Narrowband internet of things multi-parameter water quality monitoring and sensor self-calibration method

Abstract

This invention discloses a narrowband IoT method for multi-parameter water quality monitoring and sensor self-calibration, belonging to the field of water quality monitoring and IoT technology. The method involves deploying multi-parameter water quality sensors at water supply nodes to collect raw readings; nodes with low network coverage transmit these readings via low-power narrowband IoT. Edge gateways identify sensor drift locally based on exponentially weighted moving average control charts and only transmit abnormal frames. The cloud platform stores data in buckets according to nodes and constructs regional cross-plant federated water quality baselines, distinguishing between sensor drift and actual pollution through standardization. After confirming drift, least-squares fitting correction coefficients are used and downloaded over the air; if necessary, physical calibration with standard liquid is triggered, forming a monitoring-identification-calibration closed loop. This invention achieves accurate monitoring of multi-parameter water quality and online self-calibration of sensor drift under low-power, wide-coverage conditions, reducing communication overhead and maintenance costs.

Description

Technical Field

[0001] This invention relates to the fields of water quality monitoring and Internet of Things (IoT) technology, specifically to a multi-parameter water quality monitoring and sensor self-calibration method based on narrowband IoT. Background Technology

[0002] Urban and rural water supply is an essential infrastructure for ensuring the safety of drinking water and public health. Continuous monitoring of the effluent from urban and rural water plants and the water quality in the distribution networks is crucial. With the expansion of water diversion projects and the associated water supply systems, the number of water supply points has increased significantly and their distribution has become more dispersed, posing a serious challenge to the real-time and accurate acquisition of water quality parameters.

[0003] In existing technologies, water quality monitoring IoT systems typically employ a centralized architecture: water quality sensors are deployed at each water supply node, and the probe responses are uploaded to the cloud via a network, where the cloud platform handles data storage, analysis, and anomaly warnings. For example, invention patent CN114910622B discloses a calibration scheme for water quality monitoring IoT sensors. This scheme includes an inlet system, a reaction system, an online monitoring system, a control system, and a data storage and processing system. It collects probe responses through online monitoring, performs data preprocessing (timestamp verification, error deletion, and missing data completion), and then uses a deep learning model to correct the IoT sensors. The corrected water quality monitoring model is then obtained by optimizing the solution. This scheme improves the measurement accuracy of single-node sensors to a certain extent.

[0004] However, the aforementioned existing technologies have the following shortcomings: First, this scheme relies on physical calibration devices and centralized deep learning computation to calibrate a single sensor in isolation, failing to consider the commonalities in water quality and drift correlations among multiple water plants under the same water source, making it difficult to distinguish between the drift of a single sensor and actual water pollution events. Second, this scheme does not optimize for low-power, low-bandwidth transmission for water supply nodes with low network coverage, such as in rural areas, resulting in high communication costs and power consumption in narrowband IoT scenarios. Third, the calibration model update of this scheme relies on centralized optimization solutions, lacking support for on-site drift identification at the edge and cross-plant federated baselines, thus failing to achieve distributed online self-calibration. Therefore, there is an urgent need for a water quality monitoring method that can operate under low network coverage conditions, has cross-plant collaborative baselines, and can perform online self-calibration of sensor drift. Summary of the Invention

[0005] This invention aims to solve the technical problems of existing water quality monitoring IoT systems, such as isolated calibration of single sensors, lack of transmission optimization for low network coverage scenarios, and lack of cross-plant collaborative drift identification and online self-calibration capabilities. It provides a narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, which achieves accurate monitoring of multi-parameter water quality and online self-correction of sensor drift under low power consumption and wide coverage conditions.

