Heating network heater and unit drainage online monitoring method and system

Abstract

The invention discloses a heating network heater and unit drainage on-line monitoring method and system, and the method comprises the steps: carrying out the cooling conditioning of a multi-branch sampling, building a 25 + / -1 DEG C isothermal steady flow gating, and carrying out the distribution heat exchange; fusing water chemical conservation, delta P-Q consistency and working condition stability on gating data, constructing a unified state vector, and generating a distribution robust prediction interval and a self-anchoring dynamic threshold value; double thresholds and k-of-n are jointly triggered in a grading manner, and valve position fine adjustment, short pulse flushing, bypass or isolation are preferably selected under the constraint of a safety shield; a verification window is set for verification after disposal, and parameters are written back in a self-adaptive manner, so that false alarm and missing alarm are reduced, and safety and economy are improved.

Description

Technical Field

[0001] This invention relates to the field of online monitoring technology, specifically to a method and system for online monitoring of heating network heaters and unit condensate drains. Background Technology

[0002] The condensate quality of heating units and network heaters directly affects heat exchange efficiency, steam and water quality, and equipment lifespan. Current online monitoring methods largely rely on single-point conductivity, turbidity, or temperature thresholds, failing to adequately consider the impact of hydraulic disturbances such as insufficient sampling cooling, flow pulsation, and valve operation on readings, easily leading to false positives / false negatives. Deposition and resuspension in the sampling chain, cooler scaling, and micro-vaporization prolong the stabilization time, reducing the comparability of data across time periods; start-up and shutdown, seasonality, and load switching cause distribution drift, making it difficult to adapt fixed thresholds and simple moving averages. Traditional methods lack characterization of "uncertainty" and online calibration, and also lack causal decomposition of chemical anomalies (such as conductivity rise due to heat exchanger micro-leakage) and hydraulic anomalies (such as ΔP–Q mismatch and turbidity caused by resuspension). When limits are exceeded, handling often relies on manual experience or single rules, lacking optimal action selection and safety constraint frameworks, potentially leading to over-handling or delayed handling. In parallel multi-branch scenarios, crosstalk between channels and timing inconsistencies make "single-point triggering" unreliable. The lack of k-of-n consensus and cooling-off period mechanisms easily leads to jitter escalation. Furthermore, the effectiveness of these measures is rarely quantified, verified, or written back to parameters, making it difficult for the system to self-correct with seasonal changes and equipment aging. Engineering practice urgently needs a method to unify isothermal steady-current gating, physical consistency constraints, causal invariance fusion, and distributed bar prediction onto the same link. This would achieve a closed loop for data acceptance, anomaly identification, hierarchical triggering, action selection, and adaptive updates, reducing false alarms and missed alarms, and improving safety, economy, and compliance. Summary of the Invention

[0003] In view of the above-mentioned problems, the present invention is proposed.

[0004] Therefore, the technical problem solved by this invention is to reduce false alarms and missed alarms in hydrophobic monitoring and to optimize verifiable and adaptive hierarchical alarms and minimal intervention under multiple operating conditions and multiple branch conditions.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an online monitoring method for heat network heaters and unit condensate drains, comprising:

[0006] Sampling and cooling conditioning are performed on multi-sampling branches to generate isothermal reliable gating flags and allocate heat exchange resources.

[0007] Based on the data passed through the gating system, a state representation is established by combining physical consistency and causal invariance, and the prediction interval and anomaly probability of the sub-bar are obtained.

[0008] By utilizing probability and interval to jointly drive dynamic thresholds, a dual threshold is established to output graded alarms and limit the set of executable actions. The process of generating the set of executable actions includes: for the triggered event, using a safety shield to constrain the boundary and timing, and using the risk difference criterion to select the best option among valve position fine-tuning, short pulse flushing, bypassing, or isolation and generating a safety proof.

[0009] After processing, the difference between the expected and the actual results is verified in the verification window, and the results are written back to adaptively update the gating threshold, feature weights, interval calibration and policy parameters.

[0010] As a preferred embodiment of the online monitoring method for the heating network heater and the unit condensate of the present invention, the sampling and cooling conditioning includes five locations of the unit condensate and the heating network heater condensate. The condensate sample is cooled to 25℃±1℃ by a cooler. The condensate is equipped with an online monitoring device that can adjust the water quality at any time if changes are detected.

[0011] The drainage system includes: 1. Drainage from the heating network heater, 2. Drainage from the heating network heater, 1 and 2. Drainage from the exhaust cooler, 3. Drainage from the exhaust cooler, and 3 and 4. Boiler drainage. Sampling pipelines are connected to the outlet headers of each drainage pump, passing through the cooler and then connected to the online instruments.

[0012] As a preferred embodiment of the online monitoring method for heat network heaters and unit condensate drains described in this invention, the isothermal reliable gating and heat exchange resource allocation includes:

[0013] Within each sampling branch, the cooler outlet temperature and the measurement pool temperature are simultaneously monitored within a 5–10 s steady-state judgment window. The sample water is required to be stable at 25℃±1℃ and the temperature difference between the outlet and the measurement pool is not greater than 0.5℃. The stabilization time and overshoot count are recorded as the hot steady-state temperature score. Simultaneously, the sampling flow rate and the pressure difference before and after the throttling device are compared for consistency. The flow rate fluctuation and pulsation are limited to not exceeding the set threshold and the ΔP–Q relationship is in the healthy zone, forming the hydraulic steady-state flow score. The hot steady-state temperature score and the hydraulic steady-state flow score are summarized into a gating score according to the preset weight and a confidence level is given. When the score is not lower than the threshold, an isothermal reliable gating flag is output. If the score does not meet the threshold, the branch is temporarily suspended from entering the subsequent judgment and only the minimum cooling guarantee is retained.

