A radar-differential pressure dual-mode liquid level signal intelligent fusion method for chemical storage tank

Abstract

The application provides a radar-differential pressure dual-mode liquid level signal intelligent fusion method for a chemical storage tank, and belongs to the technical field of chemical storage tank liquid level detection and intelligent signal processing. The method comprises the following steps: synchronously acquiring original liquid level signals of a radar and a differential pressure sensor; based on real-time cross-correlation intensity and trend consistency of the synchronous signal pair, deciding a current operation mode; in a cooperative mutual verification mode, taking a weighted median of two signals in a sliding time window as a common reference, and performing constraint approximation correction of respective signals to the reference; and in a master-slave reconstruction mode, taking a reliable signal verified through physical pre-checking as a leading signal, and utilizing material mass balance relationship of the storage tank. Through full-process verification and fusion driven by physical constraints, the application solves the measurement distortion problem caused by sensor interference, drift and failure under complex working conditions, and realizes high-reliability, high-precision and high-safety continuous output of the liquid level signal.

Description

Technical Field

[0001] This invention belongs to the field of liquid level detection and intelligent signal processing technology for chemical storage tanks, specifically relating to an intelligent fusion method for radar-differential pressure dual-mode liquid level signals used in chemical storage tanks. Background Technology

[0002] Liquid level in chemical storage tanks is a core monitoring parameter for safe production and process control in industries such as petroleum and chemicals. The complex operating conditions inside these tanks, including high temperature, high pressure, corrosion, fluctuations in medium density, and interference from foam vapor, place extremely high demands on the accuracy, reliability, and stability of liquid level measurements. Currently, industrial sites primarily employ two main solutions: radar level gauges and differential pressure transmitters. Radar level gauges offer non-contact measurement and fast response, but are susceptible to signal distortion due to tank wall adhesion, antenna contamination, and vapor interference. Differential pressure transmitters, based on the pressure-level relationship, are structurally reliable but sensitive to changes in medium density and prone to measurement deviations due to pressure tap blockage and temperature drift.

[0003] To improve measurement reliability, dual-instrument redundancy configurations are commonly used on site. However, existing liquid level signal fusion technologies are mostly traditional methods such as simple threshold comparison and fixed weighted averaging. They do not combine the physical constraints of the storage tank process to realize signal validity verification, lack dynamic operating condition adaptation capabilities and fault self-healing mechanisms. When the sensor is abnormal or the operating conditions fluctuate, problems such as measurement distortion and output interruption are likely to occur, making it difficult to meet the high-precision, high-safety and high-reliability continuous monitoring requirements of chemical storage tanks. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent fusion method for radar-differential pressure dual-mode liquid level signals for chemical storage tanks, so as to solve the technical problems mentioned in the background art. The specific steps are as follows:

[0005] S1. Synchronous acquisition and physical pre-verification: The original liquid level signals from the radar and differential pressure sensor are acquired synchronously, and the instantaneous rate of change and static offset of the two signals are physically verified according to the tank process parameters. The synchronous signal pair with anomaly marker is output.

[0006] S2. Dynamic decision-making for operating mode: Based on the real-time cross-correlation strength and trend consistency of the synchronization signal pair, the current operating mode is determined; when the two signals are highly correlated and consistent, the "cooperative verification mode" is entered; otherwise, the "master-slave reconstruction mode" is entered.

[0007] S3. Collaborative verification mode fusion: In collaborative verification mode, the weighted median of the two signals within the sliding time window is used as a common benchmark. Each signal is then constrained to approximate the benchmark. Subsequently, the two corrected signals are averaged with equal weights to generate the first type of fusion value.

[0008] S4. Master-Slave Reconstruction Mode Integration: In the master-slave reconstruction mode, a reliable signal that has passed physical pre-verification is used as the dominant signal. The material mass balance relationship of the storage tank is used to combine the dominant signal and reverse the deduction in real time to generate a virtual reconstruction value of another signal, so that the reconstruction value is physically consistent with the dominant signal.

[0009] S5. Result generation and self-verification: Based on the operating mode decided in step S2, output the first type of fusion value in step S3 or the dominant signal value in step S4. Substitute the output value into the physical limit verification and quality balance relationship for reverse verification. If the verification fails, trigger a fault alarm and lock the output.

[0010] Furthermore, the physical limit verification in step S1 specifically includes:

[0011] Based on the safe operating liquid level of the storage tank, the maximum rated flow rate of the feed / discharge pump, and the safe density range of the material, the theoretical maximum instantaneous rate of change threshold and the allowable static offset threshold of the liquid level change are calculated.

[0012] The real-time liquid level signals collected by the radar and differential pressure sensor are compared with the liquid level value at the previous moment, and their instantaneous rate of change is calculated.

[0013] The real-time liquid level signal is compared with the historical average liquid level value of the storage tank in a static state, and the static offset of each signal is calculated.

[0014] When the instantaneous rate of change of any signal continuously exceeds the theoretical maximum instantaneous rate of change threshold, or its static offset continuously exceeds the allowable static offset threshold, the signal is marked as "abnormal"; otherwise, it is marked as "normal".

