Method and system for correcting measurement anomaly of a u-shaped vibrating tube densimeter

Abstract

The present application relates to the technical field of density measurement, and discloses a measurement anomaly correction method and system of a U-shaped vibrating tube densimeter, the method comprising: collecting a sequence of original density values of a target object, and calculating the sequence of original density values to obtain statistical characteristics of the target object; based on a preset sliding time window, performing trend prediction on the sequence of original density values to obtain a sequence of trend prediction values of the target object; performing difference analysis on the sequence of original density values and the sequence of trend prediction values to obtain a residual sequence of the target object; inputting the statistical characteristics and the residual sequence into a pre-trained anomaly probability evaluation model to output an anomaly probability of the target object; coupling data according to the anomaly probability and the residual sequence to obtain an adaptive correction factor of the target object; and applying the adaptive correction factor to a current measurement value of the target object to obtain a corrected density value of the target object; the present application can improve the efficiency of measurement anomaly correction of the U-shaped vibrating tube densimeter.

Description

Technical Field

[0001] This invention relates to the field of density measurement technology, and in particular to a method and system for correcting measurement anomalies in a U-shaped vibrating tube density meter. Background Technology

[0002] In the field of artificial intelligence technology, existing technologies have significant shortcomings in the data preprocessing and feature extraction stages for anomaly correction in U-shaped vibrating tube densitometer measurements. They fail to conduct systematic statistical feature analysis on the original density value sequence, relying solely on a single data dimension to identify anomalies, thus failing to comprehensively capture the central trend, dispersion, and variation characteristics of the data. Furthermore, they do not utilize sliding time windows for trend prediction and residual analysis, directly using the original measurement data for correction, making it difficult to separate normal trends from abnormal fluctuations in the data. This results in a lack of accurate data support for anomaly identification and an inability to predict potential measurement anomalies in advance.

[0003] Existing technologies have significant shortcomings in the assessment and correction of measurement anomalies in U-shaped vibrating tube densitometers. They lack a dedicated anomaly probability assessment model, relying solely on fixed thresholds or empirical judgments to identify anomalies. This fails to quantify the degree and confidence level of anomalies, leading to strong subjectivity and low accuracy in anomaly assessment. Furthermore, they do not generate adaptive correction factors by coupling anomaly probability with residual sequences, employing only fixed correction coefficients or simple compensation methods, which are ill-suited for different types and degrees of measurement anomalies. Finally, the lack of rationality verification after correction, coupled with direct output of correction results, easily results in correction values ​​exceeding the density range of the object, leading to poor reliability and failing to meet the practical needs of high-precision density measurement. Summary of the Invention

[0004] This invention provides a method and system for correcting measurement anomalies in a U-shaped vibrating tube density meter to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for correcting measurement anomalies in a U-shaped vibrating tube density meter, comprising:

[0006] S1. Collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical characteristics of the target object;

[0007] S2. Based on a preset sliding time window, perform trend prediction on the original density value sequence to obtain the trend prediction value sequence of the target object;

[0008] S3. Perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object;

[0009] S4. Input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object;

[0010] S5. Based on the anomaly probability and the residual sequence, perform data coupling to obtain the adaptive correction factor of the target object;

[0011] S6. Apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object.

[0012] In a preferred embodiment, the step of acquiring the original density value sequence of the target object and calculating the original density value sequence to obtain the statistical characteristics of the target object includes:

[0013] Obtain the density measurement values ​​of the target object over a period of time, and arrange the density measurement values ​​in chronological order to obtain the original density value sequence of the target object;

[0014] The original density value sequence is subjected to sliding window mean filtering to obtain the filtered density sequence of the target object;

[0015] The mean of the filtered density sequence is used as the central tendency statistic, and the standard deviation of the filtered density sequence is used as the dispersion statistic.

[0016] The coefficient of variation of the original density value sequence is obtained by performing a ratio discretization operation on the central tendency statistic and the dispersion statistic.

[0017] The central tendency statistic, the dispersion statistic, and the coefficient of variation statistic are integrated into the statistical characteristics of the target object.

[0018] In a preferred embodiment, the ratio discretization operation is performed on the central tendency statistic and the dispersion statistic to obtain the coefficient of variation statistic of the original density value sequence. The formula for calculating the coefficient of variation statistic is as follows:

[0019] ;

[0020] in, The coefficient of variation statistic is mentioned above. The aforementioned dispersion statistic, This refers to the central trend statistic.

[0021] In a preferred embodiment, the step of performing trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object includes:

[0022] The preset sliding time window is time-sequentially segmented to determine the length of the sliding time window for the target object;

[0023] Based on the length of the sliding time window, density values ​​within the window are sequentially extracted starting from the beginning of the original density value sequence to form a window density subsequence of the original density value sequence;

[0024] A linear fit is performed on the window density subsequence to obtain the fitted straight line of the window density subsequence;

[0025] Based on the fitted straight line, the trend prediction value sequence of the target object is obtained by extrapolating the subsequent sampled prediction values ​​of the window density subsequence.

[0026] In a preferred embodiment, the step of performing differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object includes:

[0027] Align the original density value sequence with the trend prediction value sequence at different times to obtain the density-trend correspondence group of the target object;

[0028] The instantaneous residual value of the target object is obtained by performing a difference operation on the density-trend correspondence group;

[0029] The instantaneous residual values ​​are collected in chronological order to form the initial residual sequence of the target object;

[0030] The initial residual sequence is smoothed to obtain the residual sequence of the target object.

[0031] In a preferred embodiment, the step of inputting the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and outputting the anomaly probability of the target object includes:

[0032] The pre-trained anomaly probability evaluation model is obtained through supervised training, which includes:

[0033] The historical density data of the target object is obtained and processed into a feature sample set;

[0034] To construct and initialize the forward propagation network structure of the target object;

[0035] The feature sample set is input into the forward propagation network structure for iterative training, and the network internal parameters of the forward propagation network structure are optimized through loss calculation and backpropagation.

[0036] When the training meets the stopping condition, the optimized parameters of the network's internal parameters are saved to obtain the anomaly probability evaluation model.

[0037] The anomaly probability assessment model includes an input layer, a hidden layer, and an output layer;

[0038] The statistical features and the residual sequence are combined into a comprehensive feature vector, and the comprehensive feature vector is input into the input layer;

[0039] After receiving the integrated feature vector in the input layer, the hidden layer performs a nonlinear feature transformation on the integrated feature vector to obtain a high-dimensional feature representation of the target object.

[0040] The high-dimensional feature representation is mapped to the output layer to determine the anomaly probability of the target object.

