each state value is added to the pseudo-count of the Dirichlet prior to obtain the posterior count; the posterior count is divided by the sum of the posterior counts of all states to obtain the probability estimate of the state; the 5 probability estimates constitute the prior distribution of the root node. The conditional probability table for the normalized leaching risk index nodes is assigned as follows: For the 625 state combinations of the four parent nodes, the sample count of five states for each parent state combination is counted in 200 state value records. The sample count of the five states under each parent state combination is added to the pseudo-count of the Dirichlet prior to obtain the posterior count of the five states under the parent state combination. The conditional probability of the five states under the parent state combination is obtained by dividing the posterior count of each state by the sum of the posterior counts of the five states under the parent state combination. The conditional probabilities of the 625 parent state combinations together constitute the conditional probability table for the normalized leaching risk index nodes.

[0052] Formally, set nodes Combining in the parent state Downward state The sample count is The pseudo-count per state of the Dirichlet symmetric prior is Then the maximum a posteriori estimate of the Dirichlet prior smoothing is: .in, This represents the currently estimated Bayesian network node; This represents the state combination of all parent nodes of this node; for the root node, this item is empty. This represents a state of the node, taking the integer value from 1 to 5. Indicates the combination of parent states Down Get state Sample count; Let represent the per-state pseudo-count of the Dirichlet symmetric prior, i.e., the number of equal pseudo-samples assigned to each state 1, taking a positive real number; Represents conditional probability Dirichlet prior smooth maximum a posteriori estimate; To calculate the summation index, integers from 1 to 5 are used. Group 1 commonly uses values ​​from 1 to 5 in industrial practice. This means assigning one spurious sample to each state, which is equivalent to Laplace plus 1 for smoothing; if the prior preference is a sharper distribution, then... (For example If the prior preferences are more evenly distributed, then we can take... (For example When the calibration data has high noise, Taking a larger value can enhance the smoothing stability. Dirichlet prior smoothing avoids the output of pure maximum likelihood estimation on the combination of parent states without observed samples. The problem arises from the indeterminacy of the form and the estimation of the probability of unobserved states as zero, which leads to the triggering of zero probability during inference.

[0053] After completing the above two calibration operations and the maximum a posteriori estimation of Dirichlet prior smoothing, the conditional probability tables of all 11 calibrated nodes are assigned values. Then, combined with the conditional probability tables of the two deterministic nodes, the migration activity index node and the reaction environment intensity index node, the entire process of assigning conditional probability tables to the Bayesian network is completed, and the Bayesian network enters a state where it can be inferred online.

[0054] During the online judgment phase, for each batch to be judged, the proportions of five occurrence forms are read from the distribution vector of occurrence forms in S1. The pH, redox potential, and chloride ion concentration of the reaction slurry are measured in situ by three electrodes at the discharge port of the chelation reactor. The dosage of chelating agent is recorded by the chelating agent metering pump, and the stirring time is recorded by the agitator control system. The above 10 observation data are mapped to integer state values ​​according to the sub-intervals of the corresponding nodes, serving as evidence observations for the five occurrence form proportion nodes and five process parameter nodes in the Bayesian network. The migration activity index node and the reaction environment intensity index node do not directly receive external evidence; their states are automatically determined by their respective mapping tables. The normalized leaching risk index node does not receive evidence but serves as a query node for posterior inference.

[0055] For the Bayesian network that has received the evidence observations, posterior probability inference is performed using the joint tree inference algorithm. The execution steps of the joint tree inference algorithm are as follows: First, a moralization operation is performed on the directed acyclic graph of the Bayesian network to obtain a moral graph. Triangulation is then performed on the moral graph to obtain a triangular graph. A connection graph is constructed using each maximal clique as a node, and spanning trees satisfying the joint tree property are extracted to obtain the joint tree. The potential functions of each maximal clique and the partition set are initialized according to the conditional probability table. The 10 evidence observations of the batch to be judged are substituted into the potential function of the corresponding maximal clique, and the potential of the non-evidence states is set to zero. Message passing is performed along both directions of the joint tree to converge the potential function of each maximal clique to the marginal joint distribution of the variable of that maximal clique. Finally, the posterior probability distribution of the node in the 5 states is obtained by marginalizing the maximal clique containing the normalized leaching risk index node. This Bayesian network has a total of 13 discrete nodes. The largest maximal clique in the joint tree contains the normalized leaching risk index node and its 4 parent nodes, totaling 5 nodes. Each state cell can complete message transmission and edge extraction in milliseconds with the computing power of a standard industrial controller.

