An artificial intelligence-based radioactive wastewater treatment supervision method and system

Abstract

The present application relates to the technical field of sewage treatment supervision, and specifically discloses a radioactive sewage treatment supervision method and system based on artificial intelligence. The method comprises the following steps: obtaining batch supervision data of sewage to be released for structured processing, obtaining batch information and establishing a batch certificate state; identifying disturbances based on the state, inferring an influence path through an artificial intelligence model to obtain a flow tracing identifier; extracting time lag response characteristics to correct measurement hidden variables, and reversely reconstructing physical working conditions to obtain a corrected batch certificate state; combining the batch information to perform forward physical analysis to generate a conservation expected state, and analyzing the physical self-consistency of actual data and the conservation expected state; and finally generating a release credibility level and outputting a differentiated supervision disposal strategy. The system comprises a batch establishment module, a flow tracing module, a state correction module, a self-consistency analysis module and a strategy construction module. The present application is beneficial to improving the reliability of the supervision strategy through objective data reconstruction and self-consistency verification.

Description

Technical Field

[0001] This invention relates to the field of wastewater treatment monitoring technology, specifically to a method and system for monitoring radioactive wastewater treatment based on artificial intelligence. Background Technology

[0002] Radioactive wastewater undergoes filtration, ion exchange, and evaporation concentration before entering temporary storage tanks. On-site monitoring relies heavily on continuous high-frequency data from online instruments and discrete low-frequency data from sampling and testing.

[0003] However, in actual operating conditions, radioactive conditions can cause radiation damage to the probe, leading to static bias in the measurement; and nuclides are very easy to adhere to and remain on the pipe wall, causing dynamic tailing of the instrument reading. Secondly, actual water treatment is not ideally continuous, and timestamp gaps or mismatches in water flow often occur between different process stages (such as below the lower limit of water flow or concentration errors).

[0004] The combined interference of physical and technological factors directly leads to discrepancies between online monitoring values ​​and sampled test values. Faced with suboptimal conditions, regulators struggle to determine whether the data anomalies are genuine exceedances caused by filtration equipment failure or false alarms triggered by sensor probe contamination and drift. This lack of traceability results in frequent delays or mandatory sampling and retesting of qualified wastewater due to false signals, or the cover-up of physical leaks, posing a significant hidden danger of illegal discharge of radioactive water.

[0005] Therefore, the present invention provides a method and system for monitoring radioactive wastewater treatment based on artificial intelligence. Summary of the Invention

[0006] The purpose of this invention is to provide an artificial intelligence-based method and system for monitoring and managing radioactive wastewater treatment, in order to solve the aforementioned background problems.

[0007] The objective of this invention can be achieved through the following technical solutions: An artificial intelligence-based method for monitoring the treatment of radioactive wastewater includes the following steps: Acquire batch monitoring data of radioactive wastewater awaiting release in the temporary storage tank, and perform batch process structured processing to obtain batch process information on the continuity of the current batch processing; and establish the certificate status of the current batch release basis based on the batch process information to determine the degree of self-consistency. Based on the approval status and associated underlying process characteristics, disturbances are identified, and disturbance feature vectors of affected states are constructed. Through artificial intelligence analysis models, the influence path of disturbance feature vectors is inferred to obtain the flow traceability identifiers of the main sources of influence on credible stability. Based on the flow traceability identifier, the time delay response characteristics of disturbance information between process nodes are extracted; the time delay response characteristics are used to correct the measurement hidden variables caused by probe radiation damage and tube wall nuclide retention in the batch certification status, and the physical conditions are reconstructed in reverse to obtain the corrected batch certification status. Based on batch process information, forward physical analysis along the process topology is performed using batch monitoring data to generate the conserved expected state of the target node; combined with the corrected batch certificate state, the physical consistency between the actual data and the conserved expected state is analyzed.

[0008] Furthermore, the process of establishing the approval status is as follows: The batch continuity is read from the batch information and used as a process penalty factor; Nuclide activity concentration data is extracted from batch regulatory data, and preset emission management limits are extracted to verify compliance with the nuclide activity concentration data, thus obtaining the release margin. Based on the consistency between the sampling and testing data and the online monitoring alignment value, as well as the release margin, the release basis for the current batch is determined to be self-consistent, and the batch certificate status is generated.

[0009] Furthermore, the process of obtaining the flow tracing identifier is as follows: The perturbation feature vector is input into a pre-trained artificial intelligence analysis model, and the reverse influence propagation is carried out along the preset process topology to obtain the influence path; Based on the source node attributes extracted from the impact path, match and output the flow traceability identifier of the main source of influence on the current batch's reliable stability.

[0010] Furthermore, the perturbation feature vector is established as follows: Extract the process features that trigger the non-preferred batch certificate status in the current batch, as well as the process feature limits; The process characteristics are arranged in the order of filtration, ion exchange, evaporation and concentration, and sampling and testing. The process features are normalized based on the process feature limits to obtain the corresponding dimensionless feature values. The dimensionless eigenvalues ​​are encapsulated into a one-dimensional array according to a fixed process sequence, which serves as the perturbation feature vector of the current batch's reliable stability affected state.