[0041] In a preferred embodiment, the step of receiving the comprehensive feature vector at the input layer and then performing a nonlinear feature transformation on the comprehensive feature vector at the hidden layer to obtain a high-dimensional feature representation of the target object includes:

[0042] Use the comprehensive feature vector as the initial input feature;

[0043] The initial input features are linearly weighted and combined to obtain the intermediate feature representation of the target object. The calculation formula for the intermediate feature representation is as follows:

[0044] ;

[0045] In the formula, This is a representation of the intermediate feature. Let be the dimension of the comprehensive feature vector. For the first The preset weight coefficients corresponding to each feature component The first in the comprehensive feature vector Each feature component This is a preset bias term;

[0046] The intermediate feature representation is subjected to a nonlinear transformation to obtain the high-dimensional features of the target object.

[0047] In a preferred embodiment, the step of data coupling based on the anomaly probability and the residual sequence to obtain the adaptive correction factor for the target object includes:

[0048] Based on the anomaly probability, determine the adjustment ratio coefficient of the residual sequence;

[0049] Based on the adjustment ratio coefficient, the residual values ​​in the residual sequence are dynamically adjusted to obtain the adjusted residual sequence of the target object.

[0050] The adjusted residual sequence is subjected to feature fusion processing to obtain the adaptive correction factor of the target object.

[0051] In a preferred embodiment, applying the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object includes:

[0052] The current density measurement value of the target object is acquired synchronously.

[0053] Based on the positive or negative attribute of the adaptive correction factor, the correction direction for the density measurement value at the current moment is determined;

[0054] According to the correction direction, the adaptive correction factor and the density measurement value at the current time are algebraically synthesized to obtain the preliminary corrected density value of the target object;

[0055] The rationality of the preliminary corrected density value is verified;

[0056] If the preliminary corrected density value is within the preset physical property density value range of the target object, then the preliminary corrected density value is directly output as the corrected density value.

[0057] If the value exceeds the range of the physical property density value, the boundary value closest to the initial corrected density value in the range of physical property density value is output as the corrected density value.

[0058] To address the aforementioned problems, the present invention also provides a measurement anomaly correction system for a U-shaped vibrating tube density meter, the system comprising:

[0059] The feature statistics module is used to collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical features of the target object;

[0060] The trend prediction module is used to perform trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object.

[0061] The differential analysis module is used to perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object;

[0062] An anomaly probability module is used to input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object.

[0063] The data coupling module is used to perform data coupling based on the anomaly probability and the residual sequence to obtain the adaptive correction factor of the target object;

[0064] The density correction module is used to apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object.

[0065] Compared with the prior art, the present invention has the following beneficial effects:

[0066] 1. This invention provides a scientific basis for density measurement anomaly correction through multi-dimensional data analysis and precise anomaly assessment. It collects the original density value sequence and extracts statistical features, combining this with a sliding time window to complete trend prediction and residual sequence analysis. These two types of key information are input into a trained anomaly probability assessment model, accurately outputting the anomaly probability. This enables rapid identification and quantitative judgment of measurement anomalies, improving the accuracy and timeliness of anomaly detection.

[0067] 2. This invention significantly improves the reliability and accuracy of density measurement results by employing adaptive correction and rationality verification. An adaptive correction factor is generated based on the coupling of anomaly probability and residual sequence, dynamically adapting to measurement data with different degrees of anomaly. After obtaining the initial correction value through algebraic synthesis, the rationality of the results is ensured through physical property density range verification, controlling the measurement error to an extremely small range and providing stable and reliable technical support for high-precision density measurement scenarios. Attached Figure Description

[0068] Figure 1 This is a flowchart illustrating a method for correcting measurement anomalies in a U-shaped vibrating tube density meter according to an embodiment of the present invention.

[0069] Figure 2 This is a functional block diagram of a measurement anomaly correction system for a U-shaped vibrating tube density meter according to an embodiment of the present invention;

[0070] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0071] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0072] This application provides a method for correcting measurement anomalies in a U-shaped vibrating tube densitometer. The execution subject of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method for correcting measurement anomalies in a U-shaped vibrating tube densitometer can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0073] Reference Figure 1 The diagram shown is a flowchart illustrating a method for correcting measurement anomalies in a U-shaped vibrating tube density meter according to an embodiment of the present invention. In this embodiment, the method for correcting measurement anomalies in a U-shaped vibrating tube density meter includes:

[0074] S1. Collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical characteristics of the target object;

[0075] In this embodiment of the invention, the step of acquiring the original density value sequence of the target object and calculating the original density value sequence to obtain the statistical characteristics of the target object includes:

[0076] Obtain the density measurement values ​​of the target object over a period of time, and arrange the density measurement values ​​in chronological order to obtain the original density value sequence of the target object;

[0077] The original density value sequence is subjected to sliding window mean filtering to obtain the filtered density sequence of the target object;

[0078] The mean of the filtered density sequence is used as the central tendency statistic, and the standard deviation of the filtered density sequence is used as the dispersion statistic.

[0079] The coefficient of variation of the original density value sequence is obtained by performing a ratio discretization operation on the central tendency statistic and the dispersion statistic.

[0080] The central tendency statistic, the dispersion statistic, and the coefficient of variation statistic are integrated into the statistical characteristics of the target object.

[0081] The ratio discretization operation is performed on the central tendency statistic and the dispersion statistic to obtain the coefficient of variation statistic of the original density value sequence. The formula for calculating the coefficient of variation statistic is as follows:

[0082] ;

[0083] in, The coefficient of variation statistic is mentioned above. The aforementioned dispersion statistic, This refers to the central trend statistic.

[0084] Using specialized equipment adapted for measuring the density of the target object, density data is continuously collected over a preset period. Each collection yields a corresponding density measurement value, ensuring continuous collection and covering critical periods where the target object's density may change. All collected density measurements are then arranged chronologically according to their collection time, forming a continuous data sequence that fully reflects the density changes of the target object within that time period, thus obtaining the target object's original density value sequence.

[0085] A fixed-length sliding window is set, the length of which is determined based on the common periodicity of density changes in the target object and the sampling frequency, ensuring effective smoothing of data fluctuations without losing key change information. Starting from the beginning of the original density value sequence, the sliding window sequentially covers the data segments in the sequence. All density measurements within the coverage area of ​​each window are summed and averaged. This average value is used as the filtered data for the corresponding position in that window. These average values ​​are arranged sequentially according to the window movement order to obtain the filtered density sequence of the target object.

[0086] The process involves iterating through all data in the filtered density sequence, summing all the data, and then dividing the sum by the total number of data points. The resulting value is the mean of the filtered density sequence. This mean is used as the central tendency statistic to characterize the level of concentration of the target object's density.