[0014] The ΔP–Q relationship represents the characteristic curve of pressure difference as a function of flow rate;

[0015] Under the constraints of total cooling capacity and cooler pressure drop limit, a rolling priority queue is constructed based on gate confidence, predicted settling time, residual temperature difference and branch importance. The quota increment of cooling valve position or bypass ratio is issued to priority branches in 5-10s time slices, and the coupled branches are subjected to 1-2s phase stagger and decreasing step to suppress crosstalk.

[0016] When the lower layer is executed, the valve position change rate is not greater than 5% / s, the minimum dwell time for a single target is 30–120s, and the branch pressure drop does not exceed the upper limit constraint.

[0017] At the end of each time slice, the temperature difference drops and stabilizes. Two consecutive slices without improvement are automatically downgraded and their quota is reclaimed for other non-compliant branches. When a branch meets the gating threshold, it maintains minimum residence and gradually reduces its quota, and the released heat exchange capacity is recovered and redistributed.

[0018] The gating flag, confidence level, allocation result, and stabilization record are used together as weight inputs for subsequent state modeling, prediction interval, and dynamic threshold setting to ensure that the data entering alarm triggering and handling decision-making are comparable and stable.

[0019] As a preferred embodiment of the online monitoring method for heat network heaters and unit condensate drains described in this invention, a unified state representation is established after the isothermal reliable gate flag is passed.

[0020] Temperature-compensated conductivity calculations were performed on the sample water data after temperature and flow stabilization, and conservation deviations were formed based on the water chemistry conservation law and the steam condensation and dilution law.

[0021] The hydraulic consistency deviation of ΔP–Q along the temperature residual and the stabilization time of the same branch are used as hydraulic evidence.

[0022] The operating condition domains are divided according to unit load, start-up and shutdown and season. The stability of candidate features in multiple domains is calculated. Only the subset of features with stability above the threshold is retained. Unstable elements are downweighted or isolated.

[0023] The conservation bias, hydraulic consistency bias, cross-domain stability, as well as the gate confidence, stabilization time, and residual temperature difference are fused into a unified state vector after being aligned by time, and the operating condition domain label and the main evidence of anomalies are output.

[0024] The unified state vector serves as the direct input to the prediction interval and triggering criterion.

[0025] As a preferred embodiment of the online monitoring method for heat network heaters and unit condensate drainage described in this invention, the generation of the Bruker prediction interval and the self-anchoring dynamic threshold includes: obtaining point predictions for compensated conductivity, turbidity, and hydraulic parameters within a set prediction window.

[0026] The gating is performed by dividing the samples into bins according to the working condition domain for interval calibration, so that the future intervals meet the coverage target within the significance range of 0.05 to 0.10, and the interval width is output to characterize the uncertainty.

[0027] A drift sensor is set up to score the distance and coverage between the predicted distribution of points and the historical calibration distribution of each domain, and to monitor the offset. When any one exceeds the limit, the calibration radius is automatically expanded or the conservatism is increased.

[0028] The baseline median of the historical health period and the median of the recent 24-hour interval are jointly used to determine the threshold anchor point, and the dynamic threshold band is formed by adaptive biasing according to the coverage score, gating confidence and ΔP–Q health.

[0029] Finally, the future interval, interval width, dynamic threshold band, and anchor point are output together with the unified state vector for subsequent hierarchical triggering and limitation of the set of allowed actions.

[0030] As a preferred embodiment of the online monitoring method for heat network heaters and unit condensate drainage described in this invention, wherein: after obtaining the dynamic threshold band and the future prediction interval, the actual trajectory and the minimum reversible intervention hypothesis trajectory are simultaneously deduced, and the limit is determined to be exceeded only when the two trajectories simultaneously satisfy the following within the duration T: the point prediction crosses the dynamic threshold band and the main body of the future prediction interval continuously crosses the threshold band.

[0031] Consistency confirmation for exceeding limits is achieved by using k-of-n consensus rules on multiple branches controlled by isothermal trusted gating.

[0032] After a limit violation is confirmed and the alarm level is upgraded, a cooling-off period begins, during which repeated upgrades are prohibited.

[0033] Alarm level is determined by a score based on interval width and coverage.

[0034] The alarm level, the set of allowed actions, and the suggested duration are output to the handling module.

[0035] The minimum reversible intervention assumption includes that the valve position is fine-tuned by no more than 2% or that a short pulse flush lasts for 3 to 10 seconds.

[0036] As a preferred embodiment of the online monitoring method for heat network heaters and unit condensate drainage described in this invention, wherein: for candidate actions in the set of permissible actions, the minimum alternative intervention is selected based on the risk difference criterion and the evidence gain threshold to obtain the handling action;

[0037] After the action is executed, the verification window is opened to perform expected-measured verification of stabilization time, temperature difference between outlet and measuring pool, ΔP–Q health, prediction interval width and conservation deviation. If successful, the dynamic threshold band is narrowed and the threshold anchor point bias is withdrawn, the relevant feature weights are increased and the normal calibration radius is restored. If unsuccessful, the calibration radius is expanded or the conservatism is increased, the relevant feature weights are reduced and the action strategy parameters are adjusted.

[0038] The strategy parameters include, but are not limited to: valve position single-step amplitude and cycle time, minimum dwell time, cooling period, short pulse flushing duration and number of cycles, bypass duration limit and concurrent number.

[0039] A heat network heater and unit condensate online monitoring system using the method described in this invention, wherein: a data acquisition unit performs sampling and cooling conditioning on multiple sampling branches, generates isothermal reliable gating flags and allocates heat exchange resources;

[0040] The analysis unit, based on the data passed through the gating system, combines physical consistency and causal invariance to establish a state representation, thereby obtaining the prediction interval and anomaly probability of the sub-bar.