[0015] Furthermore, the "dynamic decision-making of operating mode" is achieved through real-time joint analysis of the cross-correlation strength and trend consistency between signals. The decision-making logic is as follows: the absolute value of the Pearson correlation coefficient of the synchronous signal pair within the sliding time window is calculated as the cross-correlation strength, and the average absolute error of the first-order difference sequence of the two signals is calculated as the trend consistency measure. During the decision-making process, the cross-correlation strength is compared with the preset correlation judgment threshold, and the trend consistency is compared with the preset consistency judgment threshold. If and only if the cross-correlation strength is higher than its judgment threshold and the trend consistency is lower than its judgment threshold, the system determines that the two signals are in a highly coordinated state and enters the "coordinated verification mode". Otherwise, based on the signal identifier output by the physical pre-verification in step S1, the signal currently marked as "normal" is selected as the dominant signal source, and the decision-making system enters the "master-slave reconstruction mode".

[0016] Furthermore, the calculation of the cross-correlation strength is as follows: within a preset sliding time window, the Pearson correlation coefficient of the radar and differential pressure synchronous liquid level signal sequences is calculated, and the absolute value of the coefficient is used as the real-time cross-correlation strength. The calculation of the trend consistency is as follows: within the same time window, the first-order difference sequences of the two signal sequences are calculated respectively, and the average absolute error value of the two difference sequences is calculated, which is used as a quantitative measure of trend consistency.

[0017] Furthermore, the "collaborative verification mode fusion" is achieved through two stages: benchmark constraint correction and equal-weight fusion. First, the arithmetic mean of the two signals at each moment within the sliding time window is used as the benchmark to form a benchmark sequence. Then, a constraint approximation correction is performed on the radar and differential pressure signal sequences at each moment within the window: if the absolute value of the deviation from the benchmark value is not greater than the preset maximum allowable approximation error, the original value is retained; if the deviation exceeds the limit, the signal value at that moment is corrected to the benchmark value plus or minus the boundary value of the maximum allowable approximation error, thereby ensuring that the deviation between the corrected signal and the benchmark is within the preset range. Finally, the two corrected signal sequences are subjected to an equal-weighted arithmetic mean at the corresponding moment to output the first type of fusion value.

[0018] Furthermore, in the constrained approximation correction process, the determination and optimization of the maximum allowable approximation error (δ) includes:

[0019] Reference setting: Based on the steady-state measurement errors of the radar and differential pressure sensor in the cooperative working mode, the larger of the two is taken as the initial reference value of the maximum allowable approximation error;

[0020] Dynamic optimization: During the operation of the collaborative verification mode, the maximum allowable approximation error is adaptively and dynamically adjusted based on the real-time statistical characteristics of the difference between the radar signal correction value and the differential pressure signal correction value.

[0021] Furthermore, step S4 is a virtual signal generation and adaptive fusion method based on a mass conservation physics engine. Its core lies in using a verified reliable signal as the dominant signal. Based on the law of conservation of mass of the storage tank, it generates a physically consistent virtual associated signal in real time. Final output value Determined by the following adaptive model:

[0022]

[0023] in, For the physical model mismatch degree, the function For based on The method uses a monotonically increasing adjustment function to address the consistency error of the physical model. Dynamic driving enables a smooth, adaptive transition from "completely trusting the sensor" to "gradually adopting physical model predictions," thereby ensuring the physical authenticity and reliability of the output value even when the signal source may degrade.

[0024] Furthermore, step S5 is an output self-checking and fault isolation mechanism based on physical model reverse verification. The fusion result generated by S3 or S4 is used as the value to be verified, and it is substituted into the physical limit verification of S1 and the mass conservation relationship of S4 for reverse consistency verification. If the verification passes, the value is output; if the verification fails, a high-level fault alarm is immediately triggered, and the system output is locked at the most recently verified safe value, so as to achieve rigid maintenance of safe output under fault conditions.

[0025] Compared with the relevant prior art, the present application has the following main technical advantages:

[0026] This invention addresses the core pain points of existing chemical storage tank level monitoring systems, such as sensors being susceptible to operating condition interference, lack of physical verification in fusion methods, rigid modes, and unreliable output under fault conditions. It constructs a fully intelligent fusion system driven by physical constraints. By converting the tank process parameters into quantifiable instantaneous change rates and static offset thresholds, it performs pre-emptive physical verification on radar and differential pressure dual-mode signals, eliminating abnormal data at the source. This completely solves the problems of traditional fusion methods lacking rationality verification and causing measurement distortion due to erroneous signals directly participating in calculations, thus laying a solid foundation for high-precision output.

[0027] This invention achieves dynamic mode adaptive switching based on signal correlation and trend consistency. When both signals are normal, a benchmark constraint correction is used to complete high-precision smooth fusion. When a single sensor fails, a virtual signal reconstruction and self-healing output are achieved by relying on a physical model of material mass conservation. This invention completely breaks through the rigid limitations of fixed weighting and simple two-out-of-two in traditional redundant systems, and greatly improves the measurement accuracy, stability and robustness under complex working conditions.