[0087] The difference between each data point in the filtered density sequence and the central trend statistic is calculated. Each difference is squared to eliminate positive and negative effects. All squared differences are summed, and the sum is divided by the total number of data points to obtain the variance of the filtered density sequence. The square root of this variance is then taken to obtain the standard deviation of the filtered density sequence. This standard deviation is determined as the dispersion statistic to reflect the dispersion of the target object density data.

[0088] Using the central tendency statistic as the dividend and the dispersion statistic as the divisor, a division operation is performed. The result obtained through this ratio dispersion operation can comprehensively reflect the relative dispersion of the target object's density. This result is determined as the coefficient of variation statistic of the original density value sequence.

[0089] A unified statistical feature integration framework is established, incorporating the previously obtained central tendency statistics, dispersion statistics, and coefficient of variation statistics into the framework in a predetermined fixed order. This ensures that the three statistics are independent and clearly correlated within the framework, forming a comprehensive data set that can fully reflect the center, dispersion, and relative dispersion of the target object's density distribution. This set is then identified as the statistical feature of the target object.

[0090] The coefficient of variation statistic is derived from the dispersion statistic and the central tendency statistic, and is generated by integrating them through a specific ratio calculation logic. It is a core indicator that comprehensively reflects the relative dispersion characteristics of the original density value sequence.

[0091] The dispersion statistic is derived from the filtered density sequence. The difference between each data point in the sequence and the central tendency statistic is calculated. Each difference is squared to eliminate positive and negative differences. All squared differences are summed, and the sum is divided by the total number of data points in the sequence to obtain the variance. The variance is then squared to obtain the final value, which is the dispersion statistic.

[0092] The central tendency statistic is derived from the filtered density sequence. It involves iterating through all the data in the sequence and summing them up. The summation result is then divided by the total number of data points in the sequence. The value obtained through this averaging method is the central tendency statistic.

[0093] The significance of the formula lies in eliminating the influence of differences in data dimensions and numerical levels on the assessment of dispersion by using the ratio of the dispersion statistic to the central tendency statistic.

[0094] This operational logic does not rely on the absolute value of the data itself, but focuses on the relative proportion between the two, which can objectively reflect the relative dispersion of the original density value sequence and avoid the distortion of dispersion judgment caused by the different levels of central trend statistics.

[0095] This calculation method provides a comparable basis for the density dispersion characteristics of different target objects or the same target object at different time periods, providing a precise and unified quantitative basis for subsequent target object correlation analysis or judgment based on statistical characteristics.

[0096] The beneficial effect is that by arranging the density measurements of the target object in chronological order over a period of time, the original density value sequence can be obtained, and the trajectory of density change over time can be completely recorded, providing continuous and complete basic data for subsequent statistical feature calculations and ensuring the comprehensiveness of feature extraction.

[0097] A sliding window mean filter is applied to the original density value sequence. By moving the fixed window and calculating the mean, random interference and irregular fluctuations in the data are effectively smoothed, resulting in a filtered density sequence. This improves the stability and reliability of the data and eliminates noise interference for statistical calculations.

[0098] The mean of the filtered density sequence is used as the central tendency statistic to accurately characterize the central distribution level of the target object density; the standard deviation is used as the dispersion statistic to clearly reflect the dispersion of the density data. The two statistics capture the core distribution characteristics of the density data from different dimensions.

[0099] The coefficient of variation is obtained by performing a ratio discretization operation on the central tendency statistic and the dispersion statistic. This eliminates the influence of differences in data dimensions and numerical levels, objectively reflects the relative dispersion of density, and makes the density dispersion characteristics in different scenarios comparable.

[0100] By integrating three statistical measures into the statistical features of the target object, a multi-dimensional coverage of the distribution, discreteness, and relative dispersion of density data centers is achieved, comprehensively capturing the inherent patterns of the data and providing accurate and comprehensive feature inputs for subsequent anomaly probability assessment models, thereby improving the accuracy of anomaly identification.

[0101] The coefficient of variation is obtained by calculating the ratio of the central tendency statistic to the dispersion statistic, which effectively eliminates the influence of differences in data dimensions and numerical levels, and provides a unified and comparable basis for the density dispersion characteristics of different target objects or the same object at different time periods.

[0102] This calculation method focuses on the relative proportion between the two rather than their absolute numerical values, which can objectively reflect the relative dispersion of the original density value sequence and avoid the distortion of dispersion judgment caused by different levels of central trend statistics.

[0103] The coefficient of variation statistic integrates the concentrated distribution characteristics and dispersion of density data to form a single and comprehensive quantitative indicator, providing a concise and accurate feature input for subsequent anomaly probability assessment models, and helping the models to quickly capture data anomaly patterns.

[0104] The coefficient of variation statistic obtained by this formula can accurately distinguish between normal density fluctuations and abnormal dispersion, providing key data support for subsequent anomaly identification and measurement correction, and improving the pertinence and accuracy of anomaly correction by U-shaped vibrating tube densitometer.

[0105] S2. Based on a preset sliding time window, perform trend prediction on the original density value sequence to obtain the trend prediction value sequence of the target object;

[0106] In this embodiment of the invention, the step of performing trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object includes:

[0107] The preset sliding time window is time-sequentially segmented to determine the length of the sliding time window for the target object;

[0108] Based on the length of the sliding time window, density values ​​within the window are sequentially extracted starting from the beginning of the original density value sequence to form a window density subsequence of the original density value sequence;

[0109] A linear fit is performed on the window density subsequence to obtain the fitted straight line of the window density subsequence;

[0110] Based on the fitted straight line, the trend prediction value sequence of the target object is obtained by extrapolating the subsequent sampled prediction values ​​of the window density subsequence.

[0111] Based on the acquisition time interval of the original density value sequence, the historical cycle of density changes of the target object, and the accuracy requirements of trend prediction, the preset sliding time window is segmented in time. The continuous time span covered by each window after segmentation is clearly defined. This time span is the sliding time window length of the target object, ensuring that the window length can contain enough data to support trend analysis and accurately capture short-term change patterns.

[0112] Using a defined sliding time window length as a fixed truncation standard, starting from the first data point of the original density value sequence, all density values ​​within the corresponding time span are truncated to form the first window density subsequence. After the initial truncation, the sliding time window is moved sequentially along the time order of the original density value sequence by one acquisition interval. Density values ​​within the corresponding time span are truncated at the new window position to form the next window density subsequence. This process is repeated until the sliding time window covers all valid data in the original density value sequence, resulting in multiple consecutive window density subsequences.