[0006] The technical solution adopted by this invention to solve its technical problem is: a narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, characterized by the following steps: S1, deploying multi-parameter water quality sensors at the outlet and pipeline nodes of water supply nodes to collect raw readings of water quality parameters; S2, transmitting the raw readings of each multi-parameter water quality sensor to the IoT platform via a communication network, wherein water supply nodes located in low network coverage areas transmit the data via a narrowband IoT communication module in a low-power manner; S3, preprocessing the raw reading sequence at the edge gateway, and determining locally whether the corresponding sensor has malfunctioned or drifted based on an exponentially weighted moving average control chart. When a drift is suspected, only abnormal frames are sent; S4, the cloud platform stores the original readings in buckets according to the water supply node identifier, and constructs a regional cross-plant federal water quality baseline based on multiple water supply nodes. The readings of a single water supply node are standardized to identify drift correlations and pollution sources across water supply nodes; S5, when sensor drift is confirmed, the correction coefficient is obtained by least squares fitting based on the pairing window of the recent original readings and the federal baseline, and the correction coefficient is sent to the edge gateway via over-the-air download. If necessary, the calibration actuator is driven to complete physical calibration; S6, the correction coefficient is written back to the edge gateway, and the corrected readings are continuously monitored to form a monitoring-identification-calibration closed loop.

[0007] In the above-mentioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S2, the communication network includes one or more of narrowband IoT, fourth-generation mobile communication network and optical fiber, and the raw readings are uniformly encapsulated into message queue telemetry transmission messages and then sent to the IoT platform.

[0008] In the aforementioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S3, the edge gateway calculates the exponentially weighted moving average sequence. and control limits ,in , For the first The original readings from the second sampling. For smoothing coefficients, To control the limit multiple, The standard deviation of historical readings; when or The sensor was marked as potentially drifting.

[0009] In the aforementioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, step S4 establishes the regional cross-plant federal water quality baseline using the following formula: ,in , The first Historical readings mean and variance of each water supply node The total number of water supply nodes participating in the federation; readings for individual water supply nodes. of Standardize the fractions and based on Size identification of cross-water supply node drift correlation and pollution source tracing.

[0010] In the aforementioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S5, the correction coefficient... , The following was obtained through least squares fitting: Corrected reading The correction coefficient , It is downloaded to the edge gateway via over-the-air download.

[0011] In the aforementioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S5, when the deviation of the calibrated reading relative to the federal baseline still exceeds a preset threshold, the calibration execution mechanism is triggered to inject standard solution into the corresponding sensor probe to complete physical calibration. After physical calibration, the correction coefficient is re-estimated based on the reference value of the standard solution. , .

[0012] The above-mentioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S1, the water quality parameters collected by the multi-parameter water quality sensor include at least two of pH value, oxidation-reduction potential, conductivity, turbidity, and residual chlorine.

[0013] In the aforementioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S3, before determining suspected drift, the edge gateway performs preprocessing on the original reading sequence, including timestamp verification, outlier removal, and gap filling.

[0014] In the above-mentioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S3, when the edge gateway determines that a sensor is suspected of drifting, it only sends the abnormal event and reading summary frame through the narrowband IoT, while the normal reading is cached on the edge side and periodically reported in batches to reduce the communication power consumption and bandwidth occupation in areas with low network coverage.

[0015] In the above-mentioned narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method, in step S4, the cloud platform groups multiple water supply nodes according to the water source zone to which the water supply node belongs, and constructs regional cross-plant federal water quality baselines for each zone according to the water supply node identifier.

[0016] The beneficial effects of this invention are:

[0017] (1) By using narrowband IoT low-power wide-coverage transmission and edge-side local drift identification, only abnormal events and summary frames are sent, significantly reducing the communication power consumption and bandwidth occupation of water supply nodes with low network coverage in rural areas.

[0018] (2) By constructing a regional cross-plant federal water quality baseline and standardizing single-node readings, it is possible to distinguish between sensor drift and real water pollution events, thereby improving the accuracy of anomaly detection and pollution source tracing capabilities.

[0019] (3) By fitting the correction coefficients online using least squares and downloading them over the air, and with the necessary physical calibration using standard liquid, a closed loop of monitoring-identification-calibration is formed, which extends the effective service life of the sensor and reduces maintenance and reagent consumption. Attached Figure Description

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] Figure 1 This is an overall flowchart of a narrowband Internet of Things (IoT) multi-parameter water quality monitoring and sensor self-calibration method according to the present invention.