[0028] This invention pioneers a dual reverse self-verification mechanism that integrates the physical limits of the results and mass balance. In case of anomalies, it immediately triggers an alarm and locks the safety output, achieving a closed-loop protection of fault self-inspection, isolation, and safety maintenance. At the same time, it supports dynamic parameter optimization to adapt to field changes such as density fluctuations and sensor aging, effectively avoiding safety risks such as false level alarms, tank overflows, and cavitation. It significantly improves the functional safety and long-term operational reliability of chemical storage tank level monitoring, perfectly adapting to the stringent monitoring requirements of high-risk scenarios such as petroleum and chemical industries. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating the present invention. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments. However, the present invention can be implemented in many different ways and should not be construed as limited to the embodiments shown; rather, these embodiments provide those skilled in the art with implementation methods that meet applicable legal requirements.

[0031] Please refer to Figure 1 This invention provides an intelligent fusion method for radar-differential pressure dual-mode liquid level signals for chemical storage tanks, characterized by the following steps:

[0032] S1. Synchronous acquisition and physical pre-verification: The original liquid level signals from the radar and differential pressure sensor are acquired synchronously, and the instantaneous rate of change and static offset of the two signals are physically verified according to the tank process parameters. The synchronous signal pair with anomaly marker is output.

[0033] Physical limit verification specifically includes:

[0034] Based on the safe operating liquid level of the storage tank, the maximum rated flow rate of the feed / discharge pump, and the safe density range of the material, the theoretical maximum instantaneous rate of change threshold and the allowable static offset threshold of the liquid level change are calculated.

[0035] The real-time liquid level signals collected by the radar and differential pressure sensor are compared with the liquid level value at the previous moment, and their instantaneous rate of change is calculated.

[0036] The real-time liquid level signal is compared with the historical average liquid level value of the storage tank in a static state, and the static offset of each signal is calculated.

[0037] When the instantaneous rate of change of any signal continuously exceeds the theoretical maximum instantaneous rate of change threshold, or its static offset continuously exceeds the allowable static offset threshold, the signal is marked as "abnormal"; otherwise, it is marked as "normal".

[0038] This invention solves the measurement distortion problem caused by sensor interference, drift, and failure under complex working conditions by using physical constraint-driven full-process verification and fusion, and achieves high reliability, high precision, and high safety continuous output of liquid level signals.

[0039] During the system initialization phase, a unified industrial clock source is used as a reference to synchronously trigger the signal acquisition actions of the radar level gauge and differential pressure transmitter. This ensures that the two original level signals are strictly aligned in the time dimension, eliminating the impact of timing deviations on subsequent fusion calculations and forming a set of original level signals with consistent timing. Based on the process design parameters of the target storage tank, including the safe operating level height range, the maximum rated operating flow rate of the feed pump and discharge pump, and the safe density variation range of the stored medium in the tank, combined with the geometric parameters of the storage tank, the theoretical maximum instantaneous rate of change threshold and the allowable static offset threshold of the level change are derived and calculated through process physical constraints. These thresholds are entirely determined by the actual process conditions of the storage tank without any manual experience tuning, serving as a rigid physical benchmark for determining the validity of the signal.

[0040] After threshold calibration, physical limit verification is performed on the synchronously acquired radar and differential pressure raw liquid level signals. For the radar sensor signal, the real-time liquid level value acquired at the current moment is compared point-by-point with the effective liquid level value at the previous moment to calculate the instantaneous rate of change of the signal, which is used to determine whether the liquid level change rate exceeds the physical limit achievable by the process. At the same time, the real-time liquid level value is compared with the historical average liquid level reference value of the storage tank under static and stable conditions to calculate the static offset of the signal, which is used to identify slow-changing anomalies such as long-term sensor drift and zero-point offset. For the differential pressure sensor signal, the same verification logic is executed: the real-time liquid level value at the current moment is compared with the effective liquid level value at the previous moment to calculate the instantaneous rate of change of the differential pressure signal; the real-time liquid level value is compared with the historical average liquid level reference value of the storage tank under static conditions to calculate the static offset of the differential pressure signal, realizing independent and synchronous evaluation of the change rate and offset status of the two signals.

[0041] Based on the calculated instantaneous rate of change and static offset, validity determination and anomaly marking are performed on the two signals respectively. For radar sensor signals, if the instantaneous rate of change exceeds the theoretical maximum instantaneous rate of change threshold for multiple consecutive sampling periods, or if the static offset exceeds the allowable static offset threshold for multiple consecutive sampling periods, the signal is determined to be abnormal and marked.

[0042] Based on the calculated instantaneous rate of change and static offset, validity determination and anomaly marking are performed on the two signals respectively. For radar sensor signals, if the instantaneous rate of change exceeds the theoretical maximum instantaneous rate of change threshold for multiple consecutive sampling periods, or if the static offset exceeds the allowable static offset threshold for multiple consecutive sampling periods, the signal is determined to be abnormal and marked with an "abnormal" label. If neither indicator exceeds the corresponding threshold, the signal is determined to be normal and marked with a "normal" label. For differential pressure sensor signals, the same determination logic is performed: if the instantaneous rate of change exceeds the standard continuously or the static offset exceeds the standard continuously, an "abnormal" label is marked; otherwise, a "normal" label is marked. After completing the validity determination and labeling of the two signals, the abnormal / normal label information is encapsulated and integrated with the synchronously acquired original liquid level signal to output a synchronous signal pair with anomaly marking. This signal pair serves as a reliable input for subsequent operation mode decisions, signal fusion, and safety verification, eliminating invalid and erroneous data from the source and laying a physically reliable foundation for the full-process fusion calculation.