[0113] For each window density subsequence, a two-dimensional coordinate system is established with time as the horizontal axis and density value as the vertical axis. Each data point in the subsequence is labeled in the coordinate system according to its corresponding time and density value. By finding a straight line that best approximates all data points, minimizing the sum of the vertical distances from each data point to this line, this straight line is the fitting line for the window density subsequence. This line clearly reflects the overall trend of density value changes over time within the window.

[0114] Based on the slope and intercept characteristics of the fitted straight line, the rate of change of density values ​​over time and the initial baseline level are determined. Following the acquisition time intervals of the original density value sequence, and continuing the trend of the fitted straight line, the predicted density values ​​for each subsequent acquisition time of the window density subsequence are calculated. These predicted values ​​are arranged sequentially in chronological order, and the subsequent predicted values ​​corresponding to all windows are integrated to form a trend prediction value sequence for the target object, fully presenting the trend of density change of the target object in future time periods.

[0115] The beneficial effect is that by dividing the preset sliding time window into time sequences to determine the window length, and combining the acquisition interval of the original density value sequence with the density change cycle of the target object, the window length can be ensured to cover enough data to support trend analysis and accurately capture short-term change patterns, providing a scientific standard for subsequent subsequence truncation.

[0116] By sequentially extracting window density subsequences from the starting point of the original density value sequence according to the window length, the original data can be segmented and refined, avoiding the trend ambiguity caused by overall data analysis, and making the density change characteristics of different time periods more prominent.

[0117] A linear fit is performed on the window density subsequence to obtain a fitted straight line. The straight line that fits the data points clearly shows the trend of density change over time, transforming discrete density data into a linear relationship with a clear pattern of change, providing an intuitive basis for trend prediction.

[0118] Based on the fitted straight line, subsequent sampling prediction values ​​are extrapolated, and the trend characteristics of each window are continued to obtain the trend prediction value sequence. This fully restores the future change trend of the target object density, provides an accurate prediction benchmark for subsequent residual sequence calculation, helps to separate normal trends and abnormal fluctuations in the data, and provides reliable support for anomaly identification.

[0119] S3. Perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object;

[0120] In this embodiment of the invention, the step of performing differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object includes:

[0121] Align the original density value sequence with the trend prediction value sequence at different times to obtain the density-trend correspondence group of the target object;

[0122] The instantaneous residual value of the target object is obtained by performing a difference operation on the density-trend correspondence group;

[0123] The instantaneous residual values ​​are collected in chronological order to form the initial residual sequence of the target object;

[0124] The initial residual sequence is smoothed to obtain the residual sequence of the target object.

[0125] Using the acquisition time point of the original density value sequence as a benchmark, the prediction time point corresponding to each predicted value in the trend prediction value sequence is checked one by one to ensure that each density measurement value in the original density value sequence can find a corresponding trend prediction value at the same time point in the trend prediction value sequence. The density measurement values ​​and trend prediction values ​​at the same time point are paired and combined to form a one-to-one corresponding data set. These data sets are the density-trend correspondence sets of the target object.

[0126] For each density-trend correspondence group, the original density measurement value and the trend prediction value at the same time point are extracted. The original density measurement value is subtracted from the corresponding trend prediction value. The result obtained by this difference calculation can reflect the deviation between the actual density and the predicted density at that time point. This deviation result is the instantaneous residual value of the target object.

[0127] By sorting out the time sequence corresponding to all instantaneous residual values ​​and strictly following the acquisition time sequence of the original density value sequence, all calculated instantaneous residual values ​​are arranged sequentially to ensure that the position of each instantaneous residual value in the sequence is consistent with its corresponding time point. The continuous data sequence formed by this orderly aggregation is the initial residual sequence of the target object.

[0128] A fixed-length smoothing window is set, the length of which is determined based on the fluctuation of the initial residual sequence and the data acquisition frequency, ensuring effective filtering of random interference without destroying the core variation characteristics of the residuals. Starting from the beginning of the initial residual sequence, the smoothing window sequentially covers the data segments in the sequence. All instantaneous residual values ​​within the coverage area of ​​each window are summed and averaged. This average value is used as the smoothed residual value at the corresponding position of the window. These smoothed residual values ​​are arranged sequentially according to the window movement order to obtain the residual sequence of the target object.

[0129] The beneficial effect is that by aligning the original density value sequence with the trend prediction value sequence at the same time point, the actual density data and the predicted density data are accurately matched in the same time dimension, forming a density-trend correspondence group, which provides basic data without time deviation for subsequent difference calculation and ensures the accuracy of residual calculation.

[0130] The instantaneous residual value is obtained by performing difference calculation on the density-trend correspondence group, which directly quantifies the degree of deviation between the actual density and the predicted trend at each time point, clearly separating the normal trend component and the abnormal fluctuation component in the data, and making the characteristics of measurement anomalies explicit.

[0131] The instantaneous residual values ​​are collected in chronological order to form an initial residual sequence, which completely preserves the trajectory of abnormal fluctuations over time, avoids the loss of abnormal information, and provides continuous and complete fluctuation data support for subsequent smoothing and anomaly analysis.

[0132] The initial residual sequence is smoothed to effectively filter out random interference and noise in the residuals, highlight the core features of abnormal fluctuations, and obtain a more stable and reliable residual sequence. This provides high-quality features for subsequent anomaly probability assessment models and improves the accuracy of anomaly identification.

[0133] S4. Input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object;

[0134] In this embodiment of the invention, the step of inputting the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and outputting the anomaly probability of the target object includes:

[0135] The pre-trained anomaly probability evaluation model is obtained through supervised training, which includes:

[0136] The historical density data of the target object is obtained and processed into a feature sample set;

[0137] To construct and initialize the forward propagation network structure of the target object;

[0138] The feature sample set is input into the forward propagation network structure for iterative training, and the network internal parameters of the forward propagation network structure are optimized through loss calculation and backpropagation.

[0139] When the training meets the stopping condition, the optimized parameters of the network's internal parameters are saved to obtain the anomaly probability evaluation model.

[0140] The anomaly probability assessment model includes an input layer, a hidden layer, and an output layer;

[0141] The statistical features and the residual sequence are combined into a comprehensive feature vector, and the comprehensive feature vector is input into the input layer;

[0142] After receiving the integrated feature vector in the input layer, the hidden layer performs a nonlinear feature transformation on the integrated feature vector to obtain a high-dimensional feature representation of the target object.

[0143] The high-dimensional feature representation is mapped to the output layer to determine the anomaly probability of the target object.