[0041] The alarm unit uses a dual threshold driven by probability and interval to output hierarchical alarms and limit the set of executable actions.

[0042] The update unit processes the data and verifies the difference between the expected and actual values ​​within the verification window. The results are then written back to adaptively update the gating threshold, feature weights, interval calibration, and policy parameters.

[0043] A computer device includes: a memory and a processor; the memory stores a computer program, wherein: when the processor executes the computer program, it implements the steps of the method described in any one of the present invention.

[0044] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the present invention.

[0045] The beneficial effects of this invention are as follows: This invention uses isothermal steady-flow gating combined with heat exchange resource allocation to ensure the comparability and reproducibility of data entering the modeling process; it constructs a unified state vector based on water chemistry conservation and ΔP–Q consistency to distinguish between chemical and hydraulic causes, reducing misjudgments. The introduction of a split-blob prediction interval and self-anchored dynamic threshold, combined with dual-threshold triggering, k-of-n consensus, and a cooling period, significantly suppresses jitter and false alarms. Based on the boundary and temporal constraints of the safety shield, and in conjunction with risk difference and evidence gain, it selects the minimum viable alternative intervention (valve position fine-tuning, short-pulse flushing, etc.) to shorten recovery time and reduce external discharge while ensuring safety. After the intervention, a verification window is set and parameters are written back to achieve adaptive convergence of gating thresholds, feature weights, interval calibration, and strategies, maintaining long-term monitoring sensitivity and stability, and improving the safety, economy, and compliance of the heating system. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 The first embodiment of the present invention provides an overall flowchart of a method for online monitoring of heat network heaters and unit condensate.

[0048] Figure 2 A condensate system diagram of an online monitoring method for condensate draining from a heating network heater and a unit, provided for the first embodiment of the invention. Detailed Implementation

[0049] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0050] Example 1, referring to Figure 1 , Figure 2 As an embodiment of the present invention, a method for online monitoring of heat network heaters and unit condensate drains is provided, comprising:

[0051] S1: Sampling and cooling conditioning are performed on the multi-sampling branches to generate isothermal reliable gating flags and allocate heat exchange resources.

[0052] The sampling and cooling conditioning process includes five locations for the unit's condensate drains and the heating network heater drains. The condensate samples are cooled to 25℃±1℃ using a cooler. Online monitoring equipment is installed in the condensate drains to detect changes in water quality and allow for timely adjustments. The condensate drain system includes: 1. Heating network heater condensate, 2. Heating network heater condensate, 1 & 2. Exhaust cooler condensate, 3. Exhaust cooler condensate, and 3 & 4. Boiler condensate. Sampling lines are connected to the outlet headers of each condensate pump, passing through the cooler before being connected to online instruments.

[0053] The isothermal reliable gating flag and heat exchange resource allocation include:

[0054] Within each sampling branch, the cooler outlet temperature and the measuring pool temperature are simultaneously monitored within a 5–10 s steady-state judgment window. The sample water is required to be stable at 25℃ ± 1℃, and the temperature difference between the outlet and the measuring pool should not exceed 0.5℃. The stabilization time and overshoot count are recorded as the hot steady-state temperature score (obtained through a preset relational function). Simultaneously, the sampling flow rate and the pressure difference before and after the throttling device are compared for consistency. Flow fluctuations and pulsations are limited to not exceeding a set threshold, and the ΔP–Q relationship is within the healthy zone, forming the hydraulic steady-state flow score (obtained through a preset relational function). The hot... The steady-state temperature score and the hydraulic steady-state flow score are aggregated into a gating score according to preset weights and a confidence level is given. When the score is not lower than the threshold, an isothermal reliable gating flag is output. If the score does not meet the threshold, the branch is temporarily suspended from further judgment and only the minimum cooling guarantee is maintained (the purpose is to: protect instruments and pipelines: complete shutdown may lead to local overheating or deposits; maintain rapid recovery: with a minimum flow rate, short flushing or fine-tuning makes it easier to bring it back to steady state; and avoid contaminating other branches: give most of the heat exchange resources to branches that are "more likely to meet the standard immediately", so that the whole "collectively meets the standard" faster).

[0055] The ΔP–Q relationship represents the characteristic curve of pressure difference as a function of flow rate.

[0056] Under the constraints of total cooling capacity and cooler pressure drop limit, a rolling priority queue is constructed based on gating confidence, predicted settling time (calculated from near-window temperature slope and flow rate), residual temperature difference, and branch importance. Cooling valve position or bypass allocation increments are issued to priority branches in 5–10s time slices, and coupled branches are treated with 1–2s phase shifts and decreasing steps to suppress crosstalk. It's important to note that setting up a rolling priority queue under the constraints of total cooling capacity and cooler pressure drop limit aims to allocate resources to the most "cost-effective" targets: channels with high gating confidence, short predicted settling time, large residual temperature difference, and high branch importance, thereby minimizing the overall system settling time and false alarm risk. The 5–10s time slice is used to discretize continuous control, facilitating online priority reordering and alignment with the execution layer's dwell and phase strategies, avoiding frequent small-amplitude repetitions that cause "hunting" oscillations. The use of 1–2s phase shift and decreasing step size for coupled branches aims to reduce the instantaneous impact of concurrent valve position actions on the main pipe pressure drop and temperature, and suppress crosstalk amplification. The decreasing step size allows the "most promising" branch to have a larger step size, while other branches follow with smaller steps, thereby improving the overall convergence speed within the safety boundary. Lower-level execution enforces valve position change rate limits, minimum dwell time, and pressure drop limits to protect valves and heat exchangers, provide sufficient response time for the process, and ensure that the "improved / unimproved" judgment is statistically significant, preventing noise from being mistaken for a trend due to overly rapid actions.