[0022] Figure 2 This is a system architecture and hardware execution link diagram of the method of the present invention. Detailed Implementation

[0023] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0024] First, multi-parameter water quality sensors are deployed at the outlets and network nodes of the water supply nodes. In this embodiment, multi-parameter water quality sensors are deployed at the outlets of clear water tanks in urban and rural water plants, the outlets of secondary pump stations, and typical network terminal nodes. The collected water quality parameters include at least two of pH value, oxidation-reduction potential, conductivity, turbidity, and residual chlorine. pH value, oxidation-reduction potential, and conductivity are measured using composite electrode probes; turbidity is measured using a light scattering probe; and residual chlorine is measured using an electrochemical probe. Each sensor outputs raw readings via a standard serial interface. In this embodiment, the sampling period is 5 minutes. The edge gateway aggregates the probe signals via a standard serial bus and timestamps the raw readings before they enter the preprocessing stage. To suppress slow drift caused by biofouling on the probe surface, a self-cleaning mechanism is added around the probe in this embodiment, and the self-cleaning action is recorded in the reading metadata to eliminate cleaning disturbances in subsequent drift identification. For critical nodes, redundant probes with the same parameters are also deployed to provide a comparison benchmark in case of single probe failure.

[0025] In the probe installation phase, considering the humid and stormy environment of rural water plant intake pumping stations, in this embodiment, both the sensor probe and the edge gateway are placed in a protective box with an outdoor protection level that meets the requirements. The signal cable is a shielded cable and reliably grounded to suppress power frequency and lightning interference. The flow-through probe is equipped with a pre-filter and a flow stabilization section to avoid reading jumps caused by the accumulation of suspended solids and the attachment of air bubbles. The submersible probe is fixed to a preset depth and avoids the water inlet disturbance area. The installation location is selected based on the principle of representing the main water quality of the node and minimizing local disturbances.

[0026] Building upon the aforementioned deployment, the edge gateway first performs preprocessing on the raw reading sequence. Preprocessing includes three sub-steps: timestamp verification, outlier removal, and gap filling. Timestamp verification removes out-of-order frames caused by communication disorder or clock drift, ensuring the sequence time monotonically increases. Outlier removal uses historical interval-based limit judgments, marking readings significantly exceeding the reasonable range of physical quantities (e.g., pH values ​​exceeding the reasonable range, negative turbidity values) as invalid. Gap filling fills short-term gaps caused by communication frame drops using linear interpolation or forward hold between adjacent valid readings. Missing segments exceeding a preset length are marked as unusable data and trigger a re-acquisition request to ensure the continuity and stability of subsequent drift identification sequences. Furthermore, the edge gateway applies a sliding median filter to each channel reading to suppress occasional pulse interference and uses debouncing logic to prevent single spikes from falsely triggering drift criteria.

[0027] Each preprocessed reading carries metadata, including water supply node identifier, channel number, sampling timestamp, original and processed readings, quality flag bit, and edge side criterion output. This metadata is uploaded along with the message and archived in the cloud in buckets, enabling subsequent federal baseline calculation, drift discrimination, and calibration coefficient traceability to be performed in a closed loop under the same data model, thereby ensuring the traceability and data consistency of each link in monitoring, identification, and calibration.

[0028] Based on the preprocessed readings, the data enters the transmission stage. Water supply nodes located in urban areas with good network conditions transmit data to the IoT platform via 4G mobile communication networks or fiber optics; water supply nodes located in rural areas or other areas with low network coverage transmit data via narrowband IoT communication modules in a low-power mode. In this implementation, the narrowband IoT communication module is configured to operate in a periodic wake-up and transmission mode, entering sleep or power-saving mode after each transmission. It adapts to the deployment conditions of rural network points with limited power supply and battery life through discontinuous reception and extended discontinuous reception parameters. All raw readings are uniformly encapsulated into message queue telemetry transmission messages, written to the cloud time-series database through the IoT platform access layer, and the message transmission layer uses an encrypted channel to ensure data integrity and identity authenticity. In the event of a short-term network interruption, the edge gateway caches the unsent messages locally and retransmits them in order after the link is restored to ensure data continuity.

