Intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion

Abstract

The application is particularly a garbage leachate zero discharge intelligent supervision method based on multi-parameter fusion, relates to the garbage leachate treatment supervision technical field, and comprises the following steps: determining the time delay propagation window of each adjacent section pair based on the hydraulic connection topology between each adjacent process section of the process chain and the nominal hydraulic residence time of each section, calculating the actual propagation delay value and the attenuation coefficient of each water quality parameter between the adjacent process sections through cross-correlation analysis, and generating a process chain cascade propagation parameter matrix.In the application, the cross-correlation analysis extracts the actual propagation delay value and the attenuation coefficient of each parameter between the adjacent sections from the historical multi-parameter time series of each process section, generates a process chain cascade propagation parameter matrix, and expands the action range of multi-parameter fusion from the end single node to the time delay correlation between all sections of the whole chain.

Description

Technical Field

[0001] This invention relates to the field of landfill leachate treatment and monitoring technology, and in particular to an intelligent monitoring method for zero-discharge of landfill leachate based on multi-parameter fusion. Background Technology

[0002] Zero-discharge treatment processes for landfill leachate typically involve multiple stages in series, including conditioning and homogenization, two-stage AO biological treatment, ultrafiltration, nanofiltration, reverse osmosis, and evaporation crystallization. Significant differences in hydraulic retention time and buffering effects exist between these stages; water quality anomalies from upstream stages can take hours or even days to propagate to the final effluent outlet.

[0003] Current monitoring methods rely on real-time comparison of multiple parameters at the end-of-line effluent point, which can only detect abnormal states that have reached the end point and cannot detect the cascading propagation of abnormalities along the process chain. When multiple parameters at the end point exceed the standard and trigger an alarm, the abnormal water volume has often already entered the membrane system in large quantities, causing increased membrane fouling or a surge in concentrate volume. These problems result in a severe lag in regulatory intervention, making it impossible to provide early warning and source intervention before abnormal water volume enters subsequent membrane systems. Summary of the Invention

[0004] The purpose of this invention is to propose an intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion in order to solve the above-mentioned problems.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A smart monitoring method for zero-discharge landfill leachate based on multi-parameter fusion includes: Acquire real-time data reported by online sensors in each process segment of the leachate zero-discharge treatment process chain, align them by process segment number and timestamp, and generate a multi-process segment synchronous multi-parameter time series dataset; Based on the hydraulic connection topology between adjacent process segments in the process chain and the nominal hydraulic residence time of each segment, the time delay propagation window of each adjacent segment pair is determined. Through cross-correlation analysis, the actual propagation delay value and attenuation coefficient of each water quality parameter between adjacent process segments are calculated, and the process chain cascade propagation parameter matrix is ​​generated. The multi-process segment synchronous multi-parameter time series dataset is input into a pre-trained cascaded state propagation prediction network, which outputs the multi-parameter prediction sequence of water effluent from each process segment within a future preset time window; wherein, the attention layer of the cascaded state propagation prediction network introduces the process chain cascaded propagation parameter matrix as a structured prior constraint. The predicted sequence of multiple parameters of effluent from each process section is compared with the process protection threshold and the terminal zero discharge compliance threshold parameter by parameter. When the predicted value of the parameter of the upstream process section exceeds the process protection threshold, the expected exceedance magnitude and expected arrival time of the abnormality propagation to the terminal effluent node are calculated based on the process chain cascade propagation parameter matrix, and a graded regulatory early warning signal is generated. Based on the aforementioned graded monitoring and early warning signals, matching items are retrieved from the process control strategy library to generate parameter adjustment instructions for the process segment at the source of the anomaly.

[0006] Preferably, determining the time-delay propagation window for each adjacent segment pair includes: For each pair of adjacent process segments, the nominal hydraulic residence time of the downstream process segment is obtained as the reference delay. The time delay propagation window of the adjacent segment pair is determined with the reference delay as the center and the preset multiplier range as the boundary. For each pair of adjacent process sections and each type of water quality parameter, extract the historical time series of the same parameter of the effluent from the upstream process section and the influent from the downstream process section, calculate the cross-correlation function of the two series within the time delay propagation window, and take the time offset corresponding to the peak value of the cross-correlation function as the actual propagation delay value of the parameter. After aligning the upstream effluent sequence and the downstream influent sequence with the actual propagation delay value, the mean amplitude ratio between the downstream influent sequence and the upstream effluent sequence is calculated, and the mean amplitude ratio is used as the attenuation coefficient of this parameter.