[0011] Furthermore, the process of obtaining the revised approval status is as follows: Establish a decoupling relationship between the measurement error, which includes static bias terms, dynamic hysteresis terms, and reconstructed nuclide activity concentration; The radionuclide activity concentration sequence of the source node is extracted and shifted and decayed according to the time delay response characteristics to generate the expected concentration sequence of the downstream. Under the condition of minimizing the error between the reconstructed radionuclide activity concentration sequence and the expected concentration sequence, the measurement error decoupling relationship is solved iteratively to separate the measurement latent variables. The isolated latent variables are used to compensate and correct the original data, restore the nuclide concentration in the water, and generate a corrected certification status.

[0012] Furthermore, the process of obtaining the time-delay response characteristics is as follows: Extract the operational data of the source abnormal node locked by the flow tracing identifier and the associated downstream monitoring node within the target time window, and construct the initial response matrix of the system's dynamic transmission characteristics; The initial response matrix is ​​time-series aligned and calculated to analyze the time-delay response parameters and decay response parameters between the source and downstream sequences, and then fused to generate time-delay response features.

[0013] Furthermore, the method for analyzing the physical self-consistency is as follows: Based on batch process information, forward physical analysis along the process topology is performed using batch monitoring data to generate the conserved expected state of the target node. The actual batch monitoring data is synchronously mapped to a two-dimensional state space, the deviation between the actual state and the conserved expected state is compared, and the physical self-consistency is obtained based on the degree of deviation.

[0014] Furthermore, it also includes: The release credibility level of the current batch is generated based on the revised approval status and physical consistency; and counterfactual robustness verification is performed using flow traceability identifiers to output differentiated regulatory handling strategies.

[0015] Furthermore, the process of generating differentiated regulatory and handling strategies is as follows: Establish historical safety manifolds and release characteristic parameters, project the release characteristic parameters onto the historical safety manifolds, generate safety cluster manifolds, and lock the distribution coordinate points corresponding to the release characteristic parameters; Calculate the manifold topological distance between the distributed coordinate points and the safe clustering manifold, and convert the manifold topological distance into a continuous release confidence level; Based on the flow direction tracing identifier, a directional disturbance vector is generated, and a reverse inference analysis is performed in the historical safe manifold to generate the corresponding inference confidence level; Calculate the confidence level drop rate from the release confidence level to the deduced confidence level drop, and use the absolute value of the drop rate as the decision sensitivity of the system's anti-interference bottom line; The clearance trust level and decision sensitivity are compiled into electrical modulation pulses of the underlying hardware devices, and differentiated regulatory handling strategies are generated.

[0016] An artificial intelligence-based monitoring system for radioactive wastewater treatment includes the following modules: Batch establishment module: used to acquire batch supervision data, perform batch structured processing to obtain batch information, and establish batch certificate status based on batch information; Flow tracing module: Based on the approval status and associated underlying process characteristics, disturbances are identified and disturbance feature vectors are constructed; through an artificial intelligence analysis model, the influence path of the disturbance feature vectors is inferred to obtain the flow tracing identifier; State correction module: Extract time-delay response features based on flow direction tracing identifiers; use time-delay response features to correct the measurement latent variables of the batch certificate state, and reconstruct the physical working conditions in reverse to obtain the corrected batch certificate state; Self-consistency analysis module: Based on batch process information, it performs forward physical analysis along the process topology using batch supervision data to generate a conserved expected state; and combines the modified batch certificate state to analyze the physical self-consistency. Strategy construction module: Generates release credibility level based on modified approval status and physical consistency; and uses flow traceability identifiers to perform counterfactual robustness verification, outputting differentiated regulatory handling strategies.

[0017] The beneficial effects of this invention are as follows: 1. By extracting temporal and material transfer breakpoints between adjacent stages, batch continuity is quantified. The continuous fluctuations of the process are directly incorporated into the decision-making logic, so that the release status is no longer limited to static single test values, but rather integrated into the dynamic process integrity considerations. Batch continuity is used as a process penalty factor to adjust the occupancy ratio, thereby establishing the batch certification status. After establishing the batch certification status, a perturbation feature vector is constructed and mapped to a graph attention network with a process causal topology as the static base map, deeply binding the apparent data fluctuations with the physical connections of the underlying process equipment. By extracting the influence path output of the convergence node and tracing the source identifier, it directly points to specific physical anomalies such as equipment interception failure or sensor drift, which helps improve the targeting of problem localization and analysis efficiency.

[0018] 2. For flow direction tracing markers, extracting time delay and attenuation response characteristics and performing trend decomposition on the original measurement data helps to eliminate system artifacts caused by probe radiation damage and tube wall nuclide retention under harsh radioactive conditions. Using the separated latent measurement variables to compensate and correct the original data allows for the reconstruction of water nuclide concentrations that reflect the true physical conditions, improving the reliability of release criteria under extreme interference environments.