[0057] When the lower layer is executed, the valve position change rate is not greater than 5% / s, the minimum dwell time for a single target is 30–120s, and the branch pressure drop does not exceed the upper limit constraint.

[0058] At the end of each time slice, the temperature difference decreases and stabilizes. Branches with two consecutive unimproved slices are automatically downgraded and their quotas are reclaimed for use by other non-compliant branches. Once a branch meets the gating threshold, it maintains minimum dwell time and its quota is gradually reduced, with the released heat exchange capacity being reclaimed and redistributed. Recording the temperature difference decrease and stabilizes at the end of the slice and downgrading and reclaiming quotas for branches with two consecutive unimproved slices aims to avoid burying resources in channels that are difficult to improve in the short term, freeing up capacity to support branches that are easier to meet the standard, and accelerating "collective compliance." For compliant branches, only minimum dwell time is maintained and their quotas are gradually reduced to prevent overcooling and unnecessary energy consumption, while the released heat exchange capacity is redistributed to those that have not yet met the standard.

[0059] By combining the gating flag, confidence level, allocation result, and stabilization record as weight inputs for subsequent state modeling, prediction intervals, and dynamic threshold settings, the comparability and stability of the data entering alarm triggering and handling decisions are ensured. Packaging the gating flag, gating confidence level, allocation result, and stabilization record as weight inputs for subsequent state modeling, prediction intervals, and dynamic threshold settings aims to allow the backend algorithm to "perceive the quality of frontend data and resource scheduling history," learning and triggering only on isothermal, stable, clean, and consistent samples, thus reducing threshold drift and alarm jitter from the source. Simultaneously, it provides traceable evidence for subsequent handling effect verification and parameter adaptation, forming a closed loop from resource allocation to model judgment to strategy update.

[0060] S2: Based on the data passed through the gating system, a state representation is established by combining physical consistency and causal invariance to obtain the prediction interval and anomaly probability of the sub-bar.

[0061] After the isothermal reliable gating flag is passed, a unified state representation is established:

[0062] Temperature-compensated conductivity was calculated for the sample water data after temperature and flow stabilization, and conservation deviations were established based on water chemistry conservation and steam condensation and dilution laws. The hydraulic consistency deviation ΔP–Q of the same branch, along with the temperature residual and stabilization time, were used as hydraulic evidence.

[0063] The operating condition domains are divided according to unit load, start-up and shutdown, and season. The stability of candidate features in multiple domains is calculated. Only the subset of features with stability above the threshold is retained, and unstable elements are downweighted or isolated.

[0064] The samples are labeled according to a three-dimensional operating condition domain, with loads divided into three to five levels, and start-up and shutdown categorized as start-up, shutdown, and steady state, and seasons as winter, summer, and transition. A three-dimensional Cartesian combination constitutes a composite domain g. Only data passing through S1 gating is used in this step; each domain must meet a minimum sample size requirement, otherwise, when that domain participates in the comparison, the parent domain or global backoff is used. Robust statistics and distribution distance: For any candidate feature, the robust median and scale are calculated in each domain g, and then the mean shift and distribution difference are compared pairwise for each domain, synthesizing a cross-domain stability score. Intra-domain robust statistics:

[0065] μ j,g =median(f j |g),s j,g =1.4826×MAD(f j |g)

[0066] μ j,g s is the median of feature j in the domain g; j,g is the robust scaling estimate of feature j in domain g; MAD is the statistic of absolute median deviation; 1.4826 is the coefficient that reduces MAD to the Gaussian equivalent standard deviation.

[0067] Two-domain standardized mean shift:

[0068]

[0069] Δμ j (g1, g2) represents the standardized median difference of feature j in the domains g1 and g2; ε is a minimal positive number used for numerical stability.

[0070] Robust normalization of the first-order Wasserstein distance between two domains:

[0071]

[0072] W j (g1, g2) is the normalization of the first-order Wassstein distance of the characteristic distribution; 1 is the first-order Wassstein distance operator; F j,g Let be the empirical distribution function of feature j in the domain g; IQR be the interquartile range of the merged samples; and be a very small positive number.

[0073] Normalized mutual information between features and domain labels:

[0074]

[0075] denoted as , where is the normalized mutual information between features and domain labels; MI is the mutual information operator; g is the composite domain label; H is the Shannon entropy of the features; and is a minimal positive number.

[0076] Instability and stability:

[0077]

[0078] U j The overall instability of feature j; For all pairs of fields, Δμ j The mean or upper quantile aggregate value; For W of all field pairs j The mean or upper quantile aggregate value; α is the weight of the mean displacement term; β is the weight of the distribution distance term; γ is the weight of the mutual information term; S j The cross-domain stability score ranges from zero to one.

[0079] Weighting and isolation of unstable elements:

[0080]

[0081] New weights for feature j; The underlying weights for this feature can be derived from professional priors or historical regression; θ iso If the isolation threshold is below this value, the feature is directly masked; θ full The full-weight threshold is set to not reduce weight if it exceeds this value; p is the smoothing exponent, which controls the steepness of the weight reduction curve.

[0082] The conservation bias, hydraulic consistency bias, cross-domain stability, as well as the gating confidence, stabilization time, and residual temperature difference are fused into a unified state vector after being aligned by time, and the operating condition domain label and anomaly master evidence are output; wherein, the unified state vector serves as the direct input for the prediction interval and triggering criterion.