[0007] Preferably, the cascaded state propagation prediction network adopts an encoder-decoder structure, wherein the encoder encodes the multi-parameter sequences of each process segment into process segment state vectors; the attention layer introduces the process chain cascaded propagation parameter matrix as a structured prior constraint, including: The state vectors of each process segment are used as query and key-value pairs in attention calculation. For each process segment... and process section Attention scores between The adjusted attention score is: ; in, For the process segment in the process chain cascade propagation parameter matrix To process section The attenuation coefficient corresponding to the parameter, , For process section To process section The actual propagation delay value of the corresponding parameter, It represents the maximum value of the actual propagation delay for all adjacent segments in the process chain. This is the mean of all raw attention scores in the current batch; After attention-weighted aggregation, the state vectors of each process segment are fused into a full-chain state representation. The decoder outputs the multi-parameter prediction sequence of the effluent from each process segment based on the full-chain state representation.

[0008] Preferably, the process segment and process section When there is no direct hydraulic connection between them, Take zero, Take the maximum value This leads to the adjusted attention score. .

[0009] Preferably, each temporal coding branch in the encoder also outputs intermediate layer hidden state features; before the attention layer, the intermediate layer hidden state features of each process segment are concatenated according to the adjacent segment pairs in the process chain cascade propagation parameter matrix, and after mapping by the fully connected layer, inter-segment interaction features are generated. The inter-segment interaction features are concatenated with the corresponding process segment state vector and then input into the attention layer.

[0010] Preferably, the step of calculating the expected exceedance magnitude and expected arrival time of the anomaly propagation to the final effluent node based on the process chain cascade propagation parameter matrix includes: Extract the actual propagation delay value and attenuation coefficient of each adjacent segment pair between the abnormal triggering process segment and the terminal water outlet node under the corresponding parameter category from the process chain cascade propagation parameter matrix; The actual propagation delay values ​​of each adjacent segment are summed along the propagation path to obtain the estimated arrival time; The difference between the predicted parameter value of the abnormal trigger point and the process protection threshold is multiplied by the attenuation coefficient of each adjacent segment on the propagation path to obtain the expected residual abnormal amount. The expected residual abnormal amount is then compared with the end-of-pipe zero emission compliance threshold to obtain the expected exceedance range.

[0011] Preferably, the generation of tiered regulatory early warning signals includes: When the expected exceedance is greater than the first amplitude threshold and the expected arrival time is less than the first time threshold, the highest level warning signal is generated; when the expected exceedance is greater than the second amplitude threshold and the expected arrival time is less than the second time threshold, an intermediate level warning signal is generated; and for other cases exceeding the process protection threshold, the lowest level warning signal is generated; wherein, the first amplitude threshold is greater than the second amplitude threshold and the first time threshold is less than the second time threshold.

[0012] Preferably, before the parameter-by-parameter comparison, a sliding window smoothing process is applied to the multi-parameter prediction sequence of the effluent from each process section, and the length of the sliding window is taken as a preset proportion of the nominal hydraulic residence time of the corresponding process section.

[0013] Preferably, the process chain cascade propagation parameter matrix is ​​updated on a rolling basis according to a preset period. Each time it is updated, the actual propagation delay value and attenuation coefficient of each adjacent segment pair are recalculated using historical data of the most recent preset duration, and the old process chain cascade propagation parameter matrix is ​​replaced.

[0014] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention extracts the actual propagation delay value and attenuation coefficient of each parameter between adjacent segments from the historical multi-parameter time series of each process segment through cross-correlation analysis, and generates a process chain cascade propagation parameter matrix, thereby expanding the scope of multi-parameter fusion from the end single node to the time delay correlation between all segments of the entire chain.

[0015] 2. This invention, through tiered monitoring and early warning signals and a supporting process control strategy library, enables the monitoring platform to push early warning signals and issue parameter adjustment instructions before abnormal water volume enters the subsequent membrane system, achieving proactive compliance control across the entire zero-discharge chain. The process chain cascade propagation parameter matrix is ​​updated on a rolling basis according to a preset period, allowing the prediction results to adapt to the drift of operating conditions due to seasonal changes or changes in influent water quality, avoiding prediction deviations caused by operating condition drift. Attached Figure Description

[0016] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0018] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0019] Example 1 Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0020] In this embodiment, it includes: Zero-discharge treatment processes for landfill leachate typically involve multiple stages in series, including conditioning and homogenization, two-stage AO biological treatment, ultrafiltration, nanofiltration, reverse osmosis, and evaporation crystallization. Significant differences in hydraulic retention time and buffering effects exist between these stages, meaning that water quality anomalies from upstream stages can take hours or even days to propagate to the final effluent outlet. Current monitoring methods, based on real-time comparison of multiple parameters at the final effluent outlet, can only detect anomalies that have reached the end point and cannot detect the cascading propagation of anomalies along the process chain. When multiple parameters at the end point exceed limits and trigger alarms, the abnormal water volume has often already entered the membrane system in large quantities, causing increased membrane fouling or a surge in concentrate volume, resulting in a severe delay in regulatory intervention.