[0043] Beneficial effects: This embodiment establishes a rigid validity judgment standard from the signal acquisition end through a physical constraint pre-verification mechanism, which completely avoids the core defect of abnormal signals directly participating in the calculation in traditional fusion methods. It can operate stably under complex working conditions such as high temperature, high pressure, corrosion, medium density fluctuation, and foam vapor interference in the tank, and meets the stringent requirements for safety monitoring, metering and accounting and functional safety of chemical storage tanks.

[0044] S2. Dynamic Decision-Making of Operating Mode: Based on the real-time cross-correlation strength and trend consistency of the synchronization signal pair, the current operating mode is determined; when the two signals are highly correlated and consistent, the "cooperative verification mode" is entered; otherwise, the "master-slave reconstruction mode" is entered.

[0045] The "Dynamic Decision-Making of Operating Mode" is achieved through real-time joint analysis of the cross-correlation strength and trend consistency between signals. Specifically, the decision-making logic is as follows: the absolute value of the Pearson correlation coefficient of the synchronized signal pair within the sliding time window is calculated as the cross-correlation strength, and the average absolute error of the first-order difference sequences of the two signals is calculated as the trend consistency measure. During decision-making, the cross-correlation strength is compared with a preset correlation threshold, and the trend consistency is compared with a preset consistency threshold. If and only if the cross-correlation strength is higher than its threshold and the trend consistency is lower than its threshold, the system determines that the two signals are in a highly coordinated state and enters the "Coordinated Verification Mode." Otherwise, based on the signal identifier output by the physical pre-verification in step S1, the signal currently marked as "normal" is selected as the dominant signal source, and the decision-making system enters the "Master-Slave Reconstruction Mode."

[0046] The calculation of cross-correlation strength is as follows: within a preset sliding time window, the Pearson correlation coefficient of the radar and differential pressure synchronous liquid level signal sequences is calculated, and the absolute value of the coefficient is used as the real-time cross-correlation strength. The calculation of trend consistency is as follows: within the same time window, the first-order difference sequences of the two signal sequences are calculated respectively, and the average absolute error value of the two difference sequences is calculated, which is used as the quantitative measure of trend consistency.

[0047] After receiving the synchronization signal pair with an anomaly mark output in step S1, the system enters the dynamic decision-making stage of the operating mode. In this stage, the system performs real-time quantitative analysis of the coordination degree of the two signals to achieve accurate determination and automatic switching of the operating mode.

[0048] Within the system's preset sliding time window, correlation calculations are performed on the synchronous liquid level signal sequences output by the radar sensor and the differential pressure sensor to obtain the Pearson correlation coefficient between the two signals. The absolute value of this coefficient is used as the real-time cross-correlation strength to intuitively reflect the degree of correlation between the two liquid level signals in terms of numerical changes. Within the same sliding time window, first-order difference processing is performed on the radar liquid level signal sequence and the differential pressure liquid level signal sequence respectively to obtain a difference sequence that can characterize the real-time trend of signal changes. Then, the mean absolute error is calculated on the two sets of difference sequences, and the calculation result is used as a quantitative measure of trend consistency to objectively reflect the degree of matching between the two signals in terms of change trends.

[0049] The system compares the real-time cross-correlation strength with the preset correlation judgment threshold, and at the same time compares the trend consistency measurement result with the preset consistency judgment threshold. The final decision on the operating mode is made through the joint judgment of the two indicators.

[0050] When the cross-correlation strength of the two signals is higher than the correlation threshold and the trend consistency is lower than the consistency threshold, the system determines that the two signals are in a highly coordinated state and automatically enters the coordinated verification mode, providing a mode basis for subsequent high-precision coordinated fusion processing. When the two signals do not meet the high coordination judgment condition, the system selects the signal currently marked as normal as the dominant signal source based on the signal identification information output by the physical pre-verification in step S1, and automatically decides to enter the master-slave reconstruction mode to ensure that stable and reliable liquid level results can still be output under the condition of abnormal or degraded single signal.

[0051] Beneficial effects: By jointly analyzing the real-time cross-correlation strength and trend consistency, dynamic decision-making of the operating mode is achieved. It can accurately identify the cooperative state of two liquid level signals, ensuring stable and reliable mode judgment. It can automatically switch between cooperative verification mode and master-slave reconstruction mode according to the actual working condition of the signal. It can give full play to the advantages of dual-mode redundancy fusion when both signals are working normally, and can automatically select a reliable signal that has been physically pre-verified as the dominant signal source when a single signal is abnormal or degraded. It can adapt to complex working conditions and sensor state changes without manual intervention. It fundamentally solves the problems of rigid mode, slow switching, single judgment and poor fault tolerance of traditional redundant monitoring systems. It significantly improves the adaptive ability, anti-interference ability and operational stability of the entire liquid level fusion system, and provides a solid guarantee for continuous, reliable and high-precision monitoring of liquid level in chemical storage tanks.