[0144] After receiving the comprehensive feature vector at the input layer, the hidden layer performs a nonlinear feature transformation on the comprehensive feature vector to obtain a high-dimensional feature representation of the target object, including:

[0145] Use the comprehensive feature vector as the initial input feature;

[0146] The initial input features are linearly weighted and combined to obtain the intermediate feature representation of the target object. The calculation formula for the intermediate feature representation is as follows:

[0147] ;

[0148] In the formula, This is a representation of the intermediate feature. Let be the dimension of the comprehensive feature vector. For the first The preset weight coefficients corresponding to each feature component The first in the comprehensive feature vector Each feature component This is a preset bias term;

[0149] The intermediate feature representation is subjected to a nonlinear transformation to obtain the high-dimensional features of the target object.

[0150] Historical density measurement data of the target object over a period of time are collected. Valid data are filtered according to the same collection criteria and time intervals as those used to obtain the original density value sequence. These historical density data are then processed in sequence with sliding window mean filtering, statistical feature extraction, trend prediction, and residual sequence calculation. The statistical features corresponding to each historical data are associated and paired with the residual sequence. At the same time, the actual abnormal state label corresponding to each paired data is labeled to form a feature sample set containing input features and corresponding labels.

[0151] Based on the feature complexity of the target object density data and the requirements for anomaly identification, a forward propagation network structure comprising an input layer, hidden layers, and an output layer is designed. The number of nodes in each layer and the inter-layer connection methods are clearly defined. The number of nodes in the input layer matches the dimension of the comprehensive feature vector, the hidden layer has a fixed number of nodes for feature transformation, and the number of nodes in the output layer corresponds to the output dimension of the anomaly probability. Initial values ​​are assigned to the connection weights, biases, and other parameters within the network. These initial values ​​are generated according to a fixed random distribution rule to ensure that the network structure has initial computational capabilities, thus completing the construction and initialization of the forward propagation network structure for the target object.

[0152] The feature sample set is divided into a training sample subset and a validation sample subset at a fixed ratio. The training sample subset is first input into the constructed forward propagation network structure. The network processes the sample features sequentially, from the input layer to the hidden layer and then to the output layer, obtaining the predicted output for each training sample. The predicted output is compared with the corresponding actual abnormal state labels in the sample set, and the difference between the two is quantified using a fixed loss calculation method to obtain the loss value. Based on the magnitude and trend of the loss value, the connection weights and bias parameters within the network are adjusted in the reverse direction of network propagation to reduce the loss value. This process of sample input, forward propagation, loss calculation, and parameter adjustment is repeated until a predetermined number of iterations are completed. During the iteration process, the network's predictive performance is periodically tested using the validation sample subset to dynamically monitor the training effect.

[0153] Fixed training stopping conditions are set, including a preset maximum number of iterations, a loss value below a fixed threshold, or the prediction accuracy of a subset of validation samples reaching a preset standard. The number of iterations, changes in loss value, and prediction accuracy are continuously monitored during training. When any one of these stopping conditions is met, the training process is immediately terminated, and the optimized parameters of all connection weights and biases within the network are saved. These parameters record the anomaly recognition patterns learned by the network from the feature sample set. The forward propagation network structure after saving the parameters is determined as the anomaly probability evaluation model.

[0154] The central tendency statistic, dispersion statistic, and coefficient of variation statistic in the statistical features are organized and expanded into a one-dimensional feature vector in a fixed order. All data in the residual sequence are also expanded into a one-dimensional vector in chronological order. The two one-dimensional vectors are then concatenated end to end in a preset fixed order to ensure that all information in the statistical features and residual sequence is completely preserved and arranged in an orderly manner, forming a comprehensive feature vector that includes the density distribution characteristics and deviation characteristics of the target object. This comprehensive feature vector is then input into the input layer of the anomaly probability assessment model.

[0155] After receiving the comprehensive feature vector, the input layer transmits all the feature data in the vector to the hidden layer according to a fixed transmission rule. The hidden layer processes the received feature data through an internally preset nonlinear transformation rule, breaks the linear correlation between features, mines the complex correlation information hidden in the data, and transforms the processed feature data into a high-dimensional feature representation that can more accurately represent abnormal correlation patterns.

[0156] The hidden layer passes the obtained high-dimensional feature representation to the output layer. The output layer processes the high-dimensional feature representation through fixed mapping rules, transforming it into a value within a specific range. This value directly corresponds to the probability of the target object being abnormal. This value is the abnormal probability of the target object, thus completing the determination and output of the abnormal probability.

[0157] The input layer directly passes the received integrated feature vector to the hidden layer. This integrated feature vector contains all the effective information of the statistical features of the target object and the residual sequence. The hidden layer directly uses it as the initial input feature for subsequent feature processing to ensure that the feature information enters the transformation process completely without omission.

[0158] The hidden layer calls the network's internal connection weights and bias parameters determined after training, and performs a weighted operation on each feature component in the initial input features, that is, multiplies each feature component by its corresponding connection weight, then sums all the weighted feature components, and then adds the preset bias parameters. Through this linear group, a fixed nonlinear transformation rule is used to process the intermediate feature representation. This rule can perform nonlinear mapping on the values ​​in the intermediate feature representation, amplify the effective feature differences, suppress irrelevant information interference, break the linear correlation limit between initial features, and uncover the complex correlation patterns hidden inside the features. The processed result is determined as the high-dimensional features of the target object.

[0159] The intermediate feature representation originates from the comprehensive feature vector, preset weight coefficients, and preset bias terms. The preset weight coefficients are parameters configured for each feature component of the comprehensive feature vector when the hidden layer of the anomaly probability assessment model performs linear weighted combination. The determination of their values ​​is explained in two stages: initialization and training optimization. Initialization generates features based on a uniform random distribution of 0 to 1. The initial values ​​of the statistical feature components are biased towards the range of 0.7 to 1.0, the residual sequence components are biased towards the range of 0.0 to 0.3, and the bias terms are taken as 0.01 to 0.05. Training optimization is based on the initial values ​​and is adjusted through iterative training. The weights of the statistical feature components are stabilized at 0.90 to 0.98, the residual sequence components are stabilized at 0.03 to 0.10, and the bias terms are taken as 0.02 to 0.03. These features are generated through specific linear combination operation logic and represent the feature form after preliminary information fusion of the initial input features.

[0160] The comprehensive feature vector is derived from the combination of statistical features and residual sequences. The central tendency statistic, dispersion statistic, and coefficient of variation statistic in the statistical features are expanded into a one-dimensional feature vector in a fixed order. Then, the residual sequence is expanded into a one-dimensional vector in time order. The two one-dimensional vectors are concatenated end to end in a preset order to form a comprehensive feature vector containing the density distribution features and deviation features of the target object. Each independent information unit contained therein is a feature component.