[0029] Building upon the aforementioned transmission stage, the edge gateway uses an exponentially weighted moving average control chart to locally determine whether the sensor has malfunctioned or drifted. Specifically, the edge gateway calculates an exponentially weighted moving average sequence for the readings of each channel.

[0030]

[0031] Based on this, the upper and lower control limits are calculated.

[0032]

[0033] in

[0034]

[0035] In the formula For the first The original readings from the second sampling. As a smoothing coefficient, this embodiment takes... ; To control the limiting multiple, this embodiment takes ; The standard deviation of historical readings is obtained statistically during the initial baseline establishment phase of deployment. In the initial warm-up phase after method startup, the control limits are directly given by the historical baseline values. Once the local sequence accumulates to a preset sample size (50 samples in this embodiment), the control limits are switched to time-varying limits derived from the above formula to balance startup stability and sensitivity to slow drift. When a certain reading meets... or When a sensor channel is suspected of drifting, the edge gateway marks it as such and generates an abnormal event frame. By sending abnormal frames only when a drift is suspected, and caching normal readings at the edge and reporting them in batches periodically, this embodiment can reduce the daily uplink message volume of nodes with low network coverage by more than an order of magnitude, significantly saving communication resources for narrowband IoT. Multiple channels are determined independently without interference; any channel exceeding its limit does not affect the normal reporting of other channels.

[0036] Meanwhile, the cloud platform stores the raw readings in buckets according to the water supply node identifiers, and constructs a regional cross-plant federated water quality baseline based on multiple water supply nodes. Since multiple water plants within the same water source zone share common water quality characteristics, this invention utilizes the participation of federated systems... The water supply nodes are cooperative nodes, and the regional federated mean and variance are constructed as follows:

[0037]

[0038] in , The first The historical mean and variance of each water supply node are obtained from the stable operating condition readings of that node over a recent period. To determine the total number of water supply nodes participating in the federation, this embodiment uses the same water source zone. A water plant. To account for seasonal and operational changes, the federal baseline is updated on a rolling basis according to a preset cycle (7 days in this example), and an old sample is downweighted by introducing a forgetting factor to prevent the baseline from being solidified by long-term historical means. When constructing the baseline, the cloud platform truncates the statistics of participating nodes to remove outliers from individual nodes, improving the robustness of the baseline. The cloud platform provides real-time readings of a single water supply node. conduct Score standardization:

[0039]

[0040] when Exceeding the preset standardized threshold (in this embodiment, it is taken as...) When this occurs, the node's reading is determined to deviate from the regional federal baseline. If only a single node shows a large deviation, it is considered a deviation. If the readings of other nodes are normal, it is likely that the sensor at that node is drifting; if multiple nodes simultaneously show large readings in the same direction... This approach tends to identify events as genuine water pollution incidents, thus enabling drift correlation and pollution source tracing across water supply nodes. The aforementioned federal baseline is constructed according to water source zones, with each zone calculated independently and stored in separate bins, avoiding misjudgments caused by differences in the baseline water quality of different water sources.

[0041] To further enhance the automation of calibration, once the cloud platform confirms that a sensor has drifted based on the federated baseline, it initiates a self-calibration process. This invention retrieves the sensor's most recent raw readings. Compared with the corresponding federal baseline reference value Paired windows are formed, and least squares fitting is used to obtain linear correction coefficients. , :

[0042]

[0043] The corrected reading is calculated using the following formula:

[0044]

[0045] Correction coefficients obtained from fitting , The data is downloaded over the air to the corresponding edge gateway, which then uses the corrected readings for monitoring and reporting. In this embodiment, the pairing window consists of 24 consecutive hours of reading pairs, and the window length can be adjusted between 12 and 48 hours depending on node communication conditions. When the correlation coefficient within the pairing window is lower than a preset level (in this embodiment, the correlation coefficient is less than 0.9), the linear model is deemed insufficient, and piecewise or higher-order fitting is used, or robust reweighting is applied to outliers within the window before fitting. To prevent individual abnormal samples from distorting the coefficients, this embodiment applies reasonable boundary constraints to the fitted coefficients; coefficients exceeding the boundaries are truncated and recorded in the alarm to avoid erroneous coefficients contaminating subsequent monitoring.