[0052] S3. Collaborative verification mode fusion: In the collaborative verification mode, the weighted median of the two signals within the sliding time window is used as a common benchmark. Each signal is then constrained to approximate the benchmark. Subsequently, the two corrected signals are averaged with equal weights to generate the first type of fusion value.

[0053] The "collaborative verification mode fusion" is achieved through two stages: benchmark constraint correction and equal-weight fusion. First, the arithmetic mean of the two signals at each moment within the sliding time window is used as the benchmark to form a benchmark sequence. Then, a constraint approximation correction is performed on the radar and differential pressure signal sequences at each moment within the window: if the absolute value of the deviation from the benchmark value is not greater than the preset maximum allowable approximation error, the original value is retained; if the deviation exceeds the limit, the signal value at that moment is corrected to the benchmark value plus or minus the boundary value of the maximum allowable approximation error, thereby ensuring that the deviation between the corrected signal and the benchmark is within the preset range. Finally, the two corrected signal sequences are subjected to an equal-weighted arithmetic mean at the corresponding moment, and the first type of fusion value is output.

[0054] In the constrained approximation correction process, the determination and optimization of the maximum allowable approximation error (δ) includes:

[0055] Reference setting: Based on the steady-state measurement errors of the radar and differential pressure sensor in the cooperative working mode, the larger of the two is taken as the initial reference value of the maximum allowable approximation error;

[0056] Dynamic optimization: During the operation of the collaborative verification mode, the maximum allowable approximation error is adaptively and dynamically adjusted based on the real-time statistical characteristics of the difference between the radar signal correction value and the differential pressure signal correction value.

[0057] After the system enters the collaborative verification mode, it first performs point-by-point synchronous acquisition and time-series alignment processing on the radar measurement signal and differential pressure measurement signal within a preset sliding time window. This ensures that the two signals maintain strict consistency in the time dimension, providing a unified data foundation for subsequent collaborative analysis and mutual verification. After the time-series alignment is completed, the system constructs a common reference sequence within the sliding window based on the steady-state measurement characteristics and engineering application constraints of the two signals. This reference sequence is based on the statistical average value of the two signals and is smoothed and optimized in combination with the sensor's own measurement accuracy indicators to form a continuous, stable, and practically accurate reference standard.

[0058] After the common reference sequence is constructed, the system performs constrained approximation correction operations on the radar signal and differential pressure signal respectively. For each sampling time, the system calculates the real-time deviation between the radar signal and the common reference, and the real-time deviation between the differential pressure signal and the common reference, respectively. The two deviations are compared with the preset maximum allowable approximation error. If the deviation between the signal value and the common reference is within the maximum allowable approximation error range, the signal is determined to be in the normal operating range, and the signal value at the current time remains unchanged. If the deviation between the signal value and the common reference exceeds the maximum allowable approximation error range, the signal is determined to have abnormal fluctuations, and boundary constraint correction is performed on the signal value at the current time to forcibly adjust it to the legal range formed by the common reference and the maximum allowable approximation error, so that the two signals always remain within a stable, reasonable, and reliable value range.

[0059] During the constraint approximation correction process, the system dynamically and adaptively adjusts the maximum permissible approximation error to ensure the flexibility and applicability of the correction rules. The initial value of the maximum permissible approximation error is based on the maximum measurement error of the radar sensor and the differential pressure sensor under steady-state operating conditions. During continuous system operation, the system statistically analyzes the difference distribution characteristics between the two corrected signals in real time. Based on indicators such as the root mean square error, extreme values, and trends of the differences, the maximum permissible approximation error is periodically updated, enabling the correction constraints to automatically adapt to actual operating conditions such as drift, aging, and environmental interference during long-term sensor operation, thereby improving the stability and reliability of the correction process.

[0060] After completing the constrained approximation correction and dynamic parameter adjustment of the two signals, the system performs equal-weighted fusion calculation on the corrected radar signal and the corrected differential pressure signal at the same time. The arithmetic mean of the two signals is used as the stable output of the collaborative verification fusion result at the current time. This fusion result is verified by triple constraints of common benchmark constraints, deviation correction constraints and dynamic adaptive constraints. It can effectively suppress the data distortion caused by random noise, abnormal jumps and local faults of single signals, and provide continuous, stable, accurate and reliable fusion data support for subsequent data verification, status judgment and system decision-making.

[0061] Beneficial Effects: By constructing a complete collaborative fusion processing system encompassing time-series synchronous acquisition, common reference sequence constraints, constrained approximation correction, and dynamic adaptive error adjustment, high-precision fusion output can be achieved when both signals are in normal collaborative working condition. This effectively suppresses problems such as random noise, instantaneous interference, local fluctuations, and slow sensor drift present in single-channel signals, ensuring a high degree of consistency between radar and differential pressure signals in both time and numerical dimensions. Furthermore, it can automatically adjust correction constraints based on the actual working state of the sensors and changes in on-site conditions, significantly improving the system's adaptability and long-term operational stability in complex chemical environments, media fluctuations, and equipment aging scenarios. The equal-weighted arithmetic averaging achieves complementary advantages and cross-verification of the dual-mode measurement signals, further reducing the impact of single-channel signal anomalies on the overall measurement results. This significantly improves the accuracy, smoothness, and reliability of the fused liquid level values, providing high-quality, high-stability, and high-reliability data support for subsequent result self-verification, safety alarm triggering, and full-process safety monitoring of storage tanks. This comprehensively enhances the measurement performance and engineering applicability of the entire dual-mode intelligent liquid level fusion system.