[0019] 3. Based on the corrected batch status, forward physical calculations are performed according to the node changes in batch process information to generate a conserved expected state reflecting water flux and nuclide residue level by level. Deviation analysis of actual batch monitoring data yields physical consistency, which helps identify hidden contradictions where data from a single monitoring point is normal but the overall material flow is unbalanced, and mitigates the risk of misjudgment caused by local instrument failures. Release confidence levels are calculated. The spatial differential gradient of the confidence level drop is calculated to quantify the system's sensitivity to specific physical disturbances. The confidence level and decision sensitivity are compiled into electrical modulation pulses, facilitating the conversion of abstract high-dimensional model analysis into concrete hardware operation instructions. Attached Figure Description

[0020] The invention will now be further described with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart of a radioactive wastewater treatment and monitoring method based on artificial intelligence according to the present invention; Figure 2 This is a flowchart of the process for generating flow direction tracing identifiers in this invention; Figure 3 This is a functional module diagram of an artificial intelligence-based radioactive wastewater treatment and monitoring system according to the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Example 1: like Figure 1 and Figure 2 As shown, an artificial intelligence-based method for monitoring the treatment of radioactive wastewater includes the following steps: S10. Obtain batch monitoring data of radioactive wastewater awaiting release in the temporary storage tank, and perform batch process structured processing to obtain batch process information on the continuity of the current batch processing process; and establish the certificate status of the current batch release basis based on the batch process information. The process of obtaining batch regulatory data and acquiring batch progress information is as follows: S101. After receiving the request to release radioactive wastewater from the temporary storage tank from the control system, read the multi-source heterogeneous data and batch monitoring data of the front-end supporting links of the temporary storage tank, and perform time-series alignment processing according to the same batch number. Specifically, the process of reading multi-source heterogeneous data and batch regulatory data is as follows: High-frequency time-series data are read from the online monitoring instruments of the filtration, ion exchange, and evaporation concentration stages at the front end of the temporary storage tank, and low-frequency discrete data are read from the sampling and testing stage. Each batch of monitoring data includes fixed-length fields arranged in sequence: process identifier, batch number, radionuclide activity concentration data, cumulative treated water volume data, and collection timestamp; For the time-series data of nuclide activity concentration collected by online monitoring instruments, the Hamper filter criterion is used to remove isolated outliers; Based on the collection timestamp of the sampling and testing process, online monitoring data within adjacent time ranges in the same batch number are selected and aligned using a sliding window to obtain the online monitoring alignment value corresponding to the sampling and testing data. The width of the sliding window is pre-configured based on the online monitoring sampling cycle and the sampling and testing interval, and is recorded in the control system parameter table; S102. Establish batch process files based on batch supervision data, compare the time connection relationship between adjacent links and the cumulative water volume transfer relationship to obtain batch process information; Specifically, the process of obtaining approval information is as follows: A batch process file is established using the batch number as an index. Data from the same batch are arranged in the process sequence of filtration, ion exchange, evaporation and concentration, and sampling and testing, forming a process record sequence. For segments with continuous online records, the earliest collection timestamp in the segment is used as the segment start timestamp, and the latest collection timestamp in the segment is used as the segment end timestamp. For steps with only discrete records, the acquisition timestamp of the discrete record shall be used as the recording timestamp of that step. Read the timestamps of adjacent steps one by one. When the difference between the start timestamp or record timestamp of the subsequent step and the end timestamp of the preceding step is greater than the corresponding dynamic time base threshold, it is counted as a time breakpoint. The preferred dynamic time base threshold for the filtration to ion exchange stage is 25 minutes, the preferred dynamic time base threshold for the ion exchange to evaporation and concentration stage is 30 minutes, and the preferred dynamic time base threshold for the evaporation and concentration to sampling and analysis stage is 45 minutes. The water transfer relationship between adjacent stages is evaluated one by one: when the adjacent stage is from filtration to ion exchange or from ion exchange to evaporation and concentration, the water transfer retention rate is obtained by dividing the cumulative water volume treated in the subsequent stage by the cumulative water volume treated in the preceding stage; when the water transfer retention rate is lower than the lower limit threshold of water transfer, it is counted as a material breakpoint, and the lower limit threshold of water transfer is preferably 0.96. When the preceding stage is evaporation and concentration, the theoretical concentration ratio recorded by the evaporation and concentration equipment control system is read, and the cumulative water volume processed in the preceding stage is divided by the theoretical concentration ratio to obtain the expected subsequent water volume. When the deviation of the cumulative water volume processed in the subsequent stage from the expected subsequent water volume exceeds the concentration error threshold, it is counted as a material breakpoint. The preferred concentration error threshold is 4%. Add the number of timing breakpoints to the number of material breakpoints to get the total number of breakpoints; Divide the total number of breakpoints by the number of adjacent inspection items actually completed in this batch to obtain the normalized breakpoint penalty value; When the normalized breakpoint penalty value is greater than 1, it is counted as 1; subtract the normalized breakpoint penalty value from 1 to obtain the batch continuity falling in the interval between 0 and 1; The batch continuity, number of time breakpoints, number of material breakpoints, time difference of each breakpoint, water transfer retention rate, concentration deviation value and corresponding judgment result are packaged into batch information. The process of generating the certificate status based on the batch information is as follows: S111. Read the batch continuity as the process penalty factor, extract the preset emission management limit value to verify the compliance of the radionuclide activity concentration data, and obtain the release margin. The initial occupancy ratio of each nuclide is obtained by dividing the activity concentration data of each nuclide to be tested in the sampling and testing process by the corresponding emission management limit. The maximum initial occupancy limit among all the nuclides to be tested is used as the control occupancy limit. When the batch continuity is 0, it is directly determined that the current batch does not meet the continuity requirements for release basis, and the release margin is recorded as 0; When the batch continuity is greater than 0, calculate the ratio of the control limit ratio to the batch continuity to obtain the corrected limit ratio subject to process continuity penalty. The release margin is obtained by subtracting the corrected occupancy ratio from 1; when the calculation result is less than 0, the release margin is recorded as 0, and the batch is regarded as a batch that does not meet the release margin requirement. S112. Based on the consistency between the sampling and testing data and the online monitoring alignment value, as well as the release margin, determine the self-consistency of the release basis for the current batch and generate the batch certificate status. Specifically, the method for generating the approval status is as follows: Calculate the relative deviation between the sampled test values ​​and the noise-reduced and aligned online monitoring values ​​for the same nuclide, and take the maximum relative deviation among all the nuclides to be tested as the consistency deviation of the current batch; read the critical deviation value determined by the historical safe batches at a 95% confidence level; If the consistency deviation of the current batch is not greater than the critical deviation value, then proceed to the integrity check. When the batch process file fields are complete and there is no logical rollback of the timestamp, the batch process continuity is multiplied by the release margin to obtain the degree of self-consistency. When the degree of self-consistency is greater than the set benchmark, the approval status is set to allow release, and the set benchmark is preferably 0.80; when the degree of self-consistency is not greater than the set benchmark, the approval status is set to postpone release. If the approval process file has missing fields or timestamp logic rollback, the self-consistency level will be recorded as 0, and the approval status will be set to data incomplete and pending supplementation. If the consistency deviation of the current batch is greater than the critical deviation value, the batch certificate status will be set to sample retention and retesting, and the corresponding nuclide, deviation value, related links and current absolute timestamp will be written into the retesting work order for review. The approval status, self-consistency level, and absolute timestamp when the status was generated are all bound together as the approval status record for the current batch.