[0056] like Figure 5 As shown, after obtaining the posterior probability distribution of the normalized leaching risk index nodes in the five states, the final aggregation of the leaching risk probability of the batch to be judged is performed: the sum of the three posterior probabilities corresponding to the critical state, the exceeding state, and the severely exceeding state is taken as the leaching risk probability of the batch to be judged. Formalistically, let's assume... , , Let represent the posterior probabilities of the normalized leaching risk index nodes obtained from posterior inference in the critical state, the over-limit state, and the severely over-limit state, respectively. Then we have: .in, This represents the leaching risk probability of the batch to be judged, with a value falling within the closed interval between 0 and 1; , , denoted as the probabilities of the critical state, the over-limit state, and the severely over-limit state in the posterior probability distribution of the normalized leaching risk index nodes in the five states, respectively. They are dimensionless and take values ​​between 0 and 1 in the closed interval.

[0057] Incorporating critical states into the aggregation process stems from a cautious release engineering approach: a critical state corresponds to the leaching concentration of at least one target heavy metal being close to but not exceeding the limit specified in GB 16889. Including this in risk aggregation allows batches closer to the regulatory threshold to automatically trigger re-inspection or supplementation. As an optional implementation, in scenarios with extremely stringent regulations, a portion of the posterior probability of the acceptable state can be further superimposed. China and Israel can construct a more risk-averse aggregation method; conversely, in a scenario with lax regulation, only the sum of the two items, exceeding the limit and serious exceeding the limit, can be taken.

[0058] Thus, for each batch to be judged, the leaching risk probability of that batch is obtained through evidence discretization, conditional probability table lookup, joint tree inference, and summation of three posterior probabilities. This leaching risk probability, together with the stabilization pass probability threshold and the stabilization warning probability threshold, enters the judgment and supplementation closed loop of S3.

[0059] Leaching risk probability obtained from stage S2 , which is a scalar falling within the closed interval between 0 and 1. To balance regulatory risk and industrial economics, S3 adopts a three-level judgment structure, taking three actions—release, re-inspection, and supplementation—in the three intervals of low, medium, and high leaching risk probability, respectively.

[0060] The three-tier judgment structure is derived from two probability thresholds: a stabilization pass probability threshold and a stabilization warning probability threshold. The stabilization pass probability threshold is denoted as... The stabilization early warning probability threshold is denoted as .in, This represents the stabilization pass probability threshold, which is the upper limit of the leaching risk probability for direct release. Its value falls within the closed interval between 0 and 1. This represents the stabilization warning probability threshold, i.e., the lower limit of the leaching risk probability that necessitates the supplementation process; its value also falls within the closed interval between 0 and 1. (Mandatory constraint) , making the closed interval Open interval Closed interval The three elements are mutually exclusive and their union covers each other. The range of values ​​is determined so that the leaching risk probability of any batch to be judged will fall into exactly one level, thus avoiding ambiguity or overlap in the judgment.

[0061] The specific values ​​for the two probability thresholds are determined by each treatment plant based on the stringency of regulations, the acceptable probability of exceeding standards in random inspections, and operational economics; common combinations of values ​​used in industrial practice are as follows: , The allowable threshold of 0.10 means that only batches with an actual probability of exceeding the limit of no more than 10% are allowed to be released directly, while the mandatory supplementary threshold of 0.30 means that batches with an actual probability of exceeding the limit of more than 30% will be forcibly included in the supplementary threshold. (Middle range) For batches with an actual probability of exceeding standards between 10% and 30%, chemical re-testing is used, replacing probability inference with on-site chemical testing as the final determination basis. This preserves regulatory certainty while avoiding excessive supplementation. Optionally, in landfill scenarios in core urban areas with extremely strict regulations, [the following can be used]. Further down to 0.05 and Reduced to 0.20 to increase safety margin; in scenarios with relatively relaxed regulations and prioritizing industrial economics, it can be... Relaxed to 0.15 and Relax to 0.40 to reduce unnecessary additions. Regardless of which set of values ​​is selected, it must always be maintained. .