[0161] The preset weight coefficients are parameters that are optimized and determined during the training of the anomaly probability assessment model. During the model training phase, through iterative training and backpropagation of the feature sample set, the weight values ​​are assigned to each feature component according to the importance of different feature components to the anomaly probability assessment, so as to adjust the influence of each feature component in the calculation.

[0162] The preset bias term is a parameter set during the initialization of the anomaly probability assessment model and optimized during training. In the initial stage, it is assigned a basic value according to fixed rules, and then continuously adjusted through loss calculation and backpropagation. It is used to compensate for possible systematic biases in linear combination operations, so that the intermediate feature representation is more in line with the actual data pattern.

[0163] The significance of the formula lies in achieving linear information fusion of the initial input features by weighting and integrating each feature component in the comprehensive feature vector and compensating for bias.

[0164] This operational logic highlights the role of important feature components based on preset weight coefficients, weakens the interference of secondary feature components, and at the same time corrects the baseline level of the linear combination through bias terms, integrating scattered multi-dimensional feature information into a unified intermediate feature representation.

[0165] This linear combination method can retain the core information of each feature component and establish a preliminary correlation, laying the foundation for subsequent nonlinear transformation to mine complex feature patterns, ensuring that the feature processing process is orderly and efficient, and improving the ability of the anomaly probability assessment model to identify the abnormal state of the target object.

[0166] The beneficial effect is that by processing the historical density data of the target object to form a feature sample set, the model can be provided with training data that fits the actual application scenario, ensuring that the anomaly recognition rules learned by the model are practical and adaptable.

[0167] Construct and initialize a forward propagation network structure containing an input layer, hidden layers, and an output layer, clarify the function and connection logic of each layer, provide a stable network framework for feature processing and anomaly probability output, and ensure the orderliness of model operations.

[0168] By iteratively training and calculating loss, and optimizing the network's internal parameters through backpropagation, the model's prediction error is continuously reduced, enabling the model to accurately capture the correlation between statistical features, residual sequences, and abnormal states, thereby improving the accuracy of anomaly probability assessment.

[0169] When the training meets the stopping condition, the optimized parameters are saved to obtain a stable anomaly probability evaluation model, avoiding model performance defects caused by overtraining or undertraining, and ensuring that the model can stably output reliable results.

[0170] By combining statistical features with residual sequences into a comprehensive feature vector, integrating density distribution features and fluctuation bias features, the model is provided with comprehensive and multidimensional input information, avoiding the one-sidedness of evaluation caused by a single feature input.

[0171] The hidden layer performs nonlinear feature transformation on the comprehensive feature vector, breaking the linear constraints between features, mining the complex anomaly patterns hidden in the data, obtaining high-dimensional feature representations, and improving the model's ability to identify subtle anomalies.

[0172] The output layer maps high-dimensional feature representations to anomaly probabilities, enabling quantitative output of the degree of anomaly. This replaces traditional experience-based judgments or fixed threshold judgments, making anomaly assessment more objective and accurate, and providing a scientific basis for subsequent adaptive correction.

[0173] By directly using the comprehensive feature vector as the initial input feature, the multidimensional information of statistical features and residual sequences is fully preserved, avoiding information loss during feature transfer and providing comprehensive and complete basic data support for subsequent feature transformation.

[0174] The initial input features are linearly weighted and combined. By pre-setting weight coefficients, the influence of key feature components is highlighted and the interference of secondary features is weakened. At the same time, the bias term is used to compensate for the systematic bias of linear operation, and multi-dimensional feature information is accurately integrated to obtain intermediate feature representation, making feature association more logical.

[0175] The linear weighted combination operation transforms the scattered comprehensive feature vectors into a unified intermediate feature representation, simplifying the processing difficulty of subsequent nonlinear transformations, while preserving the core correlation between features, thus laying the foundation for in-depth mining of abnormal patterns.

[0176] By performing nonlinear transformations on intermediate feature representations, the linear correlation between features is broken, the differences in potential abnormal features are amplified, and the complex abnormal patterns hidden in the data are uncovered, making the features more discriminative.

[0177] Through a two-step process of "linear integration + nonlinear transformation", a high-dimensional feature representation is finally obtained. This not only integrates the core information of the basic features, but also enhances the complex patterns related to anomalies, providing high-quality feature input for the output layer to accurately map the anomaly probability and significantly improving the accuracy of anomaly identification.

[0178] S5. Based on the anomaly probability and the residual sequence, perform data coupling to obtain the adaptive correction factor of the target object;

[0179] In this embodiment of the invention, the step of data coupling based on the anomaly probability and the residual sequence to obtain the adaptive correction factor for the target object includes:

[0180] Based on the anomaly probability, determine the adjustment ratio coefficient of the residual sequence;

[0181] Based on the adjustment ratio coefficient, the residual values ​​in the residual sequence are dynamically adjusted to obtain the adjusted residual sequence of the target object.

[0182] The adjusted residual sequence is subjected to feature fusion processing to obtain the adaptive correction factor of the target object.

[0183] Referring to the numerical value of the anomaly probability and the preset proportional mapping rule, the core rule of the proportional mapping is that the anomaly probability and the adjustment proportional coefficient are positively correlated: anomaly probability With adjustment ratio coefficient For a strictly linear positive correlation, the specific proportional relationship is as follows: ,in Range of values 0 indicates no abnormality, and 1 indicates a serious abnormality. Range of values As the probability of an anomaly increases, the adjustment ratio increases accordingly. The more severe the anomaly, the stronger the adjustment. Furthermore, a fixed range of adjustment ratios can be defined according to the anomaly probability interval, achieving a precise match between the degree of anomaly and the adjustment strength.

[0184] The determined adjustment ratio is applied to each residual value in the residual sequence. Through fixed multiplication operations, each residual value is multiplied by the adjustment ratio to achieve dynamic adjustment of all data in the residual sequence. When the probability of anomaly is high, the residual value is adjusted more significantly, while when the probability of anomaly is low, the residual value remains relatively stable. The continuous data sequence formed after adjustment is the adjusted residual sequence of the target object.

[0185] The adjusted residual sequence is processed by feature fusion. All data in the adjusted residual sequence are traversed and the mean of these data is calculated. This mean can comprehensively reflect the overall characteristics and deviation level of the adjusted residual sequence. This mean is determined as the adaptive correction factor of the target object, ensuring that the correction factor can fully integrate the effective information of the adjusted residual sequence and provide an accurate basis for subsequent data correction.