[0046] In implementation, the aforementioned correction coefficients are downloaded and distributed over the air using a reliable transmission mechanism with version numbers and verification: the cloud platform assigns an incrementing version number to each distributed coefficient packet and calculates an integrity verification value. The edge gateway receives the packet, verifies it, and then enables it. If the new coefficient causes the corrected reading to deviate further from the baseline, it automatically reverts to the previous available version and reports the anomaly, thus forming a secure closed loop of distribution, verification, and reversion. This mechanism ensures that the self-calibration process remains controllable even in rural scenarios where communication is unreliable.

[0047] In the aforementioned over-the-air download mechanism, the coefficient package is signed by the cloud platform with a key before it is distributed. The edge gateway verifies the validity of the signature before it can be enabled, preventing illegal or erroneous coefficients from being injected into the execution link. When a new water supply node is added to the federation, the baseline of the neighboring nodes with the same source is used as a cold start reference. After it accumulates enough stable samples, it is automatically merged into the federation baseline calculation, thereby achieving elastic expansion of the federation scale without affecting the discrimination stability of existing nodes.

[0048] In practice, if, after the aforementioned coefficient correction, the deviation of the corrected reading relative to the federal baseline still exceeds a preset threshold (in this embodiment, the relative deviation exceeds 8%), it is determined that the sensor has a hard offset exceeding the linear correction range or probe contamination. This triggers the calibration actuator to inject a standard solution into the corresponding sensor probe to complete physical calibration. The standard solution concentration is a well-known standard reference concentration in the art. After injection, the probe response under the standard solution environment is collected, and the correction coefficient is re-estimated based on the standard solution reference value. , The re-estimation results are then written back to the edge gateway via over-the-air download, restoring sensor measurement accuracy without human intervention. The calibration actuator, driven by a controller, uses a solenoid valve and a micro-pump to inject and flush quantitative standard solutions. After physical calibration, normal monitoring automatically resumes. In this embodiment, physical calibration involves multi-point standard solution injection, collecting probe responses at low and high concentrations of standard solutions. The re-estimation coefficient is calibrated at two points to improve linearity over a wide measurement range. The timing, volume, and residence time of injection and flushing are preset by the controller according to the probe type, and the deviation before and after each physical calibration is recorded to evaluate calibration effectiveness.

[0049] In terms of the execution details of physical calibration, the calibration actuator presets the injection and rinsing sequence according to the probe type: first, the probe surface and flow path are rinsed with pure water to remove residues; then, low-concentration and high-concentration standard solutions are injected sequentially. Each standard solution remains at the probe until the reading stabilizes, and a reference response is collected. The linear coefficient is re-estimated based on the two-point response and the known concentration of the standard solution; subsequently, it is rinsed again with pure water and normal water sample flow is restored. The controller records the injection volume, residence time, and deviation before and after calibration for each injection. When the residual deviation of multiple consecutive physical calibrations is less than the preset level, the probe is deemed to have recovered and the latest coefficient is set to effective; if the residual deviation still exceeds the limit, a probe aging or damage alarm is generated and on-site replacement is prompted. The above physical calibration actions are automatically triggered at the edge side according to a preset cycle or when linear self-calibration is insufficient to correct the problem, without the need for on-site operation by maintenance personnel.

[0050] During the continuous operation phase of the aforementioned closed loop, the edge gateway persistently stores the currently effective correction coefficients and version numbers on local non-volatile media. Even if communication is interrupted or the cloud platform is temporarily unavailable, the edge side continues monitoring and uploading with the most recently effective coefficients, ensuring monitoring continuity. After communication is restored, the cloud and edge sides are aligned with the version numbers to avoid coefficient rollback conflicts. When a water supply node leaves the federation due to long-term shutdown, the cloud platform removes it from the federation baseline calculation of the corresponding partition and reconstructs the baseline, while other nodes are unaffected. When a node reconnects, it is re-integrated using the aforementioned cold start reference logic, thereby ensuring the long-term robustness and maintainability of the federation baseline.