[0062] The specific execution process of the three response actions is described below.

[0063] Tier 1: When the leaching risk probability is less than or equal to the stabilization pass probability threshold, i.e. At this point, the batch meets the conditions for direct release. The control system sets the batch status to stable and qualified and pushes it to the human-machine interface. It opens the discharge valve of the chelation reactor, allowing the chelated solidified product to be transported to the landfill operation area via the landfill conveyor line. The weighing sensor at the beginning of the landfill conveyor line records the discharge mass. After the discharge is completed, the discharge valve of the chelation reactor is closed, the batch status is updated to the released status, and the batch record database retains the leaching risk probability, five process parameter observations, morphological distribution vector, and release timestamp for traceability.

[0064] Level 2: When the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, i.e. At this point, the batch enters the re-inspection process. The control system sets the batch status to "pending verification" and pushes it to the human-machine interface. The discharge valve of the chelation reactor is not opened temporarily. The automatic sampling robot is triggered to collect 50 grams of chelated solidified product at the discharge port of the chelation reactor as a re-inspection sample. The re-inspection sample is then transported to the on-site chemical analysis laboratory in a sealed container.

[0065] After the re-inspection samples arrive at the laboratory, they undergo laboratory leaching toxicity retesting according to the acetic acid buffer solution leaching toxicity test method specified in HJ / T300. The leaching concentration of each target heavy metal in the leachate is determined using inductively coupled plasma mass spectrometry or inductively coupled plasma atomic emission spectrometry to obtain the actual leaching toxicity test results for this batch. The control system compares the leaching concentration of each target heavy metal with the corresponding leaching toxicity limit in GB16889: if the leaching concentration of all target heavy metals does not exceed the corresponding limit, the batch is deemed qualified and the same release procedure as in stage 1 is performed; if the leaching concentration of at least one target heavy metal exceeds the corresponding limit, the batch is deemed unqualified and is transferred to the supplementation process or subjected to additional treatment according to on-site regulations. Once the re-inspection process is initiated, the actual leaching toxicity test results, rather than Bayesian network posterior inference, are used as the final judgment basis for this batch, serving as an objective verification of the inference results.

[0066] Tier 3: When the leaching risk probability is greater than or equal to the stabilization warning probability threshold, i.e. At this point, the batch enters the replenishment process. The control system sets the batch status to leaching risk and pushes it to the human-machine interface, closes the discharge valve of the chelation reactor, opens the solenoid valve of the replenishment channel, and adds chelating agent to the chelation reactor by the chelating agent metering pump according to the preset ratio.

[0067] The method for determining the supplementation ratio is as follows: the additional chelating dose for each supplementation is taken as a preset ratio of the chelating dose already added in the first chelation reaction of that batch. This means that the amount added this time is recorded as [amount]. .in, This indicates the additional chelation dose added this time, in units of... Consistent (e.g., all are kilograms or fly ash dry basis mass percentages); This indicates the total chelating dose already administered to this batch prior to this supplement (including the initial dose and the cumulative dose from previous supplements); This indicates the supplementary ratio, and is a dimensionless value, typically between 0.2 and 0.5. Typically, it is set to 0.2. Optionally, The treatment plant can dynamically adjust the risk level based on historical data: the probability of initial leaching risk is only slightly higher than... batch, Take a smaller value (e.g., 0.15) to maintain a gentle replenishment; the probability of initial leaching risk is significantly higher than... batches (e.g.) ), Use a larger value (e.g., 0.4) to quickly restore the sufficiency of the reaction supply. The additional amount is metered into the chelating reactor by the chelating agent metering pump and accumulated in the cumulative chelating agent dosage record for that batch.