[0024] S20. Based on the approval status and associated underlying process characteristics, disturbance identification is performed, and a disturbance feature vector of the affected status is constructed. The disturbance feature vector is used to perform influence path reasoning on the disturbance feature vector through an artificial intelligence analysis model to obtain the flow source identification of the main sources of influence on credible stability. The process of identifying disturbances based on the approval status and associated underlying process features, and constructing the disturbance feature vector of the affected state, is as follows: S201. Analyze the underlying process characteristics that trigger non-optimal approval states, perform multi-dimensional perturbation identification, and construct perturbation feature vectors for states where credibility and stability are affected. Among them, non-preferred approval status includes temporary release, incomplete data pending supplementation, or sample retention for retesting; Specifically, the process of constructing the perturbation feature vector is as follows: Extract process features that trigger non-optimal batch certificate status in the current batch. The process features include the duration of the timing breakpoint in the process, the water transfer deviation value in the process, and the dual-path consistency deviation value of the nuclide. The above process characteristics are arranged in the order of filtration, ion exchange, evaporation and concentration, and sampling and testing. The dynamic time base threshold, the lower limit threshold, the concentration error threshold, and the critical deviation value are referred to as the process characteristic limits corresponding to each process characteristic. Divide the duration of the time series breakpoint by the corresponding dynamic time base threshold, divide the extracted water transfer retention rate by the corresponding transfer lower limit threshold, divide the extracted water concentration deviation value by the corresponding concentration error threshold, and divide the nuclide dual-path consistency deviation value by the critical deviation value to obtain the dimensionless feature value, that is, to achieve normalization processing. The dimensionless eigenvalues ​​are encapsulated into a one-dimensional array according to a fixed process sequence, which serves as the perturbation feature vector of the current batch’s reliability stability affected. The process of using an artificial intelligence analysis model to infer the impact path of disturbance feature vectors and obtain the source identification of the main sources of influence on credible stability is as follows: S211. Input the disturbance feature vector into the pre-trained artificial intelligence analysis model, and propagate the influence in reverse along the preset process topology to obtain the influence path; Specifically, the process of inferring the influence path is as follows: Construct a process causal topology graph consisting of nodes and directed edges. Nodes represent temporary storage tanks, filter components, ion exchange beds, evaporators, and various online monitoring instruments, while directed edges represent the direction of physical material flow or data transmission. Graph attention networks are used as the basic architecture of the artificial intelligence analysis model, and the process causal topology graph is used as the static base graph of the graph attention network. Each normalized feature value in the perturbation feature vector is mapped to the corresponding anomaly trigger node in the static base map, serving as the initial feature of the graph node; The startup graph attention network enables the abnormal triggering node to pass on the influence weights to adjacent nodes in the opposite direction of the process flow, and aggregates the features of adjacent nodes through the attention mechanism; When the aggregated impact weight of an upstream node is greater than the tracing threshold, the upstream node is determined to be a convergence node, and the connected subgraph from the abnormal trigger node to the convergence node is extracted as the impact path. The source tracing threshold is the lower boundary of the aggregation weight corresponding to the true root cause node determined statistically in the graph attention network by pre-extracting historical real anomaly source tracing samples. The preferred source tracing threshold is 0.75. S212. Based on the source node attributes in the extracted impact path, match and output the flow traceability identifier of the main source of influence on the current batch's reliable stability. Specifically, the process of matching and outputting the flow tracing identifier is as follows: The most upstream convergence node in the influence path is identified as the root cause node; Read the pre-configured attribution mapping table, which defines the correspondence between various types of nodes and flow tracing identifiers; Compare the attribute information of the root cause node with the attribution mapping table: If the root cause node is a material handling device (such as a filter component or ion exchange bed), output the flow direction traceability identifier of "equipment retention failure or material leakage". If the root cause node is an online monitoring instrument, and only a single instrument node is abnormal, then output a flow tracing identifier of "sensor drift or probe contamination"; If the root cause node is a temporary storage tank or a source control valve, then output a flow tracing identifier for "sudden change in influent water quality"; If the attribute information of the root cause node does not belong to the above mapping relationship, the flow tracing identifier of "undefined exception class" will be output, and the manual verification mechanism will be triggered.