[0083] Time alignment and gating mask:

[0084] y k =mean{y(t)|t∈[t] k -Δ,t k GatePass(t) = 1

[0085] y_k is the representative value of quantity y in the k-th time slice; y(t) is the original time series of quantity y; t_k is the end time of the k-th S1 allocation time slice; Δ is the half-width of the alignment window, ranging from one to five seconds; GatePass(t) is the value that is only included in the average when the gate pass indication is one; T_hold is the maximum holding time of the most recent valid representative value, used for temporary storage when the gate sample is insufficient.

[0086] Robust standardization of healthy period references:

[0087]

[0088] z r(x) represents the robust standardization result for quantity x; median ref (x) represents the median of the healthy reference window; IQR ref (x) represents the interquartile range of the reference window during the healthy period; ε is a very small positive number used to avoid division by zero and improve numerical stability.

[0089] Summary of evidence from both chemical and hydraulic perspectives:

[0090]

[0091] E chem,k E represents the chemical evidence score at time k. hyd,k The hydraulic evidence score at time k; For chemical-side characteristics, such as temperature-compensated conductivity and conservation bias; This includes hydraulic-side characteristics such as ΔP-Q deviation, differential pressure, high-frequency energy, and flow pulsation. The weights of feature j are updated based on cross-domain stability; f j,k z is the time-aligned representative value of feature j at time k; r (·) represents the robust normalized operator obtained according to formula H.

[0092] X k =[z r (σ 25,k 0,z r (ConsBias k ),z r (dev ΔP-Q,k ),z r (HF ΔP,k ),z r (CV Q,k ),

[0093] z r (ΔT res,k ),z r (τ settle,k ),GateConf k ,S σ ,S ConsBias ,S dev E chem,k ]

[0094] X k Let σ be the unified state vector column at time k; 25,k To compensate for temperature to the alignment value of the conductance at time k at 25 degrees Celsius and via z r Standardization; ConsBias k The alignment value of the water chemistry conservation deviation at time k and after z r Standardization; deg ΔP-Q,kThe normalized deviation of the ΔP-Q health band at time k is the alignment value and z-axis value. r Standardization; HF ΔP,k The alignment value of the differential pressure high-frequency energy at time k and after z r Standardization; CV Q,k The aligned value of the flow variation coefficient at time k and after z r Standardization; ΔT res,k The aligned value of the residual temperature difference at time k and after z r Standardization; τ settle,k To predict the alignment value of the steady-state time at time k and via z r Standardization; GateConf k The alignment value of the gate confidence at time k is set to zero to one; S σ The transdomain stability constant term corresponding to temperature-compensated conductance is used for threshold or interval weighting; S ConsBias S is the transdomain stability constant term corresponding to the conservation bias; dev E is the transdomain stability constant term corresponding to the ΔP-Q deviation; chem,k E represents the score for evidence from the chemical side. hyd,k The score for hydraulic evidence; q k The encoding of the operating condition field label can be either one-hot or embedded.

[0095] The generation of the segmented prediction interval and self-anchoring dynamic threshold includes: obtaining point predictions for compensated conductivity, turbidity, and hydraulic measurements within a set prediction window. Interval calibration is performed on the gated samples, categorized by operating condition domain, to ensure that future intervals meet the coverage target within a significance range of 0.05–0.10, and the interval width is output to characterize uncertainty. Significance 0.05–0.10: In statistics, α means 1-α = 90%–95% coverage. α = 0.05 → 95% coverage; α = 0.10 → 90% coverage.

[0096] A drift sensor is set up to score the distance and coverage between the predicted distribution of points and the historical calibration distribution of each domain, and to monitor the offset. When any one exceeds the limit, the calibration radius is automatically expanded or the conservatism is increased (first expand the calibration radius (extend the residual statistics window, fall back from the current bucket to the parent bucket / global bucket in the operating domain and increase the minimum sample size), and then increase the conservatism (widen the quantile to thicken the prediction interval, perform a safety side bias on the dynamic threshold band according to the interval width, increase the duration of the dual threshold and reduce the set of allowed actions, and increase the redundancy of the safety shield)).

[0097] First, making point predictions for compensated conductivity, turbidity, and hydraulic measurements only masks uncertainty, easily leading to false alarms that appear to exceed limits but are actually normal fluctuations during load switching, start-up / shutdown, or seasonal changes. Therefore, point predictions are first made on homogeneous samples passing through the gating system, and then the prediction intervals are calibrated by binning according to operating condition domains (load × start-up / shutdown × season), fixing the statistical significance α within the acceptable engineering band of 0.05–0.10 (i.e., 90%–95% coverage). The purpose is threefold: first, to clearly express the "expected fluctuation range" using intervals, so that triggering no longer depends on isolated points; second, to ensure consistent coverage across domains, avoiding systematic over-tightening or under-tightening in certain domains; and third, to use the output interval width as an online uncertainty measure, providing a quantitative basis for subsequent graded alarms, action intensity, and safety redundancy (the thicker the interval, the greater the uncertainty, and the more conservative the action).

[0098] Secondly, long-term drift or short-term disturbances in operating conditions can distort the "previously calibrated intervals." To address this, a drift sensor is installed to monitor two types of signals: the distance between the point prediction residual distribution and the historical calibration distribution, and the deviation of the actual coverage rate from the target coverage (90%–95%). The design aims to quickly identify the "model-reality" misalignment using dual evidence of distribution distance and coverage deviation. If either exceeds the limit, adaptive correction is performed in the order of "radius first, then conservatism": first, the calibration radius is expanded (extending the residual statistical window, reverting from the current sub-bucket to the parent / global bucket, increasing the minimum sample size) to increase sample stability and reduce estimation variance; if this still fails to meet the target, the conservatism is increased (widening the quantiles to thicken the interval, applying a safety bias to the dynamic threshold band, extending the duration of the dual thresholds, tightening the allowed action set, and increasing the safety shield redundancy). The purpose of this two-stage strategy is to prioritize data aggregation to offset noise without sacrificing sensitivity, and only when there is concrete evidence of drift is the safety redundancy increased proportionally, thereby stabilizing alarm quality and ensuring safe handling.