[0062] S4. Master-Slave Reconstruction Mode Integration: In the master-slave reconstruction mode, a reliable signal that has passed physical pre-verification is used as the dominant signal. The material mass balance relationship of the storage tank is used to combine the dominant signal and reverse the deduction in real time to generate a virtual reconstruction value of another signal, so that the reconstruction value is physically consistent with the dominant signal.

[0063] Step S4 is a virtual signal generation and adaptive fusion method based on a mass conservation physics engine. Its core lies in using a verified reliable signal as the dominant signal. Based on the law of conservation of mass of the storage tank, it generates a physically consistent virtual associated signal in real time. Final output value Determined by the following adaptive model:

[0064]

[0065] in, For the physical model mismatch degree, the function For based on The method uses a monotonically increasing adjustment function to address the consistency error of the physical model. Dynamic driving enables a smooth, adaptive transition from "completely trusting the sensor" to "gradually adopting physical model predictions," thereby ensuring the physical authenticity and reliability of the output value even when the signal source may degrade.

[0066] After the system enters master-slave reconstruction mode, it immediately activates the virtual signal generation and adaptive fusion mechanism based on the mass conservation physics engine, with physical constraints as the core throughout the process to ensure the authenticity and reliability of the output value.

[0067] First, the system reads the signal status identifier from the physical pre-verification output in step S1, automatically filters and locks the currently marked normal signal as the dominant signal. The signal has passed dual physical verification of instantaneous rate of change and static offset, and has the rigid constraint condition of reliable measurement input;

[0068] Next, based on the storage tank's geometric parameters, the safe range of material density, the rated constraints of inlet and outlet flow rates, and the law of conservation of mass, a liquid level-mass physical mapping engine is constructed to guide the signal. To ensure real-time input, a reverse deduction is performed using a mass conservation physical model to generate virtual associated signals that are physically consistent with the current operating conditions, material state, and liquid level. This ensures that the virtual signal and the dominant signal remain consistent in terms of process logic and physical changes, without jumps, drifts, or physical contradictions;

[0069] Subsequently, the mismatch degree of the physical model is calculated. This is used to characterize the real-time physical deviation between the virtual accompanying signal and the dominant signal, reflecting the consistency error between the model and the sensor. The system has a built-in mismatch-based feature. monotonically increasing adjustment function This function follows Increases smoothly and gradually, with It decreases smoothly and achieves dynamic adaptive adjustment;

[0070] Final output value Calculated in real time by the adaptive fusion model: ;

[0071] When the physical model mismatch When it is small, the adjustment function The output is relatively small; the system relies primarily on the sensor and the output value is close to... ;

[0072] When the physical model mismatch As it gradually increases, the adjustment function Dynamic improvement: The system gradually adopts the prediction results of the physical model, achieving a smooth transition from "completely trusting the sensors" to "gradually adopting the predictions of the physical model";

[0073] When the signal source drifts, degrades, or becomes abnormal, the system can still maintain the physical authenticity and continuity of the output value based on the mass conservation physical model, without jumping, distortion, or interruption of the output;

[0074] The entire process is driven by a physics engine and requires no manual intervention. It can automatically reconstruct and adaptively fuse virtual signals in scenarios such as single-channel sensor failure, signal degradation, and sudden changes in operating conditions, ensuring that the liquid level output always meets the physical constraints of the process and the requirements for safe operation.

[0075] Beneficial Effects: By constructing a master-slave reconstruction and adaptive fusion mechanism driven by a physics engine based on mass conservation, the liquid level output can maintain continuous stability, physical consistency, and reliability under various complex operating conditions, including normal, drifting, degradation, or abnormal sensor signals. It is dominated by normal signals that have passed physical pre-verification, while dynamically generating virtual accompanying signals highly matched to the operating conditions, materials, and liquid level status based on the mass conservation physical model. Seamless connection and signal reconstruction can be achieved when a single sensor fails without additional hardware, significantly reducing the hardware cost and engineering implementation difficulty of multi-sensor redundant monitoring systems. Furthermore, through physical... The dynamic coupling of physical model mismatch degree and monotonically increasing adjustment function enables a smooth transition from complete trust in sensors to gradual adoption of physical model predictions. This effectively avoids output anomalies caused by sensor jumps, drift, and distortion, ensuring that the liquid level output always conforms to the process physical constraints. It significantly improves the stability, reliability, and safety of the system under complex operating conditions, sudden changes, and long-term continuous operation. No manual intervention or parameter tuning is required throughout the process. It has strong self-adaptability and high versatility, and can be widely used for high-precision liquid level monitoring in various chemical storage tanks, closed containers, and complex environments with high and low temperatures and strong interference, providing a solid guarantee for the safe operation of industrial processes.