[0186] The beneficial effect is that the adjustment ratio coefficient of the residual sequence is determined based on the probability of anomalies, so that the ratio coefficient is directly linked to the degree of anomaly. The more severe the anomaly, the larger the coefficient and the stronger the adjustment. This achieves accurate matching between the correction factor and the anomaly state, and avoids insufficient or excessive correction caused by fixed coefficients.

[0187] The adjustment ratio coefficient is used to dynamically adjust all residual values ​​in the residual sequence, so that each residual value can be adjusted in a targeted manner according to the current degree of anomaly, forming an adjusted residual sequence. This retains the core information of the anomaly in the residuals while eliminating irrelevant interference, thus improving the effectiveness of the correction basis.

[0188] The adjusted residual sequence is subjected to feature fusion processing to integrate the overall deviation characteristics and variation patterns of the sequence, resulting in a single and comprehensive adaptive correction factor. This avoids the random influence of a single residual value and ensures that the correction factor can comprehensively reflect the overall trend of abnormal fluctuations.

[0189] The entire data coupling process deeply integrates the quantitative assessment of anomaly probability with the fluctuation characteristics of residual sequences. The resulting adaptive correction factor has dynamic adjustment capabilities and can adapt to different types and degrees of measurement anomalies. It provides a precise and flexible core basis for subsequent density value correction, significantly improving the pertinence and reliability of the correction.

[0190] S6. Apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object.

[0191] In this embodiment of the invention, applying the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object includes:

[0192] The current density measurement value of the target object is acquired synchronously.

[0193] Based on the positive or negative attribute of the adaptive correction factor, the correction direction for the density measurement value at the current moment is determined;

[0194] According to the correction direction, the adaptive correction factor and the density measurement value at the current time are algebraically synthesized to obtain the preliminary corrected density value of the target object;

[0195] The rationality of the preliminary corrected density value is verified;

[0196] If the preliminary corrected density value is within the preset physical property density value range of the target object, then the preliminary corrected density value is directly output as the corrected density value.

[0197] If the value exceeds the range of the physical property density value, the boundary value closest to the initial corrected density value in the range of physical property density value is output as the corrected density value.

[0198] By using specialized measurement equipment adapted to the density acquisition of the target object, the density data of the object is acquired in real time at the specific moment when density correction is required, ensuring that the acquired data can accurately reflect the actual density state of the target object at the current moment. The acquired data is the density measurement value of the target object at the current moment.

[0199] The numerical attributes of the adaptive correction factor are judged. If the adaptive correction factor is positive, the correction direction is determined to increase the density measurement value at the current time. If the adaptive correction factor is negative, the correction direction is determined to decrease the density measurement value at the current time. This judgment clarifies the operation logic of subsequent data synthesis and ensures that the correction direction is completely matched with the characteristics of the adaptive correction factor.

[0200] According to the determined correction direction, an algebraic synthesis operation is performed between the adaptive correction factor and the density measurement value at the current moment. If the correction direction is to increase, the density measurement value at the current moment is summed with the adaptive correction factor; if the correction direction is to decrease, the absolute value of the density measurement value at the current moment is subtracted from the absolute value of the adaptive correction factor. The value obtained through this algebraic synthesis method is the preliminary corrected density value of the target object.

[0201] Based on the physical properties, material attributes, and historical density data of the target object, a pre-defined range of density values ​​under normal conditions is established. This range comprehensively covers the density values ​​that the target object may exhibit under reasonable operating conditions. The preliminary corrected density value is compared with the pre-defined range to determine whether the preliminary corrected density value falls within this range, thus completing the rationality verification of the preliminary corrected density value.

[0202] When the rationality verification result shows that the preliminary corrected density value is within the preset physical property density value range, it means that the preliminary corrected density value conforms to the actual physical property law of the target object. No further adjustment is needed. The preliminary corrected density value can be directly used as the corrected density value of the target object for output to ensure the accuracy and effectiveness of the output results.

[0203] When the rationality verification result shows that the preliminary corrected density value exceeds the preset physical property density value range, the difference between the preliminary corrected density value and the upper and lower limits of the physical property density value range are calculated respectively. The two differences are compared, and the boundary value with the smaller difference is selected as the corrected density value of the target object and output. This ensures that the output corrected density value is always within a reasonable physical property range and avoids abnormal values ​​that do not conform to reality.

[0204] The beneficial effects are that the density measurement value at the current moment is obtained synchronously, ensuring the time synchronization between the measurement data and the correction process, avoiding inaccurate correction due to time deviation, and providing real-time and effective basic data for subsequent accurate correction.

[0205] The correction direction is determined by the positive or negative attribute of the adaptive correction factor, so that the correction operation is accurately matched with the actual situation of abnormal deviation, ensuring the rationality of the correction direction and avoiding the expansion of error due to reverse correction.

[0206] The adaptive correction factor is algebraically synthesized with the density measurement value at the current moment according to the correction direction. The preliminary corrected density value is obtained through targeted calculation logic, so as to realize the directional compensation for measurement anomalies and quickly offset abnormal deviations.

[0207] The rationality of the preliminary corrected density value is verified by introducing the preset density value range of the target object as the judgment standard, which effectively filters out unreasonable correction results that exceed the physical properties of the object and ensures the scientific nature of the correction value.

[0208] The corrected density value is output according to different situations. When the initial corrected value is within a reasonable range, it is output directly to ensure the correction efficiency. When it exceeds the range, the closest boundary value is selected for output to avoid abnormal values ​​that do not conform to the actual physical properties of the object, thus maximizing the reliability and practicality of the correction results.

[0209] The entire correction process, from data acquisition, direction determination, algebraic synthesis to rationality verification, forms a closed-loop processing mechanism. It not only adapts to measurement anomalies of different degrees, but also ensures that the correction results are consistent with reality through the constraints of the material property range, significantly improving the measurement accuracy and data reliability of the U-shaped vibrating tube density meter.

[0210] like Figure 2 The diagram shown is a functional block diagram of a measurement anomaly correction system for a U-shaped vibrating tube density meter provided in an embodiment of the present invention.

[0211] The measurement anomaly correction system 100 for a U-shaped vibrating tube density meter described in this invention can be installed in an electronic device. Depending on the functions implemented, the measurement anomaly correction system 100 may include a feature statistics module 101, a trend prediction module 102, a differential analysis module 103, an anomaly probability module 104, a data coupling module 105, and a density correction module 106. The module described in this invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0212] In this embodiment, the functions of each module / unit are as follows:

[0213] The feature statistics module 101 is used to collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical features of the target object.

[0214] The trend prediction module 102 is used to perform trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object.

[0215] The differential analysis module 103 is used to perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object;

[0216] The anomaly probability module 104 is used to input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object.