[0068] After the chelating agent is added, close the solenoid valve of the addition channel, start the mixer control system, and continue mixing for the preset addition time. The preset addition time is recorded as follows: The typical duration is 10 to 20 minutes. This indicates the preset time for additional stirring, in minutes, and should be a positive real number. Typically, the time is 10 to 20 minutes. The control system monitors the mixing time in real time via the speed feedback signal from the mixer control system and a mixing time counter. When the cumulative mixing time reaches... When the time is up, stop stirring and record the cumulative stirring time for that batch. .

[0069] After the addition and stirring are completed, the control system initiates a re-collection of process parameters for this batch: at the outlet of the chelation reactor, the pH, redox potential, and chloride ion concentration of the reaction slurry are re-measured in situ using pH, redox potential, and chloride ion selective electrodes. The cumulative chelating agent dosage for this batch is read from the chelating agent metering pump record as the updated evidence observation value for the chelating agent dosage, and the cumulative stirring time for this batch is read from the mixer control system record as the updated evidence observation value for the stirring time. In each iteration of the addition cycle, the evidence observation values ​​for the five occurrence format percentage nodes are not re-collected; instead, the occurrence format distribution vector obtained when the batch first entered S1 is used, because simply adding chelating agent and continuing stirring during the addition stage is insufficient to cause a statistically distinguishable migration of the occurrence formats of heavy metals.

[0070] After the process parameters are reacquired, the control system inputs the retained morphological distribution vector and the updated evidence observations of the five process parameters into the Bayesian network, and re-executes the discretization, joint tree inference, and summation of the three posterior probabilities in stage S2 to obtain the updated leaching risk probability. The control system has The comparison is then performed again with the stabilization pass probability threshold: If The replenishment cycle terminates, and the status of this batch is updated from leaching risk indicator to stabilization qualified indicator. The control system opens the discharge valve of the chelation reactor, and the chelated solidified product of this batch enters the landfill operation area via the landfill conveyor line, completing the release; if The replenishment cycle continues: the control system closes the discharge valve of the chelation reactor again and opens the solenoid valve of the replenishment channel, according to the preset ratio. Add chelating agent (at this time) The value is updated to the cumulative chelating agent dosage for this batch up to the end of the previous cycle. Stirring continues. Repeat the above re-acquisition and S2 inference process, and continue this cycle until... Release the goods promptly and complete the final disposal of the batch.

[0071] Formally, let the iteration index of the supplementary loop be... , No. The leaching risk probability after the end of the cycle is denoted as: , No. The cumulative chelating agent dosage for this batch after the cycle is completed is recorded as follows: , No. The cumulative mixing time for this batch after the cycle is completed is recorded as follows: The supplementary loop follows a recursive relationship. and The loop exit condition is .in, Indicates the index of the iteration of the supplementary loop, starting from 1 and taking a positive integer; Indicates the first The leaching risk probability obtained by re-executing in phase S2 after the round replenishment cycle ends; Indicates the first The cumulative chelating agent dosage for this batch after the completion of the replenishment cycle, in units of... Consistent; Take the chelation dose added during the first chelation reaction of this batch; Indicates the first The cumulative mixing time for this batch after the replenishment cycle is completed, in units of... Consistent; Take the stirring time during the first chelation reaction of this batch.

[0072] like Figure 6 As shown, the leaching risk probability of different batches gradually decreases with the increase of replenishment cycles and tends to stabilize below the pass probability threshold. The replenishment cycle iterates automatically according to a recursive relationship, but in actual industrial systems, a cycle upper limit must be set to cope with extreme operating conditions. Let the preset upper limit of the replenishment cycle be denoted as . Typically, the values ​​are 5 to 8. This indicates the maximum number of replenishment cycles allowed, i.e., the maximum number of replenishment cycles that can be performed before the batch of chelated and cured products reaches the stabilization qualification mark, and is taken as a positive integer. The engineering basis for selecting 5 to 8 is: in Under the given conditions, after five rounds of replenishment, the cumulative dosage of the chelating agent in this batch reached the initial dosage. After 8 rounds of replenishment, it reached [a certain number of times]. If the leaching risk probability is still not suppressed below the stabilization qualification probability threshold, then the characteristics of this batch of fly ash have exceeded the effective treatment range of this chelation process. achieve and Still greater than When the control system executes a cycle timeout exit, it sets the batch status to a cycle timeout exit flag and triggers a human-machine interface alarm. On-site technicians then transfer the batch to the hazardous waste disposal unit, return it to the front-end process for secondary pretreatment, or perform additional chemical testing. Optionally, a cumulative chelating agent dosage limit or a single batch total disposal time limit can also be set as... In addition to the independent exit protection conditions, once any one of the upper limits is triggered, the loop will time out and exit.