[0021] This embodiment provides an intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion. By extracting the propagation delay value and attenuation coefficient of each parameter between adjacent segments of the process chain, and using a cascaded state propagation prediction network with fused time delay correlation, the method can predict the impact of anomalies on the final effluent when the anomaly is still in the upstream process segment, thereby achieving upstream early warning and source intervention.

[0022] In this embodiment, each process segment of the leachate zero-discharge treatment process chain is equipped with online sensors, and each sensor is connected to the data processing server of the monitoring platform through a data acquisition gateway. The monitoring platform is configured with a pre-trained cascaded state propagation prediction network and a pre-stored process control strategy library.

[0023] A smart monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to an embodiment of the present invention includes the following steps: Step 1: Obtain the synchronous multi-parameter time series dataset for each process segment. The system acquires real-time data from online sensors deployed in each process segment of the leachate zero-discharge treatment process chain, aligns the data by process segment number and timestamp, and generates a multi-process segment synchronous multi-parameter time series dataset.

[0024] The reported data for each process stage includes COD, ammonia nitrogen, total nitrogen, conductivity, and pH values ​​of the influent and effluent, as well as operating parameters such as hydraulic retention time, sludge concentration, membrane flux, and operating pressure. Since these parameters have different dimensions, Z-score standardization is applied to each parameter sequence before inputting them into the cascaded state propagation prediction network to eliminate the impact of these dimensional differences on network training.

[0025] It should be noted that the above alignment by process segment number and timestamp refers to mapping the sensor-reported data from each process segment to the same time reference axis. The sampling periods of each sensor may differ; the alignment process uses a linear interpolation algorithm to unify the parameter sequences to a preset minimum sampling interval. For example, if the COD sensor in the biochemistry segment samples every 5 minutes, while the membrane flux sensor in the ultrafiltration segment samples every 1 minute, the COD sequence is extended to one data point per minute through linear interpolation, ensuring that the parameter sequences of each process segment correspond point-by-point in the time dimension.

[0026] Step 2: Generate the process chain cascade propagation parameter matrix Based on the actual hydraulic connection topology between adjacent process segments in the process chain and the nominal hydraulic residence time of each segment, the time-delay propagation window between adjacent segments is calculated. Historical time series of multiple parameters of the influent and effluent of each process segment are extracted, and the actual propagation delay and attenuation coefficient of each water quality parameter between adjacent process segments are calculated through cross-correlation analysis. The actual propagation delay and attenuation coefficient are organized according to process segment pairs and parameter categories to generate a process chain cascade propagation parameter matrix.

[0027] Step 2 may specifically include: Step 201: Determine the time-delay propagation window. For each pair of adjacent process segments in the process chain, obtain the nominal hydraulic residence time of the downstream process segment as the reference delay. Using the reference delay as the center and a preset multiplier range as the boundary, determine the time-delay propagation window of the adjacent segment pair. For example, when the nominal hydraulic residence time of a certain process segment is 6 hours and the preset multiplier range is 0.5 to 2.0, the time-delay propagation window of the adjacent segment pair is 3 hours to 12 hours.

[0028] Step 202: Calculate the actual propagation delay value. For each pair of adjacent process sections and each type of water quality parameter, extract the historical time series of the same parameter from the effluent of the upstream process section and the influent of the downstream process section. Calculate the cross-correlation function of the two series within the time-delay propagation window, and take the time offset corresponding to the peak value of the cross-correlation function as the actual propagation delay value of the parameter between the adjacent process sections.

[0029] Step 203: Calculate the attenuation coefficient. For each pair of adjacent process sections and each type of water quality parameter, after aligning the upstream effluent sequence and downstream influent sequence based on the actual propagation delay value, calculate the average amplitude ratio between the downstream influent sequence and the upstream effluent sequence, and use the average amplitude ratio as the attenuation coefficient of the parameter between the adjacent process sections.

[0030] Step 204: Organize the process chain cascade propagation parameter matrix. Organize the actual propagation delay values ​​and attenuation coefficients corresponding to each parameter of each adjacent segment pair, using row indices for process segment pairs and column indices for parameter categories, to generate the process chain cascade propagation parameter matrix.

[0031] It should be noted that the length of the historical time series used in the above cross-correlation analysis is no less than a preset multiple of the maximum nominal hydraulic residence time in each adjacent segment pair, to ensure the statistical significance of the cross-correlation function. For example, when the maximum nominal hydraulic residence time is 24 hours and the preset multiple is 30, the length of the historical time series is no less than 720 hours.

[0032] In this embodiment, to adapt to the drift of process chain operating conditions due to seasonal changes or influent water quality variations, the process chain cascade propagation parameter matrix is ​​updated on a rolling basis according to a preset period. Each update re-executes steps 201 to 204 using historical data of the most recent preset duration, replacing the old process chain cascade propagation parameter matrix. This rolling update ensures that the process chain cascade propagation parameter matrix reflects the actual propagation characteristics of each parameter under the current operating conditions, avoiding prediction deviations caused by operating condition drift.