[0051] To address the independence of measurement channels within a multi-parameter probe, this invention models and solves the drift identification and self-calibration of each channel separately, avoiding the impact of drift correction on other channels. When multiple channels of the same probe exhibit unidirectional drift at the same time, it is preferentially determined to be an abnormality in the probe's power supply or signal link, triggering an overall self-test, rather than blindly correcting each channel separately. This prevents hardware faults from being misjudged as independent sensor drifts and introducing additional errors.

[0052] Corresponding to the aforementioned steps, this invention continuously writes the correction coefficients back to the edge gateway and continuously monitors the corrected readings, forming a closed loop of perception—transmission—edge identification—cloud analysis—self-calibration execution—data back. In this closed loop, the local drift identification on the edge side and the cross-plant federated baseline on the cloud side complement each other: the edge side is responsible for low-latency, low-bandwidth anomaly screening, while the cloud side is responsible for cross-node correlation discrimination and calibration coefficient calculation. The two work together to control communication overhead while ensuring discrimination accuracy. Both correction coefficients and calibration events are recorded in the node-level audit log for operation and maintenance traceability.

[0053] Based on the aforementioned anomaly detection, this invention jointly detects sensor malfunctions and actual contamination, forming the following decision logic: when a certain node... If the threshold is exceeded and the exponentially weighted moving average control chart of the verification channel also continuously exceeds the limit, while the readings of adjacent nodes remain stable, sensor drift is preemptively identified and self-calibration is initiated; when multiple nodes synchronously exhibit the same direction at similar times... When the threshold is exceeded but the edge-side control chart does not exceed the limit independently, it is judged as a suspected contamination event and a contamination source tracing alarm is generated, which is then transferred to the operation and maintenance personnel for review. In this case, self-calibration is not triggered to avoid masking the true contamination; when the edge-side control chart exceeds the limit but... If the score does not exceed the threshold, it is determined to be a short-term disturbance in a single node and reported as an abnormal event for observation. Through the above collaborative discrimination logic, this invention improves the timeliness of detecting contamination events while reducing the false calibration rate.

[0054] In the implementation of joint discrimination, this invention combines the edge-side control map's out-of-limit state with the cloud-side control map's state. The combination of fractional limit exceedance states is used for two-dimensional discrimination: when both exceed the limit, it is judged as sensor drift and self-calibration is initiated; when only exceeds the limit... When the score exceeds the limit but the control chart does not, it is judged as suspected contamination and an alarm is generated; when only the control chart exceeds the limit... If the score does not exceed the limit, it is judged as a short-term disturbance of a single node and observed as an abnormal event; if neither exceeds the limit, normal monitoring is maintained. This two-dimensional discrimination reduces the probability of false action under transitional conditions by a single criterion, and ensures that the drift calibration and pollution alarm actions do not overlap logically.

[0055] As another optional implementation, the water supply nodes participating in the federation in the aforementioned method are not limited to the same water source zone, but can be grouped according to administrative or dispatching jurisdiction to construct the federation baseline. The construction formula and standardization judgment logic are consistent with those described above, with only adaptive adjustments made to the grouping dimension. When a node is temporarily unable to upload data due to maintenance or failure, that node will not participate in the current round of federation baseline calculation. It will be automatically merged in after the data is restored, thereby ensuring the robustness of the federation baseline. In deployment scenarios with only a single water supply node and no cross-node federation conditions, redundant reference probes deployed within the same node can be used as the benchmark for that node. The reference probe readings and the main probe readings form a paired window to perform the aforementioned least-squares self-calibration. Its calibration logic is equivalent to the cross-plant federation method, and those skilled in the art can choose to implement one based on the scale of the nodes on site.