[0073] While specific embodiments of the present invention have been described above, those skilled in the art should understand that these specific embodiments are merely illustrative. Those skilled in the art can omit, substitute, and modify the details of the above methods and systems in various ways without departing from the principles and essence of the present invention. For example, combining the above method steps to perform substantially the same function and achieve substantially the same result according to substantially the same method falls within the scope of the present invention. Therefore, the scope of the present invention is defined only by the appended claims.

Claims

1. A method for determining the chelation, solidification, and stabilization of fly ash from waste incineration, characterized in that, Includes the following steps: S1: Obtain the terahertz absorption spectrum of the fly ash sample of the batch to be judged, extract spectral feature quantities according to the 5 occurrence morphology feature intervals to obtain spectral feature vector; input the spectral feature vector into the partial least squares discriminant-regression joint model composed of the partial least squares discriminant analysis submodule and the partial least squares regression submodule in sequence for inversion, and then perform zero lower limit truncation and summation normalization to obtain the occurrence morphology distribution vector containing the proportion of the 5 occurrence morphologies. S2: Construct a Bayesian network consisting of 13 discrete nodes, including 5 occurrence morphology proportion nodes, 5 process parameter nodes, migration activity index nodes, reaction environment intensity index nodes, and normalized leaching risk index nodes. The normalized leaching risk index nodes are divided into 5 states: low risk state, qualified state, critical state, out-of-limit state, and severely out-of-limit state. The morphological distribution vector and the collected process parameters are used together as evidence observations of the Bayesian network. Posterior probability inference is performed on the nodes of the normalized leaching risk index. The sum of the posterior probabilities corresponding to the critical state, the over-limit state and the severe over-limit state is taken as the leaching risk probability of the batch to be judged. S3: When the stabilization pass probability threshold is less than the stabilization warning probability threshold; when the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released; when the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, the leaching toxicity retest is initiated; when the leaching risk probability is greater than or equal to the stabilization warning probability threshold, the chelating agent is added, and after the addition is completed, S2 is executed again until the leaching risk probability is less than or equal to the stabilization pass probability threshold, the chelated solidified product corresponding to the batch to be judged is released.

2. The method as described in claim 1, characterized in that, The steps in S1 to obtain the terahertz absorption spectrum of the fly ash sample to be judged include: drying the fly ash sample to be judged in a constant temperature oven at 105 degrees Celsius for 3 hours, then ball milling and sieving to a particle size of less than 75 micrometers; weighing 0.2 grams of the ground fly ash powder and pouring it into a polyethylene circular pressing mold, and pressing it under 10 MPa pressure for 2 minutes to obtain a circular sheet sample with a diameter of 13 mm and a thickness of 1 mm; and performing terahertz time-domain spectroscopy on the circular sheet sample to obtain the terahertz absorption spectrum.

3. The method as described in claim 2, characterized in that, The steps for acquiring terahertz time-domain spectrometry include: transmitting terahertz pulses with a frequency range of 0.1 terahertz to 3 terahertz through a circular sheet sample using a terahertz time-domain spectrometer, and sampling the terahertz pulses after they have passed through the circular sheet sample to obtain the sample terahertz time-domain signal; acquiring the blank reference terahertz time-domain signal of the air background using the same procedure; performing Fourier transform on the sample terahertz time-domain signal and the blank reference terahertz time-domain signal to obtain the sample terahertz frequency domain signal and the reference terahertz frequency domain signal, respectively; and taking the negative logarithm of the ratio of the frequency domain amplitudes of the sample terahertz frequency domain signal and the reference terahertz frequency domain signal to obtain the terahertz absorption spectrum.