[0033] Step 3: Predict the multi-parameter sequence of effluent from each process stage. The multi-process section synchronous multi-parameter time series dataset is input into the pre-trained cascaded state propagation prediction network, which outputs the multi-parameter prediction sequence of the effluent from each process section within a preset time window.

[0034] The cascaded state propagation prediction network employs a multi-input multi-output encoder and decoder structure. The encoder encodes the multi-parameter sequences of each process segment into process segment state vectors. The state vectors of each process segment are weighted and correlated through an attention layer according to the time delay relationship in the cascaded propagation parameter matrix of the process chain, and fused to form a full-chain state representation. Based on the full-chain state representation, the decoder outputs the predicted multi-parameter effluent sequences of each process segment within a preset future time window.

[0035] Step 3 may specifically include: Step 301: Encode the process segment state vector. Input the multi-parameter time series of each process segment into the corresponding time-series coding branch in the encoder. Each time-series coding branch uses a recurrent neural network with the same structure to encode the input multi-parameter time series into a fixed-dimensional process segment state vector. The dimensions of the state vectors for each process segment are the same.

[0036] Step 302: Integrate the full-chain state representation. Input the state vectors of each process segment into the attention layer. In the attention layer, the state vectors of each process segment participate in attention calculation as query and key-value pairs. The calculation of attention weights introduces the process chain cascade propagation parameter matrix as a structured prior constraint: for the attention weight between any two process segment state vectors, the actual propagation delay value and attenuation coefficient corresponding to the process chain cascade propagation parameter matrix are used as bias terms and superimposed on the attention score. After attention weighted aggregation, the state vectors of each process segment are fused into the full-chain state representation.

[0037] It should be noted that the way the process chain cascade propagation parameter matrix is ​​introduced as a bias term in the above attention weights is as follows: for each process segment and process section Attention scores between The attenuation coefficients corresponding to the propagation parameter matrix of the process chain cascade are used. As a scaling factor, the actual propagation delay value After normalization and applied as an additive bias, the adjusted attention score is: ; in, For process section For process section Adjusted attention score To query the side process section number, This refers to the numbering of the key-side process section. For the process segment in the process chain cascade propagation parameter matrix To process section The attenuation coefficient corresponding to the parameter, The original attention score before adjustment. For the actual propagation delay value The normalized mapping will Mapping to The numerical range of the same dimension, specifically, ,in For the process segment in the process chain cascade propagation parameter matrix To process section The actual propagation delay value of the corresponding parameter, It represents the maximum value of the actual propagation delay for all adjacent segments in the process chain. For all raw attention scores in the current batch The mean.

[0038] It should be noted that when the process section and process section When there is no direct hydraulic connection between them, Take zero, the corresponding adjusted attention score Degenerate into only retain The additive bias term, i.e. ;at this time In Take the maximum value of the actual propagation delay for all adjacent segments in the process chain. , making Take the upper limit of normalization, 1, and multiply by The additive bias term takes the average of the original attention scores for the current batch to ensure that the attention scores between segments without direct hydraulic connections have definite values ​​rather than zero, thus preventing the complete collapse of the association weights of the attention layer for unconnected segment pairs. For adjacent segment pairs with direct hydraulic connections, This reflects the amplitude retention ratio of upstream anomalous signals after propagation to the downstream, enabling the attention layer to distinguish between strongly coupled and weakly coupled segment pairs.

[0039] Step 303: Decode the prediction sequence. The full-chain state representation is input into the decoder. The decoder adopts a multi-output temporal decoding structure symmetrical to the encoder, and is connected to a fully connected output layer at the end. The output dimension of the fully connected output layer is the product of the number of effluent parameters of each process segment and the number of preset time window steps, respectively outputting the effluent multi-parameter prediction sequence of each process segment within the future preset time window. The effluent multi-parameter prediction sequence output by the fully connected output layer is a value in the Z-score normalized space. After the inverse Z-score transformation of the corresponding parameters, it is restored to the parameter prediction value in the original dimension, which is used for comparison with the process protection threshold and the end-of-pipe zero discharge compliance threshold in the subsequent step 4.

[0040] In this embodiment, to enhance the ability of the cascaded state propagation prediction network to capture the nonlinear coupling effect between process segments, each temporal coding branch in the encoder outputs intermediate layer hidden state features in addition to the process segment state vector. Before the attention layer in step 302, a cross-segment feature interaction layer is added: the intermediate layer hidden state features of each process segment are concatenated according to the adjacent segment pairs in the process chain cascaded propagation parameter matrix, mapped by a fully connected layer to generate inter-segment interaction features, which are then concatenated with the corresponding process segment state vectors before being input into the attention layer. This cross-segment feature interaction layer enables the cascaded state propagation prediction network to capture fine-grained relationships between adjacent segments at the feature level before fusion at the state vector level.