[0025] Example 2: Please see Figure 1 As shown, an artificial intelligence-based method for monitoring the treatment of radioactive wastewater includes the following steps: S30. Extract the time-delay response characteristics of disturbance information between process nodes based on flow direction traceability identifiers; use the time-delay response characteristics to correct the measurement hidden variables caused by probe radiation damage and tube wall nuclide retention in the batch certification status, and reverse reconstruct the physical conditions to obtain the corrected batch certification status. The process of extracting the time-delay response characteristics of disturbance information between process nodes based on flow direction tracing identifiers is as follows: S301. Extract the operational data of the source abnormal node locked by the flow tracing identifier and the associated downstream monitoring node within the target time window, and construct the initial response matrix of the system's dynamic transmission characteristics. Specifically, the process of constructing the initial response matrix is ​​as follows: Simultaneously extract the time series of radionuclide activity concentrations of the source abnormal node and the downstream monitoring node, as well as the corresponding cumulative treated water volume time series; According to the collection time sequence, the time, nuclide activity concentration and cumulative treated water volume at the same time or adjacent times after alignment are arranged into matrix rows to generate an initial response matrix containing time column, concentration column and water volume column; S302. Perform time series alignment calculation on the initial response matrix, analyze the time delay response parameters and decay response parameters between the source and downstream sequences, and fuse them to generate time delay response features. Specifically, the process of fusing and generating time-delay response features is as follows: The dynamic time warping algorithm is used to calculate the shortest warped path between the time series of nuclide activity concentrations at the source node and the time series of nuclide activity concentrations at the downstream node in the initial response matrix; Extract the average time offset between the timestamps of the source and downstream sequences aligned along the shortest regular path, and use it as the time delay response parameter; The normalized inner product is calculated for the source and downstream sequences aligned along the shortest regular path, and the result of the normalized inner product is used as the decay response parameter. By splicing the time delay response parameters and the decay response parameters in a fixed order, the time delay response characteristics between process nodes are formed. The process of correcting the latent measurement variables caused by probe radiation damage and tube wall radionuclide retention in the certification state using time-delay response characteristics, and then reconstructing the physical conditions in reverse to obtain the corrected certification state, is as follows: S311. Establish a decoupling relationship between the measurement error, which includes static bias term, dynamic hysteresis term and reconstructed nuclide activity concentration; Specifically, the process of establishing the measurement error decoupling relationship is as follows: Obtain the original radionuclide activity concentration sequence of downstream monitoring nodes within the target time window; Trend decomposition was performed on the original radionuclide activity concentration sequence, and the slowly changing and persistent low-frequency components were extracted as static bias terms to reflect the measurement drift caused by probe radiation damage. The tailing component that continues to decay after the concentration peak is extracted as a dynamic hysteresis term to characterize the measurement tailing caused by the retention of radionuclides in the tube wall. The original nuclide activity concentration sequence is represented as the superposition relationship of reconstructed nuclide activity concentration, static bias term and dynamic hysteresis term, and the measurement error decoupling relationship is obtained; S312. Using the acquired time-delay response characteristics as correction constraints, the measurement error decoupling relationship is solved iteratively to separate the measurement latent variables. Specifically, the iterative solution process is as follows: The radionuclide activity concentration sequence of the source abnormal node is shifted on the time axis according to the time delay response parameter, and the amplitude of the shifted sequence is multiplied by the decay response parameter to calculate and generate the expected concentration sequence of the downstream monitoring node. Under the condition of minimizing the error between the reconstructed nuclide activity concentration sequence and the expected concentration sequence, the amplitudes of the static bias term and the dynamic hysteresis term are iteratively adjusted. After the iteration stabilizes, the static bias term is taken as the latent measurement variable caused by probe radiation damage, and the dynamic hysteresis term is taken as the latent measurement variable caused by tube wall nuclide retention. S313. The separated latent variables of measurement are used to compensate and correct the original data, restore the concentration of nuclides in the water body to reconstruct the physical conditions, and generate the corrected certification status. Specifically, the process of generating a revised approval status is as follows: In the original batch monitoring data of nuclide activity concentration sequence, static bias term and dynamic lag term are simultaneously subtracted to obtain the reconstructed nuclide activity concentration; Substitute the reconstructed nuclide activity concentration into the original release criteria, recalculate the revised release margin and self-consistency, write the updated verification results into the batch process file, and output the revised batch certificate status after removing measurement system artifacts.