4. The method as described in claim 1, characterized in that, The five occurrence forms are exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state; the five occurrence form characteristic intervals are the characteristic intervals of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state corresponding to the five occurrence forms; in S1, the maximum absorption intensity within each occurrence form characteristic interval is taken as the spectral characteristic quantity of the corresponding occurrence form characteristic interval, and the five spectral characteristic quantities are spliced ​​together in the listed order of the five occurrence form characteristic intervals to obtain the spectral characteristic vector; the occurrence form distribution vector is composed of the proportion of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state in the listed order.

5. The method as described in claim 4, characterized in that, The partial least squares discriminant analysis submodule takes the spectral feature vector as input to identify the dominant occurrence morphology category of the batch to be judged; the partial least squares regression submodule takes the spectral feature vector and the dominant occurrence morphology category as input to regress and output the preliminary predicted proportion of the five occurrence morphologies of the batch to be judged. The zero lower bound truncation process involves replacing values ​​less than zero in the initial percentage prediction with zero. The summation and normalization process involves dividing each of the five values ​​after truncation by the zero lower bound by the sum of the five values. After zero lower bound truncation and summation normalization, the storage morphology distribution vector is obtained; The calibration process of the partial least squares discriminant-regression joint model includes: selecting 200 fly ash samples with known distributions of storage morphologies to form a calibration fly ash sample set; performing Tessier continuous extraction experiments on each calibration fly ash sample in the calibration fly ash sample set to obtain the measured proportions of the five storage morphologies, and collecting terahertz absorption spectra; collecting terahertz absorption spectra of the Tessier extracts corresponding to the five storage morphologies after drying, grinding, and tableting, and using the frequency range corresponding to the full width at half maximum (FWHM) of the absorption peaks in the collected terahertz absorption spectra as the initial frequency band boundaries of the corresponding storage morphology characteristic intervals; using the terahertz absorption spectra and measured proportions of all calibration fly ash samples as training data, iteratively solving the partial least squares discriminant analysis (PLSA) algorithm and the partial least squares regression (PLR) algorithm to obtain the discriminant matrix of the PLSSA submodule and the regression matrix of the PLSSA submodule, and iteratively optimizing the initial frequency band boundaries to obtain the final frequency band boundaries of the five storage morphology characteristic intervals.

6. The method as described in claim 1, characterized in that, The five occurrence state percentage nodes are the exchangeable state percentage node, the carbonate-bound state percentage node, the iron-manganese oxide-bound state percentage node, the organic matter and sulfide-bound state percentage node, and the residue state percentage node; the five process parameter nodes are the reaction slurry pH node, the reaction slurry redox potential node, the reaction slurry chloride ion concentration node, the chelating agent dosage node, and the stirring time node; among the 13 discrete nodes, except for the normalized leaching risk index node, the remaining 12 nodes are divided into 5 states at equal intervals according to their respective value ranges; The normalized leaching risk index node is segmented into four segments: risk compliance threshold, risk warning threshold, value 1, and risk severe exceedance threshold. The four segment values ​​increase sequentially. Value 1 indicates that the normalized leaching quotient of at least one target heavy metal reaches the leaching toxicity limit corresponding to GB16889. The four segment values ​​sequentially divide the normalized leaching risk index node into five states: low risk, compliance, critical, exceedance, and severe exceedance.

7. The method as described in claim 6, characterized in that, The normalized leaching risk index is defined as follows: For each batch of chelated solidified product, leachate is obtained according to the HJ / T300 acetic acid buffer solution leaching toxicity test method. The leaching concentration of each target heavy metal in the leachate is divided by the leaching toxicity limit corresponding to GB16889 to obtain the normalized leaching quotient of each target heavy metal limit. The maximum value among all the normalized leaching quotients is taken as the normalized leaching risk index of the corresponding batch. The target heavy metal is at least one of the inorganic pollutants with leaching toxicity limits specified in GB16889. The directed connection relationship between the nodes in the Bayesian network is as follows: the five occurrence form proportion nodes lead to directed edges pointing to the migration activity index node; the reaction slurry pH node, the reaction slurry redox potential node, and the reaction slurry chloride ion concentration node lead to directed edges pointing to the reaction environment intensity index node; the migration activity index node, the reaction environment intensity index node, the chelating agent dosage node, and the stirring time node lead to directed edges pointing to the normalized leaching risk index node.