[0076] S5. Result generation and self-verification: Based on the operating mode decided in step S2, output the first type of fusion value in step S3 or the dominant signal value in step S4. Substitute the output value into the physical limit verification and mass balance relationship for reverse verification. If the verification fails, trigger a fault alarm and lock the output.

[0077] Step S5 is an output self-check and fault isolation mechanism based on physical model reverse verification. The fusion result generated by S3 or S4 is used as the value to be verified. It is substituted into the physical limit check of S1 and the mass conservation relationship of S4 for reverse consistency verification. If the verification passes, the value is output. If the verification fails, a high-level fault alarm is immediately triggered, and the system output is locked at the most recently verified safe value, so as to achieve rigid maintenance of safe output under fault conditions.

[0078] Based on the operating mode of the dynamic decision in step S2, the system automatically selects the corresponding output value as the result to be verified: if it is the collaborative verification mode, the first type of fusion value generated in step S3 is selected; if it is the master-slave reconstruction mode, the adaptive fusion value of the dominant signal and the virtual reconstruction signal output in step S4 is selected. The system takes the result to be verified as the detection object and starts the dual self-check process of physical limit reverse verification and quality balance consistency verification.

[0079] First, perform a physical limit reverse verification. Substitute the result to be verified into the theoretical maximum instantaneous rate of change threshold and the allowable static offset threshold set in step S1 to verify whether the liquid level value and rate of change are within the safe range allowed by the tank process, and whether there are any abnormal jumps that exceed the physical boundary. If the value or rate of change exceeds the preset threshold range, the verification is directly determined to fail.

[0080] Then, reverse verification of the mass balance relationship is performed. The result to be verified is substituted into the mass conservation physical model built based on the storage tank geometric parameters, medium density, and inlet and outlet flow rates. The corresponding material volume and material mass are deduced to determine whether they satisfy the storage tank material balance relationship and whether they conform to the process physical logic.

[0081] If both checks pass, the system determines the result is valid and outputs the value as the final liquid level value. If either check fails, the system immediately determines the current output is unreliable, triggers a high-level fault alarm signal, and automatically executes the output locking mechanism to keep the output value at the safe value of the most recent successful double check until the signal returns to normal. This ensures that the output value is safe, reliable, does not jump, and is not distorted even when the sensor fails, the data is abnormal, or the model drifts, providing a rigid and reliable guarantee for the continuous safe operation of chemical storage tanks.

[0082] Beneficial effects: Through a dual physical constraint reverse verification mechanism, closed-loop self-checking and reliable locking of the fused output results are achieved. Under complex operating conditions such as sensor drift, data anomalies, or model deviations, invalid outputs can be automatically identified and eliminated, ensuring that the final output value is always within a physically reasonable and process-safe range. Based on physical limit verification, abnormal jumps exceeding process boundaries can be quickly identified, avoiding erroneous outputs caused by instantaneous interference. On the other hand, based on mass conservation self-verification, the rationality and consistency of the results can be guaranteed from the underlying mechanism, eliminating false data without physical meaning. When any verification fails, the system automatically locks the most recent valid output, achieving safe maintenance under fault conditions and completely eliminating risks such as liquid level jumps, false alarms, and loss of control. This significantly improves the reliability, stability, and safety of the entire monitoring system, providing a rigid guarantee for the continuous, stable, and safe operation of the storage tank.

[0083] The above embodiments illustrate only one implementation of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A radar-differential pressure dual-mode liquid level signal intelligent fusion method for chemical storage tanks, characterized in that, Includes the following steps: S1. Synchronous acquisition and physical pre-verification: The original liquid level signals from the radar and differential pressure sensor are acquired synchronously, and the instantaneous rate of change and static offset of the two signals are physically verified according to the tank process parameters. The synchronous signal pair with anomaly marker is output. S2. Dynamic decision-making of operating mode: Based on the real-time cross-correlation strength and trend consistency of the synchronization signal pair, the current operating mode is determined. When the two signals are highly correlated and consistent, the "cooperative verification mode" is entered; otherwise, the "master-slave reconstruction mode" is entered. S3. Collaborative verification mode fusion: In collaborative verification mode, the weighted median of the two signals within the sliding time window is used as a common benchmark. Each signal is then constrained to approximate the benchmark. Subsequently, the two corrected signals are averaged with equal weights to generate the first type of fusion value. S4. Master-Slave Reconstruction Mode Integration: In the master-slave reconstruction mode, a reliable signal that has passed physical pre-verification is used as the dominant signal. The material mass balance relationship of the storage tank is used to combine the dominant signal and reverse the deduction in real time to generate a virtual reconstruction value of another signal, so that the reconstruction value is physically consistent with the dominant signal. S5. Result generation and self-verification: Based on the operating mode decided in step S2, output the first type of fusion value in step S3 or the dominant signal value in step S4. Substitute the output value into the physical limit verification and quality balance relationship for reverse verification. If the verification fails, trigger a fault alarm and lock the output. 2.The radar-differential pressure dual-mode liquid level signal intelligent fusion method for a chemical storage tank according to claim 1, characterized in that, The physical limit verification in step S1 specifically includes: Based on the safe operating liquid level of the storage tank, the maximum rated flow rate of the feed / discharge pump, and the safe density range of the material, the theoretical maximum instantaneous rate of change threshold and the allowable static offset threshold of the liquid level change are calculated. The real-time liquid level signals collected by the radar and differential pressure sensor are compared with the liquid level value at the previous moment, and their instantaneous rate of change is calculated. The real-time liquid level signal is compared with the historical average liquid level value of the storage tank in a static state, and the static offset of each signal is calculated. When the instantaneous rate of change of any signal continuously exceeds the theoretical maximum instantaneous rate of change threshold, or its static offset continuously exceeds the allowable static offset threshold, the signal is marked as "abnormal"; otherwise, it is marked as "normal". 3.The radar-differential pressure dual-mode liquid level signal intelligent fusion method for a chemical storage tank according to claim 1, characterized in that, The "dynamic decision-making of operating mode" is achieved through real-time joint analysis of the cross-correlation strength and trend consistency between signals. The decision-making logic is as follows: the absolute value of the Pearson correlation coefficient of the synchronous signal pair within the sliding time window is calculated as the cross-correlation strength, and the average absolute error of the first-order difference sequence of the two signals is calculated as the trend consistency measure. During the decision-making process, the cross-correlation strength is compared with the preset correlation judgment threshold, and the trend consistency is compared with the preset consistency judgment threshold. If and only if the cross-correlation strength is higher than its judgment threshold and the trend consistency is lower than its judgment threshold, the system determines that the two signals are in a highly coordinated state and enters the "coordinated verification mode". Otherwise, based on the signal identifier output by the physical pre-verification in step S1, the signal currently marked as "normal" is selected as the dominant signal source, and the decision-making system enters the "master-slave reconstruction mode".