[0099] The baseline median of historical health periods and the median of the most recent 24-hour interval are jointly used to determine the threshold anchor point, which is then adaptively biased based on coverage scoring, gating confidence, and ΔP–Q health, forming a dynamic threshold band. This serves two purposes: first, to ensure the threshold is always "anchored" between the center of the health data and the median of the latest distribution, avoiding unilateral drift; second, to explicitly inject front-end data quality (gating confidence) and hydraulic health (ΔP–Q) into the threshold. The more reliable the front end, the tighter the threshold; if the front end is questionable, the threshold automatically loosens. This makes the threshold no longer a static constant, but a safety boundary that is "self-aware of data quality and operating conditions."

[0100] Finally, the future interval, interval width, dynamic threshold band, and anchor point are output together with the unified state vector for subsequent hierarchical triggering and limitation of the set of allowed actions.

[0101] S3: Utilize probability and interval to drive the dynamic threshold with dual thresholds, output graded alarms and limit the set of executable actions; the generation process of the set of executable actions includes: for the triggered event, use the safety shield to constrain the boundary and timing, and at the same time use the risk difference criterion to select the best option among valve position fine-tuning, short pulse flushing, bypass or isolation and generate a safety certificate.

[0102] After obtaining the dynamic threshold band and the future prediction interval, the actual trajectory and the minimum reversible intervention hypothesis trajectory are simultaneously deduced. A limit violation is determined only when both trajectories simultaneously satisfy the following conditions within a duration T: the point prediction crosses the dynamic threshold band and the main body of the future prediction interval continuously crosses the threshold band. The minimum reversible intervention hypothesis trajectory includes: a time valve position fine-tuning of no more than 2% or a short pulse flush for 3–10 seconds.

[0103] Consistency confirmation for boundary violations is achieved using a k-of-n consensus rule across multiple branches controlled by isothermal reliable gating. (Before confirming a boundary violation and deciding whether to escalate: each channel with voting rights is independently subjected to a double threshold + counterfactual judgment; channels judged as "boundary violation established" are added to the voting set; if k-of-n is satisfied within a 30–60s consensus window (e.g., n=5, k=3), then sufficient consistency is determined → escalation is allowed; otherwise, only the minimum reversible action is prompted or observation continues, and escalation is not performed.)

[0104] After a limit violation is confirmed and the alarm level is upgraded, a cooling-off period begins, during which repeated upgrades are prohibited.

[0105] The alarm level is determined by the interval width and coverage score; when the interval width is large or the coverage score is low, the level is downgraded and the duration is extended, while the safety redundancy is increased.

[0106] The alarm level, the set of allowed actions, and the suggested duration are output to the handling module. The set of allowed actions is the subset of actions (and their parameter boundaries) that are allowed to be executed after the system filters them layer by layer according to evidence → safety → uncertainty → benefit in a single out-of-limit event, and is available for the handling module to call. The action library typically includes: valve position fine-tuning, short pulse flushing, quick discharge, bypass, and isolation.

[0107] Allowing algorithmic processing of action sets (six-step method):

[0108] enter:

[0109] Unified state vector; point prediction and future prediction interval; interval width W; coverage score Score_cov; dynamic threshold band / anchor point; gate confidence GateConf; ΔP–Q health; consensus result (k-of-n); resource limit and security shield constraint.

[0110] Step 0 | Trigger Pre-processing (Whether to enter the filter):

[0111] It is judged as an upgradable overlimit and enters the action screening only when both the actual track and the minimum reversible intervention track meet the "double threshold + consensus (k-of-n, 30–60 s)".

[0112] If the minimum reversible intervention track does not meet the double threshold (while the actual track does), only the "minimum reversible actions" (fine-tuning ≤ 2%, short-pulse flushing for 3–10 s) are opened, and no further amplification of actions is performed.

[0113] Step 1 | Causal attribution gate (aligning actions with problem types):

[0114] Hydraulic side dominant: ΔP–Q deviation, ΔP high-frequency energy ↑, |dQ / dt|↑, valve step → turbidity synchronous upsurge, and conservation deviation ≈ 0.

[0115] → Candidates only include: short-pulse flushing, valve position fine-tuning, fast discharge (bypass / isolation not released temporarily).

[0116] Chemical side dominant: significant conservation deviation, σ25 increase, while ΔP–Q is healthy, dNTU / dt ≈ 0.

[0117] → Candidate order: fine-tuning → flushing (confirmation) → bypass → isolation.

[0118] Mixed / unknown: First retain fine-tuning / flushing and shorten the review window (30–60 s).

[0119] Step 2 | Safety shield projection (hard constraint filtering):

[0120] Check each candidate one by one: temperature / pressure drop boundary, total cooling capacity, valve position change rate ≤ 5% / s, minimum residence time of 30–120 s, upper limit of switching times, branch phase stagger ≥ 1–2 s. If not passed, it is excluded or automatically downgraded to a lighter action.

[0121] Step 3 | Uncertainty gate (determining the degree of conservatism using W and Score_cov):

[0122] W ≥ p90 of the healthy period or Score_cov deviation > 5% → Only fine-tuning / review / flushing are allowed; bypass / isolation are prohibited; duration and residence time are increased.

[0123] p50 ≤ W < p90 and Score_cov is normal → Flushing is opened; bypass requires passing the next threshold.

[0124] W < p50 and the double threshold is stable → Bypass / isolation can be opened (still need to pass the next step).