8. The method as described in claim 7, characterized in that, Both the migration activity index node and the reaction environment intensity index node are deterministic nodes; the conditional probability table of the migration activity index node is assigned by the migration activity mapping table; the conditional probability table of the reaction environment intensity index node is assigned by the reaction environment intensity mapping table; the migration activity mapping table is pre-prepared according to the chemical principle that the migration activity levels of exchangeable state, carbonate-bound state, iron-manganese oxide-bound state, organic matter and sulfide-bound state, and residue state decrease sequentially as defined by Tessier continuous extraction; the reaction environment intensity mapping table is pre-prepared according to the chelation reaction chemical principle that the reaction environment intensity level increases when the pH of the reaction slurry increases, the redox potential of the reaction slurry decreases, and the chloride ion concentration of the reaction slurry decreases; the conditional probability tables of the remaining 11 nodes, excluding the migration activity index node and the reaction environment intensity index node, are assigned by the calibration experiment; the calibration experiment procedure includes: performing two steps on each calibration fly ash sample in the calibration fly ash sample set described in claim 5. Calibration operations: The first step involves inverting the morphological distribution vector according to the steps described in claim 5; the second step involves performing a chelation reaction according to a preset chelating agent dosage and a preset stirring time to obtain a chelated solidified product, recording the pH, redox potential, chloride ion concentration, chelating agent dosage, and stirring time of the chelated solidified product's reaction slurry, and determining the normalized leaching risk index of the chelated solidified product according to the HJ / T300 acetic acid buffer solution leaching toxicity test method; the calibration observations obtained from the two calibration operations are discretized into state values ​​of the corresponding nodes according to the state partitioning method of the corresponding nodes in the Bayesian network, and the conditional probability tables of the 11 nodes are assigned values ​​one by one by the maximum a posteriori estimation algorithm with Dirichlet prior smoothing; the morphological distribution vector obtained by inversion according to the steps described in claim 5 is used as input evidence in both the calibration stage and the online judgment stage of the Bayesian network; the posterior probability inference in S2 is completed by the joint tree inference algorithm.

9. The method as described in claim 6, characterized in that, The process parameter acquisition steps in S2 include: measuring the pH, redox potential, and chloride ion concentration of the reaction slurry of the batch to be judged in situ at the outlet of the chelation reactor corresponding to the batch to be judged using pH electrode, redox potential electrode, and chloride ion selective electrode, respectively; recording the chelating agent dosage of the batch to be judged by the chelating agent metering pump; and recording the stirring time of the batch to be judged by the agitator control system. The evidence observation values ​​of the five process parameter nodes are the pH, redox potential, chloride ion concentration, chelating agent dosage, and stirring time of the batch to be judged.

10. The method as described in claim 1, characterized in that, In S3: When the leaching risk probability is less than or equal to the stabilization pass probability threshold, the control system outputs a stabilization pass indicator, and the chelated solidified product corresponding to the batch to be judged is released into the landfill conveyor line through the discharge valve of the chelation reactor; when the leaching risk probability is greater than the stabilization pass probability threshold but less than the stabilization warning probability threshold, the control system outputs a verification pending indicator and initiates the re-inspection process; the re-inspection process is as follows: an automatic sampling robot takes a 50-gram sample of the chelated solidified product at the discharge port of the chelation reactor, and performs laboratory tests on the sampled chelated solidified product according to the HJ / T300 acetic acid buffer solution leaching toxicity test method. Leaching toxicity retesting is conducted, and the actual leaching toxicity test results are used as the final basis for judging the batch to be judged. When the leaching risk probability is greater than or equal to the stabilization warning probability threshold, the control system outputs a leaching risk indicator and initiates the replenishment process. The replenishment process is as follows: close the discharge valve of the chelation reactor, add a preset proportion of chelating agent to the chelation reactor through the replenishment channel solenoid valve, and continue stirring for a preset replenishment stirring time. After the replenishment stirring is completed, S2 is re-executed using the morphological distribution vector of the batch to be judged until the leaching risk probability is less than or equal to the stabilization qualified probability threshold, at which point the control system outputs a stabilization qualified indicator.

Images