[0056] During implementation, it was observed that after adopting the cloud-edge collaboration and cross-plant federated baseline described in this embodiment, the communication message volume of nodes with low network coverage decreased by about an order of magnitude compared with the full-process uploading scheme. The effective measurement cycle of the sensor under the condition of slow probe drift was extended, the proportion of false alarms caused by drift was significantly reduced, and the average detection delay of real water quality anomalies was significantly shortened compared with single-node isolated discrimination.

[0057] Based on the foregoing steps and parameter descriptions, those skilled in the art can implement this invention on the hardware execution link to achieve the technical effect of accurate monitoring of multiple water quality parameters in urban and rural water supply and online self-calibration of sensors.

Claims

1. A method for narrowband Internet of Things (IoT) multi-parameter water quality monitoring and sensor self-calibration, characterized in that, Includes the following steps: S1, Install multi-parameter water quality sensors at the water outlet and pipeline nodes of the water supply node to collect raw readings of water quality parameters. S2, the raw readings of each of the multi-parameter water quality sensors are sent to the Internet of Things platform via the communication network, wherein the water supply nodes located in low network coverage areas send the data in a low-power manner via a narrowband Internet of Things communication module. S3, the original reading sequence is preprocessed at the edge gateway, and the corresponding sensor is determined locally based on the exponentially weighted moving average control chart to determine whether an abnormality or drift has occurred. When it is determined to be a suspected drift, only the abnormal frame is sent up. S4, the cloud platform stores the original readings in buckets according to the water supply node identifier, and constructs a regional cross-plant federal water quality baseline based on multiple water supply nodes. The readings of a single water supply node are standardized to identify the drift correlation and pollution source tracing across water supply nodes. S5. When sensor drift is confirmed, the correction coefficient is obtained by least squares fitting based on the pairing window of the recent raw reading and the federated baseline, and the correction coefficient is sent to the edge gateway via over-the-air download. If necessary, the calibration actuator is driven to complete the physical calibration. S6 writes the correction coefficient back to the edge gateway and continuously monitors the data with the corrected readings, forming a monitoring-identification-calibration closed loop.

2. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S2, the communication network includes one or more of narrowband Internet of Things (IoT), fourth-generation mobile communication network, and optical fiber. The raw readings are uniformly encapsulated into message queue telemetry transmission messages and then sent to the IoT platform.

3. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S3, the edge gateway calculates the exponentially weighted moving average sequence. and control limits ,in , For the first The original readings from the second sampling. For smoothing coefficients, To control the limit multiple, The standard deviation of historical readings; when or The sensor was marked as potentially drifting.

4. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S4, the regional inter-plant federal water quality baseline is constructed using the following formula: ,in , The first Historical readings mean and variance of each water supply node The total number of water supply nodes participating in the federation; readings for individual water supply nodes. of Standardize the fractions and based on Size identification of cross-water supply node drift correlation and pollution source tracing.

5. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S5, the correction coefficient , The following was obtained through least squares fitting: Corrected reading The correction coefficient , It is downloaded to the edge gateway via over-the-air download.

6. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S5, when the deviation of the calibrated reading from the federal baseline still exceeds a preset threshold, the calibration actuator is triggered to inject standard solution into the corresponding sensor probe to complete the physical calibration. After the physical calibration, the correction factor is re-estimated based on the reference value of the standard solution. , .

7. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, The water quality parameters collected by the multi-parameter water quality sensor in step S1 include at least two of the following: pH value, oxidation-reduction potential, conductivity, turbidity, and residual chlorine.

8. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S3, before determining suspected drift, the edge gateway performs preprocessing on the original reading sequence, including timestamp verification, outlier removal, and gap filling.

9. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S3, when the edge gateway determines that a sensor is suspected of drifting, it only sends the abnormal event and reading summary frame via narrowband IoT, while normal readings are cached on the edge side and reported in batches periodically, so as to reduce communication power consumption and bandwidth occupation in areas with low network coverage.

10. The narrowband IoT multi-parameter water quality monitoring and sensor self-calibration method according to claim 1, characterized in that, In step S4, the cloud platform groups multiple water supply nodes according to the water source zone to which the water supply node belongs, and constructs regional cross-plant federal water quality baselines for each zone by bucketing according to the water supply node identifier.

Images