[0125] Step 4 | Value-evidence double threshold (actions must be "cost-effective and effective"):

[0126] For each candidate, two things are evaluated simultaneously:

[0127] Risk difference ΔV≥V_min: The overall cost that can be reduced relative to "no action" (outbound risk × duration, exhaust, energy consumption, recovery time, switching fatigue).

[0128] Evidence gain ΔE≥E_min: The expected amount of narrowing or conservation deviation of W after the action reaches the threshold.

[0129] Only candidates that meet both conditions are included in the allowed set; otherwise, they are downgraded or eliminated.

[0130] Typical thresholds: V_min is 1× of the cost baseline within the station; E_min is "10–20% narrowing of W" or "conservation deviation falling back ≥30%".

[0131] Step 5 | Resources and Concurrency (Including "Can it be done simultaneously"):

[0132] If cooling / bypass capabilities are limited: sort by "unit evidence mitigation amount / cost" and take the top K entries; and execute them in phases staggered from other branches to prevent crosstalk.

[0133] Output: Allowed action set = {subset of actions retained}. (Window, cooldown requirements, safety limits). Degradation / upgrade path (invalid → next tier; successful → narrow threshold band and rollback bias).

[0134] S4: After processing, verify the difference between the expected and the actual results in the verification window, and write the results back to adaptively update the gating threshold, feature weights, interval calibration and policy parameters.

[0135] For candidate actions in the set of permitted actions, the minimum alternative intervention is selected based on the risk difference criterion and the evidence gain threshold to obtain the treatment action;

[0136] After the action is executed, the verification window is opened to perform expected-measured verification of stabilization time, temperature difference between outlet and measuring pool, ΔP–Q health, prediction interval width and conservation deviation. If successful, the dynamic threshold band is narrowed and the threshold anchor point bias is withdrawn, the relevant feature weights are increased and the normal calibration radius is restored. If unsuccessful, the calibration radius is expanded or the conservatism is increased, the relevant feature weights are reduced and the action strategy parameters are adjusted.

[0137] The strategy parameters include, but are not limited to: valve position single-step amplitude and cycle time, minimum dwell time, cooling period, short pulse flushing duration and number of cycles, bypass duration limit and concurrent number.

[0138] On the other hand, this embodiment also provides an online monitoring system for grid heaters and unit condensate, which includes: heat, characterized in that: a data acquisition unit, which performs sampling and cooling conditioning on multiple sampling branches, generates isothermal reliable gating flags and allocates heat exchange resources.

[0139] The analysis unit, based on the data passed through the gating system, combines physical consistency relationships with causal invariance to establish a state representation, thereby obtaining the prediction interval and anomaly probability of the sub-bar.

[0140] The alarm unit uses probability and interval to drive the dynamic threshold with dual thresholds, outputting hierarchical alarms and limiting the set of executable actions.

[0141] The update unit verifies the difference between the expected and actual values ​​in the verification window after processing, and writes the results back to adaptively update the gating threshold, feature weights, interval calibration, and policy parameters.

[0142] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0143] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0144] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0145] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0146] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for online monitoring of the condensate drain of a heating network heater and unit, characterized in that, include: Sampling and cooling conditioning are performed on multi-sampling branches to generate isothermal reliable gating flags and allocate heat exchange resources. Based on the data passed through the gating system, a state representation is established by combining physical consistency and causal invariance, and the prediction interval and anomaly probability of the sub-bar are obtained. By using probability and interval to drive dynamic thresholds, a tiered alarm is output and the set of executable actions is limited. The process of generating the set of executable actions includes: for the triggered event, using the safety shield to constrain the boundary and timing, and using the risk difference criterion to select the best option among valve position fine-tuning, short pulse flushing, bypassing or isolation and generating a safety certificate; After processing, the difference between the expected and the actual values ​​is verified in the verification window, and the results are written back to adaptively update the gating threshold, feature weights, interval calibration and policy parameters.

2. The online monitoring method for heat network heaters and unit condensate drains as described in claim 1, characterized in that: The sampling and cooling conditioning include five locations for the unit's condensate drain and the heating network heater's condensate drain. The condensate samples are cooled to 25℃±1℃ by a cooler. Online monitoring equipment is installed in the condensate drains to detect changes in water quality and make adjustments at any time. The drainage system includes:

1. Drainage from the heating network heater, 2. Drainage from the heating network heater, 1 and 2. Drainage from the exhaust cooler, 3. Drainage from the exhaust cooler, and 3 and 4. Boiler drainage. Sampling pipelines are connected to the outlet headers of each drainage pump, passing through the cooler and then connected to the online instruments.

3. The online monitoring method for heat network heaters and unit condensate drains as described in claim 2, characterized in that: The isothermal reliable gating flag and heat exchange resource allocation include: Within each sampling branch, the cooler outlet temperature and the measurement pool temperature are simultaneously monitored within a 5–10 s steady-state judgment window. The sample water is required to be stable at 25℃±1℃ and the temperature difference between the outlet and the measurement pool is not greater than 0.5℃. The stabilization time and overshoot count are recorded as the hot steady-state temperature score. Simultaneously, the sampling flow rate and the pressure difference before and after the throttling device are compared for consistency. The flow rate fluctuation and pulsation are limited to not exceeding the set threshold and the ΔP–Q relationship is in the healthy zone, forming the hydraulic steady-state flow score. The hot steady-state temperature score and the hydraulic steady-state flow score are summarized into a gating score according to the preset weight and a confidence level is given. When the score is not lower than the threshold, an isothermal reliable gating flag is output. If the score does not meet the threshold, the branch is temporarily suspended from entering the subsequent judgment and only the minimum cooling guarantee is retained. The ΔP–Q relationship represents the characteristic curve of pressure difference as a function of flow rate; Under the constraints of total cooling capacity and cooler pressure drop limit, a rolling priority queue is constructed based on gate confidence, predicted settling time, residual temperature difference and branch importance. The quota increment of cooling valve position or bypass ratio is issued to priority branches in 5-10s time slices, and the coupled branches are subjected to 1-2s phase stagger and decreasing step to suppress crosstalk. When the lower layer is executed, the valve position change rate is not greater than 5% / s, the minimum dwell time for a single target is 30–120s, and the branch pressure drop does not exceed the upper limit constraint. At the end of each time slice, the temperature difference drops and stabilizes. Two consecutive slices without improvement are automatically downgraded and their quota is reclaimed for other non-compliant branches. When a branch meets the gating threshold, it maintains minimum residence and gradually reduces its quota, and the released heat exchange capacity is recovered and redistributed. The gating flag, confidence level, allocation result, and stabilization record are used together as weight inputs for subsequent state modeling, prediction interval, and dynamic threshold setting to ensure that the data entering alarm triggering and handling decision-making are comparable and stable.