[0217] The data coupling module 105 is used to perform data coupling based on the anomaly probability and the residual sequence to obtain the adaptive correction factor of the target object.

[0218] The density correction module 106 is used to apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object.

[0219] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0220] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0221] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0222] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0223] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0224] Finally, 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.

Claims

1. A method for correcting measurement anomalies in a U-shaped vibrating tube densitometer, characterized in that, The method includes: S1. Collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical characteristics of the target object; The process of acquiring the original density value sequence of the target object and calculating the original density value sequence to obtain the statistical characteristics of the target object includes: Obtain the density measurement values ​​of the target object over a period of time, and arrange the density measurement values ​​in chronological order to obtain the original density value sequence of the target object; The original density value sequence is subjected to sliding window mean filtering to obtain the filtered density sequence of the target object; The mean of the filtered density sequence is used as the central tendency statistic, and the standard deviation of the filtered density sequence is used as the dispersion statistic. The coefficient of variation of the original density value sequence is obtained by performing a ratio discretization operation on the central tendency statistic and the dispersion statistic. The central tendency statistic, the dispersion statistic, and the coefficient of variation statistic are integrated into the statistical characteristics of the target object; S2. Based on a preset sliding time window, perform trend prediction on the original density value sequence to obtain the trend prediction value sequence of the target object; S3. Perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object; S4. Input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object; The step of inputting the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and outputting the anomaly probability of the target object includes: The pre-trained anomaly probability evaluation model is obtained through supervised training, which includes: The historical density data of the target object is obtained and processed into a feature sample set; To construct and initialize the forward propagation network structure of the target object; The feature sample set is input into the forward propagation network structure for iterative training, and the network internal parameters of the forward propagation network structure are optimized through loss calculation and backpropagation. When the training meets the stopping condition, the optimized parameters of the network's internal parameters are saved to obtain the anomaly probability evaluation model. The anomaly probability assessment model includes an input layer, a hidden layer, and an output layer; The statistical features and the residual sequence are combined into a comprehensive feature vector, and the comprehensive feature vector is input into the input layer; After receiving the integrated feature vector in the input layer, the hidden layer performs a nonlinear feature transformation on the integrated feature vector to obtain a high-dimensional feature representation of the target object. The high-dimensional feature representation is mapped to the output layer to determine the anomaly probability of the target object; S5. Based on the anomaly probability and the residual sequence, perform data coupling to obtain the adaptive correction factor of the target object; S6. Apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object.

2. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, The ratio discretization operation is performed on the central tendency statistic and the dispersion statistic to obtain the coefficient of variation statistic of the original density value sequence. The formula for calculating the coefficient of variation statistic is as follows: ； in, The coefficient of variation statistic is mentioned above. The aforementioned dispersion statistic, This refers to the central trend statistic.

3. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, The step of performing trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object includes: The preset sliding time window is time-sequentially segmented to determine the length of the sliding time window for the target object; Based on the length of the sliding time window, density values ​​within the window are sequentially extracted starting from the beginning of the original density value sequence to form a window density subsequence of the original density value sequence; A linear fit is performed on the window density subsequence to obtain the fitted straight line of the window density subsequence; Based on the fitted straight line, the trend prediction value sequence of the target object is obtained by extrapolating the subsequent sampled prediction values ​​of the window density subsequence.

4. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, The step of performing differential analysis between the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object includes: Align the original density value sequence with the trend prediction value sequence at different times to obtain the density-trend correspondence group of the target object; The instantaneous residual value of the target object is obtained by performing a difference operation on the density-trend correspondence group; The instantaneous residual values ​​are collected in chronological order to form the initial residual sequence of the target object; The initial residual sequence is smoothed to obtain the residual sequence of the target object.

5. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, After receiving the comprehensive feature vector at the input layer, the hidden layer performs a nonlinear feature transformation on the comprehensive feature vector to obtain a high-dimensional feature representation of the target object, including: Use the comprehensive feature vector as the initial input feature; The initial input features are linearly weighted and combined to obtain the intermediate feature representation of the target object. The calculation formula for the intermediate feature representation is as follows: ； In the formula, This is a representation of the intermediate feature. Let be the dimension of the comprehensive feature vector. For the first The preset weight coefficients corresponding to each feature component The first in the comprehensive feature vector Each feature component This is a preset bias term; The intermediate feature representation is subjected to a nonlinear transformation to obtain the high-dimensional features of the target object.

6. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, The step of coupling data based on the anomaly probability and the residual sequence to obtain the adaptive correction factor for the target object includes: Based on the anomaly probability, determine the adjustment ratio coefficient of the residual sequence; Based on the adjustment ratio coefficient, the residual values ​​in the residual sequence are dynamically adjusted to obtain the adjusted residual sequence of the target object. The adjusted residual sequence is subjected to feature fusion processing to obtain the adaptive correction factor of the target object.

7. The method for correcting measurement anomalies in a U-shaped vibrating tube density meter as described in claim 1, characterized in that, The step of applying the adaptive correction factor to the current measurement value of the target object to obtain the corrected density value of the target object includes: The current density measurement value of the target object is acquired synchronously. Based on the positive or negative attribute of the adaptive correction factor, the correction direction for the density measurement value at the current moment is determined; According to the correction direction, the adaptive correction factor and the density measurement value at the current time are algebraically synthesized to obtain the preliminary corrected density value of the target object; The rationality of the preliminary corrected density value is verified; If the preliminary corrected density value is within the preset physical property density value range of the target object, then the preliminary corrected density value is directly output as the corrected density value. If the value exceeds the range of the physical property density value, the boundary value closest to the initial corrected density value in the range of physical property density value is output as the corrected density value.

8. A measurement anomaly correction system for a U-shaped vibrating tube densitometer, characterized in that, The system for implementing the measurement anomaly correction method for a U-shaped vibrating tube density meter as described in claim 1 includes: The feature statistics module is used to collect the original density value sequence of the target object and calculate the original density value sequence to obtain the statistical features of the target object; The trend prediction module is used to perform trend prediction on the original density value sequence based on a preset sliding time window to obtain the trend prediction value sequence of the target object. The differential analysis module is used to perform differential analysis on the original density value sequence and the trend prediction value sequence to obtain the residual sequence of the target object; An anomaly probability module is used to input the statistical features and the residual sequence into a pre-trained anomaly probability evaluation model and output the anomaly probability of the target object. The data coupling module is used to perform data coupling based on the anomaly probability and the residual sequence to obtain the adaptive correction factor of the target object; The density correction module is used to apply the adaptive correction factor to the current measurement value of the target object to obtain the corrected density of the target object.

Images