[0026] S40. Based on batch process information, perform forward physical analysis along the process topology using batch monitoring data to generate the conserved expected state of the target node; combine the corrected batch certificate state to analyze the physical consistency between the actual data and the conserved expected state. The method for generating the conserved desired state of the target node by performing forward physical analysis along the process topology based on batch process information and batch monitoring data is as follows: S401. Based on batch process information, perform forward physical analysis along the process topology using batch monitoring data to generate the conserved expected state of the target node. Specifically, the process of generating the conserved expected state of the target node is as follows: A two-dimensional state space is established, consisting of cumulative treated water volume and total nuclide activity. The cumulative treated water volume is used to reflect the physical flux of the water body, and the total nuclide activity is used to reflect the remaining amount of nuclides after they migrate with the water body and are intercepted by the process. The cumulative treated water volume and total nuclide activity at the source node are used as the initial state; Based on the water volume change relationship and nuclide retention relationship corresponding to each process node in the batch information, the expected cumulative water volume and expected total nuclide activity of subsequent nodes are calculated step by step along the filtration, ion exchange, evaporation and concentration and sampling and testing stages. The expected cumulative water volume and expected total activity of nuclides calculated at each process node are connected in the process sequence to form the conserved expected state of the target node. S402. Synchronously map the actual batch monitoring data to a two-dimensional state space, compare the degree of deviation between the actual state and the conserved expected state, and obtain the physical self-consistency degree based on the degree of deviation. Specifically, the process of analyzing the physical consistency between actual data and the conserved expected state is as follows: Extract the cumulative treated water volume and total nuclide activity measured at each node from the actual batch monitoring data, and construct an actual state sequence according to the same time sequence; Calculate the deviation between the actual cumulative treated water volume and the expected cumulative treated water volume, and the deviation between the actual total activity of nuclides and the expected total activity of nuclides, respectively. The above deviations are normalized according to the corresponding release limits or process allowable deviations, and the normalized deviations are combined into a state deviation value. The physical self-consistency is obtained by subtracting the state deviation value from 1; when the state deviation value is greater than 1, it is counted as 1, so that the physical self-consistency falls within the range of 0 to 1; the higher the physical self-consistency, the more consistent the actual batch regulatory data is with the conservation expected state.

[0027] Example 3: Please see Figure 1 As shown, an artificial intelligence-based method for monitoring the treatment of radioactive wastewater also includes the following steps: S50. Generate the release credibility level of the current batch based on the modified approval status and physical consistency; and use the flow traceability identifier to perform counterfactual robustness verification and output differentiated regulatory disposal strategies. The process of generating the release confidence level for the current batch based on the modified approval status and physical consistency is as follows: S501. Establish historical safety manifolds and release characteristic parameters, project the release characteristic parameters onto the historical safety manifolds, generate safety clustering manifolds, and lock the distribution coordinate points corresponding to the release characteristic parameters. Specifically: The corrected self-consistency, corrected release margin, and physical self-consistency are read from the batch supervision data and combined to define the release characteristic parameters; A three-dimensional coordinate system is established using the corrected self-consistency, the corrected release margin, and the physical self-consistency to construct a historical safety manifold. Batch data that have been confirmed to be safely released and have no abnormal traceability under the same historical process conditions are projected into the historical safety manifold and linked to generate a safety clustering manifold. The corrected self-consistency, corrected release margin, and physical self-consistency in the current batch of regulatory data are used as three-dimensional coordinate parameters and simultaneously projected onto the historical safety manifold to lock the distribution coordinate points corresponding to the three-dimensional coordinate parameters. S502. Calculate the manifold topological distance between the distributed coordinate points and the safe clustering manifold, and convert the manifold topological distance into a continuous release confidence level; Specifically, the method for converting to a continuous release trust level is as follows: Calculate the shortest geodesic distance from the distributed coordinate points to the geometric center of the safe clustering manifold, and define the shortest geodesic distance as the manifold topological distance; Extract the pre-calibrated manifold radius constant; use the manifold radius constant as the numerator of the dividend; Calculate the sum of the manifold radius constant and the manifold topological distance, and use it as the divisor and denominator; Perform scalar division on the numerator of the dividend and the denominator of the divisor. The output of the operation is a continuous value that approaches the interval between zero and one. The continuous value is defined as the release confidence level. S503. Generate a directional disturbance vector based on the flow direction tracing identifier, perform reverse inference analysis in the historical safe manifold, and generate the corresponding inference confidence level; Specifically, the process of performing reverse inference analysis is as follows: The system analyzes the physical anomaly dimension indicated by the flow direction tracing marker and extracts the preset inference step size as the modulus of the disturbance feature. When the flow direction tracing marker points to probe contamination, a first disturbance vector is generated along the corrected self-consistency coordinate axis with the inference step size. When the flow direction tracing marker points to equipment interception failure, a second disturbance vector is generated along the physical self-consistency coordinate axis with the inference step size. When the flow direction tracing marker points to a sudden change in influent water quality, a third disturbance vector is generated along the corrected release margin coordinate axis with the inference step size. Using the distribution coordinate points as reference points, the corresponding first disturbance vector, second disturbance vector, or third disturbance vector is superimposed to generate inverse derivation coordinates; The shortest geodesic distance from the inverse deduction coordinates to the geometric center of the safe clustering manifold is recalculated as the deduction topological distance. The manifold radius constant is used as the numerator of the dividend. The sum of the manifold radius constant and the deduction topological distance is used as the denominator of the divisor. The scalar division operation is performed again to generate the corresponding deduction confidence level. S504. Calculate the confidence level drop rate from the release confidence level to the deduced confidence level, and use the absolute value of the drop rate as the decision sensitivity of the system's anti-interference bottom line. Specifically, the process of calculating the confidence level drop rate is as follows: Extract the absolute difference between the release confidence level and the inferred confidence level; calculate the Euclidean norm of the corresponding first perturbation vector, second perturbation vector, or third perturbation vector; Divide the absolute difference of the values ​​by the Euclidean norm to obtain the confidence level drop rate in that direction of inference; the absolute value of the confidence level drop rate is directly defined as the decision sensitivity of the system's anti-interference bottom line. S505: Compile the release trust level and decision sensitivity into electrical modulation pulses of the underlying hardware device, and generate differentiated regulatory handling strategies; Specifically, the process of generating differentiated regulatory and handling strategies is as follows: Extract the upper limit of the rated duty cycle of the bottom control valve; perform a scalar multiplication operation between the release confidence level and the upper limit of the rated duty cycle, and directly configure the product value output by the scalar multiplication operation as the target pulse duty cycle of the bottom control valve; use the target pulse duty cycle to drive the physical connection state of the release pipeline, and the target pulse duty cycle maps the duration ratio of the valve's normally open phase. Extract the rated sampling reference of the online sampling pump; perform a scalar multiplication operation between the decision sensitivity and the rated sampling reference, and directly configure the product value output by the scalar multiplication operation as the trigger pulse frequency of the online sampling pump; The execution density of bypass sampling retest is controlled by the trigger pulse frequency, and the time interval of the sampling action is mapped by the trigger pulse frequency. The target pulse duty cycle and trigger pulse frequency are synchronously encapsulated into a low-level execution message, which is compatible with the communication protocol of industrial programmable logic controllers. The underlying execution message is sent to the on-site hardware control cabinet as a differentiated regulatory and handling strategy, and the underlying execution message and absolute timestamp are simultaneously written into the batch regulatory and handling record.