4. The online monitoring method for heat network heaters and unit condensate drains as described in claim 3, characterized in that: After the isothermal reliable gating flag is passed, a unified state representation is established: Temperature-compensated conductivity calculations were performed on the sample water data after temperature and flow stabilization, and conservation deviations were formed based on the water chemistry conservation law and the steam condensation and dilution law. The hydraulic consistency deviation of ΔP–Q along the temperature residual and the stabilization time of the same branch are used as hydraulic evidence. The operating condition domains are divided according to unit load, start-up and shutdown and season. The stability of candidate features in multiple domains is calculated. Only the subset of features with stability above the threshold is retained. Unstable elements are downweighted or isolated. The conservation bias, hydraulic consistency bias, cross-domain stability, as well as the gate confidence, stabilization time, and residual temperature difference are fused into a unified state vector after being aligned by time, and the operating condition domain label and the main evidence of anomalies are output. The unified state vector serves as the direct input to the prediction interval and triggering criterion.

5. The online monitoring method for heat network heaters and unit condensate drains as described in claim 4, characterized in that: The generation of the prediction interval and self-anchoring dynamic threshold of the sub-bulb rod includes: obtaining point predictions for compensated conductivity, turbidity, and hydraulic parameters within a set prediction window; The gating is performed by dividing the samples into bins according to the working condition domain for interval calibration, so that the future intervals meet the coverage target within the significance range of 0.05 to 0.10, and the interval width is output to characterize the uncertainty. A drift sensor is set up to score the distance and coverage between the predicted distribution of points and the historical calibration distribution of each domain, and to monitor the offset. When any one exceeds the limit, the calibration radius is automatically expanded or the conservatism is increased. The baseline median of the historical health period and the median of the recent 24-hour interval are jointly used to determine the threshold anchor point, and the dynamic threshold band is formed by adaptive biasing according to the coverage score, gating confidence and ΔP–Q health. Finally, the future interval, interval width, dynamic threshold band, and anchor point are output together with the unified state vector for subsequent hierarchical triggering and limitation of the set of allowed actions.

6. The online monitoring method for heat network heaters and unit condensate drains as described in claim 5, characterized in that: After obtaining the dynamic threshold band and the future prediction interval, the actual trajectory and the minimum reversible intervention hypothesis trajectory are simultaneously deduced. The limit is determined to be exceeded only when the two trajectories simultaneously satisfy the following within the duration T: the point prediction crosses the dynamic threshold band and the main body of the future prediction interval continuously crosses the threshold band. Consistency confirmation for exceeding limits is achieved by using k-of-n consensus rules on multiple branches controlled by isothermal trusted gating. After a limit violation is confirmed and the alarm level is upgraded, a cooling-off period begins, during which repeated upgrades are prohibited. Alarm level is determined by a score based on interval width and coverage. The alarm level, the set of allowed actions, and the suggested duration are output to the handling module. The minimum reversible intervention assumption includes that the valve position is fine-tuned by no more than 2% or that a short pulse flush lasts for 3 to 10 seconds.

7. The online monitoring method for heat network heaters and unit condensate drains as described in claim 6, characterized in that: For candidate actions in the set of permitted actions, the minimum alternative intervention is selected based on the risk difference criterion and the evidence gain threshold to obtain the treatment action; After the action is executed, the verification window is opened to perform expected-measured verification of stabilization time, temperature difference between outlet and measuring pool, ΔP–Q health, prediction interval width and conservation deviation. If successful, the dynamic threshold band is narrowed and the threshold anchor point bias is withdrawn, the relevant feature weights are increased and the normal calibration radius is restored. If unsuccessful, the calibration radius is expanded or the conservatism is increased, the relevant feature weights are reduced and the action strategy parameters are adjusted. The strategy parameters include, but are not limited to: valve position single-step amplitude and cycle time, minimum dwell time, cooling period, short pulse flushing duration and number of cycles, bypass duration limit and concurrent number.

8. A system for online monitoring of heat network heaters and unit condensate drains using the method described in any one of claims 1-7, characterized in that: The data acquisition unit performs sampling and cooling conditioning on multiple sampling branches, generates isothermal reliable gating flags, and allocates heat exchange resources. The analysis unit, based on the data passed through the gating system, combines physical consistency and causal invariance to establish a state representation, thereby obtaining the prediction interval and anomaly probability of the sub-bar. The alarm unit uses a dual threshold driven by probability and interval to output hierarchical alarms and limit the set of executable actions. The update unit verifies the difference between the expected and actual values ​​in the verification window after processing, and writes the results back to adaptively update the gating threshold, feature weights, interval calibration, and policy parameters.

9. A computer device, comprising: A memory and a processor; the memory stores a computer program, characterized in that: when the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images