[0041] Step 4: Generate tiered regulatory early warning signals The predicted effluent parameters for each process stage are compared parameter by parameter with the corresponding process protection threshold and end-of-pipe zero-discharge compliance threshold for each process stage. When any predicted parameter in the predicted effluent parameters of an upstream process stage exceeds the process protection threshold, the expected exceedance magnitude and expected arrival time of the abnormal propagation of this parameter to the end-of-pipe effluent node are calculated based on the process chain cascade propagation parameter matrix. A tiered regulatory early warning signal is generated based on the expected exceedance magnitude and remaining response time, and the tiered regulatory early warning signal and the source process stage identifier are pushed to the regulatory platform.

[0042] Step 4 may specifically include: Step 401, parameter-by-parameter comparison. For each parameter prediction value in the multi-parameter prediction sequence of each process section's effluent, compare the predicted value with the corresponding process protection threshold for that process section. When the predicted value exceeds the process protection threshold, mark the parameter, the process section, and the corresponding prediction time point as an anomaly trigger point.

[0043] Step 402: Calculate the expected exceedance magnitude and expected arrival time at the end point. For each abnormal trigger point, extract the actual propagation delay value and attenuation coefficient of each adjacent segment pair between the process segment where the abnormal trigger point is located and the end effluent node under that parameter category from the process chain cascade propagation parameter matrix. Accumulate the actual propagation delay values ​​of each adjacent segment pair along the propagation path to obtain the expected arrival time of the parameter abnormality from the triggering process segment to the end effluent node. Multiply the difference between the predicted parameter value of the abnormal trigger point and the process protection threshold by the attenuation coefficient of each adjacent segment pair along the propagation path to obtain the expected residual abnormality amount when the parameter abnormality reaches the end effluent node. Compare the expected residual abnormality amount with the end-of-pipe zero-emission compliance threshold to obtain the expected exceedance magnitude.

[0044] Step 403: Generate tiered early warning signals. Based on the expected exceedance magnitude and expected arrival time, the early warning level is divided into multiple levels. The highest-level early warning signal is generated when the expected exceedance magnitude is greater than a first magnitude threshold and the expected arrival time is less than a first time threshold. An intermediate-level early warning signal is generated when the expected exceedance magnitude is greater than a second magnitude threshold and the expected arrival time is less than a second time threshold. The lowest-level early warning signal is generated for all other cases where the process protection threshold is exceeded but the above conditions are not met. Specifically, the first magnitude threshold is greater than the second magnitude threshold, and the first time threshold is less than the second time threshold.

[0045] Step 404: Push early warning information. The tiered regulatory early warning signal, the traceability process segment identifier, the abnormal parameter category, the expected exceedance range, and the expected arrival time are packaged into an early warning message and pushed to the regulatory platform.

[0046] It should be noted that the aforementioned process protection thresholds and end-of-pipe zero discharge compliance thresholds are two different types of thresholds. Process protection thresholds are set separately for each process stage, reflecting the upper limit of permissible effluent parameters during normal operation of each process stage. End-of-pipe zero discharge compliance thresholds are set for the final effluent outlet, reflecting the legally mandated limits for discharge compliance. Process protection thresholds are typically stricter than end-of-pipe zero discharge compliance thresholds to allow for a safety margin.

[0047] In this embodiment, to avoid frequent false alarms caused by short-term fluctuations in the effluent multi-parameter prediction sequence, a sliding window smoothing process is applied to the effluent multi-parameter prediction sequence of each process segment before the parameter-by-parameter comparison in step 401. The length of the sliding window is determined based on the nominal hydraulic residence time of the corresponding process segment, and a preset proportion of the nominal hydraulic residence time is taken as the window length. The smoothed effluent multi-parameter prediction sequence retains the trend change and filters out high-frequency fluctuation components.

[0048] Step 5: Generate process control instructions Based on the expected exceedance range, expected arrival time, and source process segment identifier in the tiered regulatory early warning signal, matching items are retrieved from the pre-stored process control strategy library to generate parameter adjustment instructions for the abnormal source process segment, and the parameter adjustment instructions are recorded in the regulatory log.

[0049] Step 5 may specifically include: Step 501: Retrieve Control Strategies. Using the source process segment identifier, abnormal parameter category, and warning level as search keys, match the corresponding control strategy entries in the process control strategy library. Each entry in the process control strategy library predefines the correspondence between process segment type, abnormal parameter category, warning level, and control action.

[0050] Step 502: Generate parameter adjustment instructions. Based on the matched control strategy entries and the expected exceedance range, specific parameter adjustment instructions are generated. When the source tracing process section is the biochemical section and the abnormal parameter is ammonia nitrogen, the parameter adjustment instructions include an aeration rate adjustment value, with the adjustment direction being to increase the aeration rate. When the source tracing process section is the upstream section of the membrane system and the abnormal parameter is conductivity, the parameter adjustment instructions include an early trigger instruction for membrane system backwashing. When the warning level is the highest level, the parameter adjustment instructions also include an influent flow rate reduction ratio.