[0028] Example 4: Please see Figure 3 As shown, an artificial intelligence-based monitoring system for radioactive wastewater treatment includes the following modules: Batch Establishment Module: Used to acquire batch monitoring data of radioactive wastewater awaiting release in the temporary storage tank, perform batch process structured processing to obtain batch process information on the continuity of the current batch processing process; and establish the certificate status of the current batch release based on the degree of self-consistency based on the batch process information. Flow traceability module: Based on the approval status and associated underlying process characteristics, disturbance identification is performed, and disturbance feature vectors of affected states are constructed; through artificial intelligence analysis models, the influence path inference of disturbance feature vectors is performed to obtain the flow traceability identifiers of the main sources of influence on credible stability; State correction module: Extracts the time-delay response characteristics of disturbance information between process nodes based on flow direction traceability identifier; uses the time-delay response characteristics to correct the measurement latent variables caused by probe radiation damage and tube wall nuclide retention in the batch certification state, and reversely reconstructs the physical conditions to obtain the corrected batch certification state. Self-consistency analysis module: Based on batch process information, it performs forward physical analysis along the process topology using batch monitoring data to generate the conserved expected state of the target node; combined with the corrected batch certificate state, it analyzes the physical self-consistency between the actual data and the conserved expected state. Strategy construction module: Generates the release credibility level of the current batch based on the modified approval status and physical consistency; and uses flow traceability identifiers to perform counterfactual robustness verification, and outputs differentiated regulatory handling strategies.

[0029] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the present invention should still fall within the scope of the present invention.

Claims

1. A method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence, characterized in that, Includes the following steps: Acquire batch monitoring data of radioactive wastewater awaiting release in the temporary storage tank, and perform batch process structured processing to obtain batch process information on the continuity of the current batch processing; and establish the certificate status of the current batch release basis based on the batch process information to determine the degree of self-consistency. Based on the approval status and associated underlying process characteristics, disturbances are identified, and disturbance feature vectors of affected states are constructed. Through artificial intelligence analysis models, the influence path of disturbance feature vectors is inferred to obtain the flow source identification of the main sources of influence on credible stability. Based on the flow traceability identifier, the time delay response characteristics of disturbance information between process nodes are extracted; the time delay response characteristics are used to correct the measurement hidden variables caused by probe radiation damage and tube wall nuclide retention in the batch certification status, and the physical conditions are reconstructed in reverse to obtain the corrected batch certification status. Based on batch process information, forward physical analysis along the process topology is performed using batch monitoring data to generate the conserved expected state of the target node; combined with the corrected batch certificate state, the physical consistency between the actual data and the conserved expected state is analyzed.

2. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 1, characterized in that: The process of establishing the aforementioned approval status is as follows: The batch continuity is read from the batch information and used as a process penalty factor; Nuclide activity concentration data is extracted from batch regulatory data, and preset emission management limits are extracted to verify compliance with the nuclide activity concentration data, thus obtaining the release margin. Based on the consistency between the sampling and testing data and the online monitoring alignment value, as well as the release margin, the release basis for the current batch is determined to be self-consistent, and the batch certificate status is generated.

3. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 1, characterized in that: The process of obtaining the flow traceability identifier is as follows: The perturbation feature vector is input into a pre-trained artificial intelligence analysis model, and the reverse influence propagation is carried out along the preset process topology to obtain the influence path; Based on the source node attributes extracted from the impact path, match and output the flow traceability identifier of the main source of influence on the current batch's reliable stability.

4. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 3, characterized in that: The perturbation feature vector is established as follows: Extract the process features that trigger the non-preferred batch certificate status in the current batch, as well as the process feature limits; The process characteristics are arranged in the order of filtration, ion exchange, evaporation and concentration, and sampling and testing. The process features are normalized based on the process feature limits to obtain the corresponding dimensionless feature values. The dimensionless eigenvalues ​​are encapsulated into a one-dimensional array according to a fixed process sequence, which serves as the perturbation feature vector of the current batch's reliable stability affected state.

5. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 1, characterized in that: The process of obtaining the revised approval status is as follows: Establish a decoupling relationship between the measurement error, which includes static bias terms, dynamic hysteresis terms, and reconstructed nuclide activity concentration; The radionuclide activity concentration sequence of the source node is extracted and shifted and decayed according to the time delay response characteristics to generate the expected concentration sequence of the downstream. Under the condition of minimizing the error between the reconstructed radionuclide activity concentration sequence and the expected concentration sequence, the measurement error decoupling relationship is solved iteratively to separate the measurement latent variables. The isolated latent variables are used to compensate and correct the original data, restore the nuclide concentration in the water, and generate a corrected certification status.

6. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 5, characterized in that: The process of obtaining the time-delay response characteristics is as follows: Extract the operational data of the source abnormal node locked by the flow tracing identifier and the associated downstream monitoring node within the target time window, and construct the initial response matrix of the system's dynamic transmission characteristics; The initial response matrix is ​​time-series aligned and calculated to analyze the time-delay response parameters and decay response parameters between the source and downstream sequences, and then fused to generate time-delay response features.

7. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 1, characterized in that: The method for analyzing the physical self-consistency is as follows: Based on batch process information, forward physical analysis along the process topology is performed using batch monitoring data to generate the conserved expected state of the target node. The actual batch monitoring data is synchronously mapped to a two-dimensional state space, the deviation between the actual state and the conserved expected state is compared, and the physical self-consistency is obtained based on the degree of deviation.

8. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 1, characterized in that: Also includes: The release confidence level of the current batch is generated based on the modified approval status and physical consistency. It also utilizes flow traceability identifiers to conduct counterfactual robustness verification and outputs differentiated regulatory and disposal strategies.

9. The method for monitoring and regulating radioactive wastewater treatment based on artificial intelligence according to claim 8, characterized in that: The process of generating differentiated regulatory and handling strategies is as follows: Establish historical safety manifolds and release characteristic parameters, project the release characteristic parameters onto the historical safety manifolds, generate safety cluster manifolds, and lock the distribution coordinate points corresponding to the release characteristic parameters; Calculate the manifold topological distance between the distributed coordinate points and the safe clustering manifold, and convert the manifold topological distance into a continuous release confidence level; Based on the flow direction tracing identifier, a directional disturbance vector is generated, and a reverse inference analysis is performed in the historical safe manifold to generate the corresponding inference confidence level; Calculate the confidence level drop rate from the release confidence level to the deduced confidence level drop, and use the absolute value of the drop rate as the decision sensitivity of the system's anti-interference bottom line; The clearance trust level and decision sensitivity are compiled into electrical modulation pulses of the underlying hardware devices, and differentiated regulatory handling strategies are generated.

10. An artificial intelligence-based radioactive wastewater treatment monitoring system, used to implement the artificial intelligence-based radioactive wastewater treatment monitoring method according to any one of claims 1-9, characterized in that, Includes the following modules: Batch establishment module: used to acquire batch supervision data, perform batch structured processing to obtain batch information, and establish batch certificate status based on batch information; Flow tracing module: Based on the approval status and associated underlying process characteristics, disturbances are identified and disturbance feature vectors are constructed; through an artificial intelligence analysis model, the influence path of the disturbance feature vectors is inferred to obtain the flow tracing identifier; State correction module: Extract time-delay response features based on flow direction tracing identifiers; use time-delay response features to correct the measurement latent variables of the batch certificate state, and reconstruct the physical working conditions in reverse to obtain the corrected batch certificate state; Self-consistency analysis module: Based on batch process information, it performs forward physical analysis along the process topology using batch supervision data to generate a conserved expected state; and combines the modified batch certificate state to analyze the physical self-consistency. Strategy construction module: Generates release credibility level based on modified approval status and physical consistency; It also utilizes flow traceability identifiers to conduct counterfactual robustness verification and outputs differentiated regulatory and disposal strategies.

Images