4. The radar-differential pressure dual-mode liquid level signal intelligent fusion method for a chemical storage tank according to claim 3, characterized in that, The calculation of the cross-correlation strength is specifically as follows: within a preset sliding time window, the Pearson correlation coefficient of the radar and differential pressure synchronous liquid level signal sequences is calculated, and the absolute value of the coefficient is used as the real-time cross-correlation strength. The calculation of the trend consistency is specifically as follows: within the same time window, the first-order difference sequences of the two signal sequences are calculated respectively, and the average absolute error value of the two difference sequences is calculated, which is used as the quantitative measure of trend consistency. 5.The radar-differential pressure dual-mode liquid level signal intelligent fusion method for chemical storage tank according to claim 1, characterized in that, The "collaborative verification mode fusion" is achieved through two stages: benchmark constraint correction and equal-weight fusion. First, the arithmetic mean of the two signals at each moment within the sliding time window is used as the benchmark to form a benchmark sequence. Then, a constraint approximation correction is performed on the radar and differential pressure signal sequences at each moment within the window: if the absolute value of the deviation from the benchmark value is not greater than the preset maximum allowable approximation error, the original value is retained; if the deviation exceeds the limit, the signal value at that moment is corrected to the benchmark value plus or minus the boundary value of the maximum allowable approximation error, thereby ensuring that the deviation between the corrected signal and the benchmark is within the preset range. Finally, the two corrected signal sequences are subjected to an equal-weighted arithmetic mean at the corresponding moment to output the first type of fusion value.

6. The radar-differential pressure dual-mode liquid level signal intelligent fusion method for a chemical storage tank according to claim 5, characterized in that, In the constrained approximation correction process, the determination and optimization of the maximum allowable approximation error (δ) includes: Reference setting: Based on the steady-state measurement errors of the radar and differential pressure sensor in the cooperative working mode, the larger of the two is taken as the initial reference value of the maximum allowable approximation error; Dynamic optimization: During the operation of the collaborative verification mode, the maximum allowable approximation error is adaptively and dynamically adjusted based on the real-time statistical characteristics of the difference between the radar signal correction value and the differential pressure signal correction value.

7. The intelligent fusion method for radar-differential pressure dual-mode liquid level signals for chemical storage tanks according to claim 1, characterized in that, The step S4 is a virtual signal generation and adaptive fusion method based on a mass conservation physical engine, the core of which is to take the reliable signal verified by the check as the leading signal , and to generate its physically self-consistent virtual companion signal in real time according to the mass conservation law of the storage tank , and the final output value is determined by the following adaptive model: in, For the physical model mismatch degree, the function For based on The method uses a monotonically increasing adjustment function to address the consistency error of the physical model. Dynamic driving enables a smooth, adaptive transition from "completely trusting sensors" to "gradually adopting physical model predictions," thereby ensuring the physical authenticity and reliability of the output values ​​even when the signal source may degrade.

8. The intelligent fusion method for radar-differential pressure dual-mode liquid level signals for chemical storage tanks according to claim 1, characterized in that, Step S5 is an output self-check and fault isolation mechanism based on physical model reverse verification. The fusion result generated by S3 or S4 is used as the value to be verified, and it is substituted into the physical limit verification of S1 and the mass conservation relationship of S4 for reverse consistency verification. If the verification passes, the value is output; if the verification fails, a high-level fault alarm is immediately triggered, and the system output is locked at the most recently verified safe value, so as to achieve rigid maintenance of safe output under fault conditions.

Images