[0051] Step 503: Record regulatory logs. Write parameter adjustment instructions, corresponding tiered regulatory warning signals, traceability process segment identifiers, abnormal parameter categories, expected exceedance ranges, expected arrival times, and instruction generation timestamps into the regulatory logs to support subsequent compliance traceability.

[0052] It should be noted that the control strategy entries in the aforementioned process control strategy library are pre-configured static rules. The control actions in each entry include, but are not limited to, aeration rate adjustment, backwash triggering, influent flow rate reduction, and chemical dosage adjustment. The correspondence between the parameter adjustment range of each control action and the expected exceedance range is defined in the control strategy entry using a piecewise linear mapping.

[0053] Example 2 The zero-discharge treatment station for leachate from a landfill consists of six interconnected process stages: a conditioning and homogenization stage, a two-stage AO biological treatment stage, an ultrafiltration stage, a nanofiltration stage, and a reverse osmosis stage. The station has a daily treatment capacity of 300 cubic meters, and each process stage is equipped with online sensors connected to a monitoring platform. During a rainy season in a certain year, the COD and ammonia nitrogen concentrations of the influent significantly increased due to rainfall. The monitoring platform initiated a complete intelligent monitoring process at 08:00 on a certain day.

[0054] The process sections are numbered sequentially as follows: Process Section 1 (Equalization and Homogenization Section), Process Section 2 (AO Biochemical Section 1), Process Section 3 (AO Biochemical Section 2), Process Section 4 (Ultrafiltration Section), Process Section 5 (Nanofiltration Section), and Process Section 6 (Reverse Osmosis Section). The final effluent outlet is the outlet of Process Section 6.

[0055] The monitoring platform's data acquisition module collects and reports data from the online sensors of each process segment at 08:00. Since the COD sensor in process segment 2 samples every 5 minutes and the membrane flux sensor in process segment 4 samples every 1 minute, the data alignment process extends the COD sequence to one data point per minute through linear interpolation, ensuring that the parameter sequences of each process segment correspond point-by-point in the time dimension. After alignment, Z-score normalization is applied to each parameter sequence to generate a multi-process segment synchronous multi-parameter time series dataset, as shown in Table 1 below.

[0056] Table 1. Original reported data (partial parameters) for each process segment at 08:00. After Z-score standardization, the standardized value of the original COD value of 4850 mg / L in the effluent of process section 1 deviates significantly in the positive direction. This deviation will be used as an abnormal signal input into the subsequent network.

[0057] The cascade propagation parameter generation module determines the time-delay propagation window based on the nominal hydraulic residence time of each adjacent segment. It then retrieves the historical time series from the past 720 hours (i.e., 30 times the maximum nominal hydraulic residence time of 24 hours), performs cross-correlation analysis on each parameter of each adjacent segment, extracts the actual propagation delay value and attenuation coefficient, and organizes and generates the process chain cascade propagation parameter matrix. Taking COD parameters as an example, the cross-correlation function between the effluent COD sequence of process segment 1 and the influent COD sequence of process segment 2 reaches its peak at an offset of 26 hours. This value is the actual COD propagation delay value from process segment 1 to process segment 2. After time alignment, the average amplitude ratio of the two sequences is 0.17, which is the corresponding attenuation coefficient, as shown in Table 2 below.

[0058] Table 2. Process chain cascade propagation parameter matrix (COD and ammonia nitrogen parameters, adjacent segment pairs) The COD attenuation coefficient for process sections 1 to 2 is 0.17, reflecting the strong reduction effect of the first biological treatment stage on upstream abnormal signals; the COD attenuation coefficient for process sections 3 to 4 is 0.78, reflecting the weak reduction characteristics of the ultrafiltration stage on the biological effluent. These parameters will be used as the attention layer of the cascaded state propagation prediction network, representing structured prior constraints.

[0059] The cascaded state prediction module inputs the multi-process segment synchronous multi-parameter time series dataset generated in step 1 into the cascaded state propagation prediction network. The encoder's six temporal coding branches encode the multi-parameter sequences of process segments 1 to 6, respectively, with each branch outputting a fixed-dimensional process segment state vector. The cross-segment feature interaction layer concatenates the hidden state features of intermediate layers from adjacent segment pairs, maps them through a fully connected layer, generates inter-segment interaction features, concatenates them with the corresponding process segment state vectors, and then enters the attention layer.

[0060] The attention layer adjusts the COD attention score for process sections 1 to 2: attenuation coefficient Take 0.17, the actual propagation delay value Take 26 hours, The maximum actual propagation delay value of 40 hours is taken as the midpoint of all adjacent segments in the process chain. This represents the average of the original attention scores for the current batch. Adjusted attention score The strong anomaly signal of process section 1 is associated with the state vector of process section 2 with controlled weights. After attention-weighted aggregation, the decoder outputs the multi-parameter prediction sequence of the effluent from each process section for the next 12 hours, which is restored to the original dimensions by inverse Z-score transformation, as shown in Table 3 below.

[0061] Table 3 shows the predicted COD sequence of process section 1 effluent (next 12 hours) output by the cascaded state propagation prediction network. The predicted sequence showed that the COD of the effluent from process section 1 would continue to exceed the process protection threshold of 3500 mg / L for the next 12 hours, and showed an upward trend. After smoothing with a sliding window, the trend of exceeding the standard was preserved, while high-frequency fluctuation components were filtered out.

[0062] The early warning generation module performs end-point propagation calculations for the COD anomaly trigger point in process segment 1 (08:00, predicted value 4850 mg / L, exceeding threshold 1350 mg / L). It extracts the propagation path parameters from process segment 1 to process segment 6 from the process chain cascade propagation parameter matrix, and accumulates the actual propagation delay values ​​of each adjacent segment along the path. The estimated arrival time is approximately 68 hours later. Multiply the excess threshold of 1350 mg / L by the attenuation coefficient of each adjacent segment: mg / L, meaning the expected residual abnormality is approximately 5.4 mg / L. The threshold for zero discharge at the end of the process is a COD of no more than 60 mg / L. The predicted background value for the reverse osmosis effluent at the end of the process is 18 mg / L. After combining these values, the expected COD at the end of the process is approximately 23.6 mg / L, which is still within the threshold. However, the deviation at process section 1 is 38.6%, which is significant, as shown in Table 4 below.

[0063] Table 4 Summary of abnormal propagation path parameters and expected terminal impacts The estimated arrival time is 68 hours, which is greater than the first time threshold (set to 12 hours). The deviation of 38.6% from the threshold at process segment 1 is significant, triggering an intermediate-level warning signal (yellow warning). The warning message encapsulates the process segment 1 identifier, the abnormal parameter COD, the estimated deviation, and the estimated arrival time, and is pushed to the monitoring platform.

[0064] The control instruction generation module uses process section 1 (regulation and homogenization section), COD anomaly, and intermediate-level early warning as search keys to match corresponding entries in the process control strategy library. Since the source process section is the regulation and homogenization section and the abnormal parameter is COD, the control actions corresponding to the matched entries are influent flow reduction and reagent dosage adjustment. Based on the expected exceedance, parameter adjustment instructions are generated according to a piecewise linear mapping relationship: influent flow reduction ratio is 20%, and flocculant dosage is increased by 15%, as shown in Table 5 below.

[0065] Table 5 Process Control Instructions and Monitoring Log Records Throughout the implementation process, the data starts from the original sensor reports at 08:00 in each process segment, and after timestamp alignment and Z-score standardization, a multi-process segment synchronous multi-parameter time series dataset is formed. The second step extracts the actual propagation delay value and attenuation coefficient of each adjacent segment pair from 720 hours of historical data to form a process chain cascade propagation parameter matrix. This matrix serves as a structured prior constraint for the attention layer in the third step, so that the state vector of process segment 1 output by the encoder is associated with each downstream segment with controlled weights. The COD prediction sequence output in the third step is restored to its original dimensions by inverse Z-score transformation. In the fourth step, it is compared with the process protection threshold to trigger an anomaly. Then, the estimated residual anomaly amount at the end and the estimated arrival time in 68 hours are calculated through the propagation path to generate an intermediate-level warning signal. The fifth step uses the warning signal as a retrieval key to match the control strategy, output parameter adjustment instructions of reducing the influent flow rate by 20% and increasing the flocculant by 15%, and writes them into the monitoring log. The entire data flow, from upstream anomaly detection to the issuance of intervention instructions at the source, completes a closed loop before abnormal water volume enters the membrane system, demonstrating the strict logical transmission relationship between each step.

[0066] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0067] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart monitoring method for zero discharge of landfill leachate based on multi-parameter fusion, characterized in that, include: Acquire real-time data reported by online sensors in each process segment of the leachate zero-discharge treatment process chain, align them by process segment number and timestamp, and generate a multi-process segment synchronous multi-parameter time series dataset; Based on the hydraulic connection topology between adjacent process segments in the process chain and the nominal hydraulic residence time of each segment, the time delay propagation window of each adjacent segment pair is determined. Through cross-correlation analysis, the actual propagation delay value and attenuation coefficient of each water quality parameter between adjacent process segments are calculated, and the process chain cascade propagation parameter matrix is ​​generated. The multi-process segment synchronous multi-parameter time series dataset is input into a pre-trained cascaded state propagation prediction network, which outputs the multi-parameter prediction sequence of water effluent from each process segment within a future preset time window; wherein, the attention layer of the cascaded state propagation prediction network introduces the process chain cascaded propagation parameter matrix as a structured prior constraint. The predicted sequence of multiple parameters of effluent from each process section is compared with the process protection threshold and the terminal zero discharge compliance threshold parameter by parameter. When the predicted value of the parameter of the upstream process section exceeds the process protection threshold, the expected exceedance magnitude and expected arrival time of the abnormality propagation to the terminal effluent node are calculated based on the process chain cascade propagation parameter matrix, and a graded regulatory early warning signal is generated. Based on the aforementioned graded monitoring and early warning signals, matching items are retrieved from the process control strategy library to generate parameter adjustment instructions for the process segment at the source of the anomaly.

2. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, Determine the time-delay propagation window for each adjacent segment pair, including: For each pair of adjacent process segments, the nominal hydraulic residence time of the downstream process segment is obtained as the reference delay. The time delay propagation window of the adjacent segment pair is determined with the reference delay as the center and the preset multiplier range as the boundary. For each pair of adjacent process sections and each type of water quality parameter, extract the historical time series of the same parameter of the effluent from the upstream process section and the influent from the downstream process section, calculate the cross-correlation function of the two series within the time delay propagation window, and take the time offset corresponding to the peak value of the cross-correlation function as the actual propagation delay value of the parameter. After aligning the upstream effluent sequence and the downstream influent sequence with the actual propagation delay value, the mean amplitude ratio between the downstream influent sequence and the upstream effluent sequence is calculated, and the mean amplitude ratio is used as the attenuation coefficient of this parameter.

3. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, The cascaded state propagation prediction network employs an encoder-decoder structure. The encoder encodes the multi-parameter sequences of each process segment into process segment state vectors. The attention layer introduces the cascaded propagation parameter matrix of the process chain as a structured prior constraint, including: The state vectors of each process segment are used as query and key-value pairs in attention calculation. For each process segment... and process section Attention scores between The adjusted attention score is: ； in, For the process segment in the process chain cascade propagation parameter matrix To process section The attenuation coefficient corresponding to the parameter, , For process section To process section The actual propagation delay value of the corresponding parameter, It represents the maximum value of the actual propagation delay for all adjacent segments in the process chain. This is the mean of all raw attention scores in the current batch; After attention-weighted aggregation, the state vectors of each process segment are fused into a full-chain state representation. The decoder outputs the multi-parameter prediction sequence of the effluent from each process segment based on the full-chain state representation.

4. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, When the process section and process section When there is no direct hydraulic connection between them, Take zero, Take the maximum value This leads to the adjusted attention score. .

5. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, Each temporal coding branch in the encoder also outputs intermediate layer hidden state features; before the attention layer, the intermediate layer hidden state features of each process segment are concatenated according to the adjacent segment pairs in the process chain cascade propagation parameter matrix, and after mapping by the fully connected layer, inter-segment interaction features are generated. The inter-segment interaction features are concatenated with the corresponding process segment state vector and then input into the attention layer.

6. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, Based on the aforementioned process chain cascade propagation parameter matrix, the expected exceedance magnitude and expected arrival time of the anomaly propagation to the final effluent node are calculated, including: Extract the actual propagation delay value and attenuation coefficient of each adjacent segment pair between the abnormal triggering process segment and the terminal water outlet node under the corresponding parameter category from the process chain cascade propagation parameter matrix; The actual propagation delay values ​​of each adjacent segment are summed along the propagation path to obtain the estimated arrival time; The difference between the predicted parameter value of the abnormal trigger point and the process protection threshold is multiplied by the attenuation coefficient of each adjacent segment on the propagation path to obtain the expected residual abnormal amount. The expected residual abnormal amount is then compared with the end-of-pipe zero emission compliance threshold to obtain the expected exceedance range.

7. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, Generate tiered regulatory early warning signals, including: When the expected exceedance is greater than the first amplitude threshold and the expected arrival time is less than the first time threshold, the highest level warning signal is generated; when the expected exceedance is greater than the second amplitude threshold and the expected arrival time is less than the second time threshold, an intermediate level warning signal is generated; and for other cases exceeding the process protection threshold, the lowest level warning signal is generated; wherein, the first amplitude threshold is greater than the second amplitude threshold and the first time threshold is less than the second time threshold.

8. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, Before the parameter-by-parameter comparison, a sliding window smoothing process is applied to the multi-parameter prediction sequence of the effluent from each process section. The length of the sliding window is a preset proportion of the nominal hydraulic residence time of the corresponding process section.

9. The intelligent monitoring method for zero discharge of landfill leachate based on multi-parameter fusion according to claim 1, characterized in that, The process chain cascade propagation parameter matrix is ​​updated on a rolling basis according to a preset period. Each time it is updated, the actual propagation delay value and attenuation coefficient of each adjacent segment pair are recalculated using the historical data of the most recent preset duration, and the old process chain cascade propagation parameter matrix is ​​replaced.

Images